Investigative journalism plays a critical watchdog role in holding power to account, and AI tools — ranging from automated document parsing to Large Language Models — are both enhancing and complicating that role. While AI systems can make investigations more efficient and scalable, they are not uniformly adopted due to constraints such as data quality, technical expertise, source confidentiality, economic resources and global inequalities. At the same time, AI companies and technologies themselves are becoming a subject of investigation, requiring journalists to scrutinize opaque algorithmic systems and powerful technology actors. Understanding these dynamics is essential for practitioners, researchers and industry leaders seeking to responsibly integrate new tools into investigative workflows while preserving journalistic integrity, independence and impact.

CNTI’s AI and Journalism Research Working Group examined how artificial intelligence (AI) is reshaping investigative journalism, drawing on a review of 44 recent academic and industry studies. This briefing synthesizes global research to assess both the opportunities and structural challenges AI introduces into investigative reporting.

AI Applications in Investigative Journalism

Investigative journalism has a distinct position within the news ecosystem, characterized — at least under ideal conditions — by extended timelines,² exceptionally rigorous verification standards and the uncovering of hidden or suppressed information often with moral or legal implications.³ In academic research, investigative journalism is described as “activist, reformer, and exposer”⁴ that is inherently adversarial, whose goal is to delineate crimes, identify victims and abusers, and advocate for social justice. Unlike routine news, which often recycles available information, investigative reporting begins with a hypothesis or a “denunciation” that must be confirmed or diluted through meticulous research and the collection of evidence, including on-the-ground reporting, confidential-source work and analysis of public records.⁵

These characteristics highlight why investigative journalism presents a uniquely complex environment for adopting new forms of automation. While AI can augment journalists’ capabilities through large-scale data processing, pattern detection, document handling and information verification, it cannot (as of now) replace the core epistemic and professional functions of investigative reporting. Investigative work often involves standalone cases and thus unique and project-specific methods of investigation: in-person “shoe leather” reporting, “dirty” or “incomplete” data with “messy” formats, hidden or withheld data,⁶ multisource verification, and high legal and ethical stakes that require human oversight given the need for high accuracy and contextual judgment.⁷ Current AI systems rely on existing data and patterns and therefore “can neither produce nor verify”⁸ genuinely new information uncovered through investigative work that is not yet available in digital form. As a result, investigative journalists tend to view AI less as a substitute for human labor and more as a tool that supports time-consuming tasks to allow reporters to focus on uniquely human aspects of the craft, such as source-building, persuasion and accountability reporting.

Altogether, the field of investigative journalism is currently navigating this tension between opportunities and limitations.

Use Cases

The range of applications of AI systems in investigative journalism represents targeted interventions across specific stages of the investigative process.

Reconstruction and Networking Mapping

In addition to performing algorithmic accountability through audits, AI tools also support modeling and reconstruction in transnational investigations. For example, AI systems have been used to align ship-tracking data with distress signals and survivor testimony to reconstruct migrant deaths in the Mediterranean.²¹ In another practice-based study — a collaboration of ​​data scientists, AI experts and journalists with the Norwegian Association for Investigative Journalism — one project involved using graph databases to map financial and regulatory networks in Norway’s petroleum sector.²²

Subterfuge and Source Protection

In regions with limited press freedom, the use of generative AI becomes closely tied to practices of subterfuge and source protection. A focus group study of 34 investigative journalists across five East African countries showed that AI tools are enabling journalists to investigate indirectly through data analysis, simulation and anonymization while minimizing exposure.²³ For example, Venezuelan news anchors use AI avatars to avoid being identified.

Innovative Storytelling

Finally, AI tools offer new storytelling formats — such as automated text generation, personalization and versioning using Natural Language Generation (NLG) and generative AI — expanding how investigative findings are produced and communicated to diverse audiences.²⁴ For example, Clarín offers a versioning tool that allows readers to access any story in six additional formats: summary, timeline, a comparison of numerical figures, a list of quotations, an index of proper names and an FAQ format.

AI Limits and Constraints in Investigative Journalism

The constraints and limitations of AI systems in investigative journalism fall under five interrelated domains: data availability and quality; professional identity and dependence on external actors; technical and epistemic limitations; linguistic barriers; and economic factors.

Where More Research Would Be Helpful

• Cross-industry comparison: More research is needed to compare how AI tools are adopted across journalistic subfields to better understand their unique applications and limitations in investigative contexts.
• Generalizable global research with investigative reporters: Much of the literature analyzed in this briefing used practice-based research techniques, bringing together investigative journalists, academics, data journalists and data scientists, to achieve a deeper understanding of investigative journalism as it is practiced today. While these provide deep insights, there is a clear opportunity for future studies to adopt broader quantitative approaches, such as large-scale surveys, to capture a more comprehensive global outlook on how investigative journalists use and perceive AI systems in their professional practice.
• Global South and context-sensitive AI: Most existing research on AI applications for investigative journalism focuses on the United States and Europe. Future work should focus more on developing and evaluating AI tools that are tailored to non-Western, non-English and culturally specific contexts in investigative journalism. 
• Emerging and evolving nature of AI: Given the rapid evolution of AI technologies and the limited number of existing studies, ongoing empirical research is needed to track their long-term impacts on investigative journalism.
• Local newsrooms: More research should examine how under-resourced newsrooms adopt, adapt to or are excluded from AI-driven investigative practices.
• Sustainable funding models: Future studies should explore viable and independent funding models that support AI innovation in investigative journalism without increasing reliance on external actors or compromising editorial autonomy.

Notable Cases

The following cases highlight that, across all regions, the consensus remains that AI tools serve as a force multiplier rather than a replacement for human labor. They enable journalists to examine data creatively and pursue stories that would otherwise be buried under the sheer scale of modern information.

Data sifting and lead generation

• AP Local News AI: In collaboration with AppliedXL, the Associated Press developed AP Local News AI. This tool monitors federal agency data to identify how national regulations impact specific local communities, effectively “anticipating” stories before they break.
• Datashare: It is a free, open-source desktop software developed by the International Consortium of Investigative Journalists (ICIJ) that enables users to index, search and analyze large volumes of documents (e.g., PDFs, emails, images) locally or on a server. 
• Nubia AI: an AI-powered data journalism platform designed to transform complex, messy datasets, such as PDFs, spreadsheets and satellite imagery, into structured data and publishable, newsroom-grade stories. Developed by Dataphyte Nigeria in 2022, it is designed to help journalists, researchers and civic organizations produce data-driven stories quickly and accurately.

Fact-checking and verification

• StatCheck: Developed by Inria (National Institute for Research in Digital Science and Technology) and RadioFrance, StatCheck automates the verification of statistical claims by cross-referencing articles against the INSEE and Eurostat databases.
• Dubawa: To combat regional misinformation in Western Africa, the Dubawa project launched an AI-powered chatbot and audio platform specifically designed to debunk viral falsehoods.

Content production and contextualization

• Odin: The Colombian outlet Cuestión Pública developed Odin, an AI system that connects past investigations to breaking news. Using retrieval-augmented generation (RAG), it searches proprietary databases and generates draft content grounded in verified reporting. The tool combines BERT for data indexing with GPT-4 for drafting, reducing production time from hours to minutes while maintaining editorial control.

Resources and Practical Guidelines 

While academic research on AI in investigative journalism remains relatively limited, a growing body of practitioner-oriented resources offers guidance on how these tools are being used in newsrooms. These sources are particularly valuable because they translate technical possibilities into replicable workflows, editorial strategies and ethical considerations. Key examples include:

• The Global Investigative Journalism Network’s guide, “Tips for Using AI as a Reporting Tool to Uncover Wrongdoing,” explains how AI can help structure large datasets, generate leads and build searchable systems, while emphasizing that tools should align with clearly defined reporting needs.
• GIJN’s Leveraging AI and Technology to Investigate Power highlights the importance of practical toolkits, low-code solutions, collaborative infrastructures and human-led investigative processes.
• GIJN’s 7 Best Practices for More Impactful Investigative Reporting, which stresses that core investigative practices — including patience, source development and narrative construction — remain central despite technological change.
• Analysis from the Center for International Media Assistance, showing how document-query systems and AI-assisted databases can reduce the time needed to analyze large document collections while still requiring verification and editorial oversight. 

Scope of included works

The reviewed studies span an international landscape, covering more than 20 countries and regions across Europe, North America, Africa and Asia and several cross-regional contexts such as Eastern Africa, Southeast Asia and Western Europe. Methodologically, the literature reflects a strong dominance of qualitative and mixed-methods approaches. Several studies also incorporated literature reviews, prototype development, usability testing and data-driven analytical techniques such as topic modeling, demonstrating substantial methodological diversity within the field.

Footnotes

1. Mari, 2024 ︎
2. Broussard, 2015; Chijioke, 2021; Čakš et al., 2025; da Silva, 2023; Ganguly, 2022; Katches, 2014 ︎
3. Boukhssas, 2026; da Silva, 2023 ︎
4. De Cooker et al., 2025 ︎
5. da Silva, 2023 ︎
6. Wellbrock, 2024 ︎
7. Fridman et al., 2025 ︎
8. Wellbrock, 2024 ︎
9. da Silva, 2023 ︎
10. Coddington, 2015 ︎
11. Lischka et al., 2025 ︎
12. Diakopoulos, 2024 ︎
13. Diakopoulos, 2024 ︎
14. Coddington, 2015 ︎
15. da Silva, 2023 ︎
16. Diakopoulos, 2024 ︎
17. Bradshaw, 2025; Cifliku & Heuer, 2025 ︎
18. Bradshaw, 2025; Stray, 2019 ︎
19. Cifliku & Heuer, 2025 ︎
20. Veerbeek, 2025 ︎
21. Mohamedy & Wandwi, 2025 ︎
22. Fridman et al., 2023 ︎
23. Gondwe, 2025 ︎
24. Bradshaw, 2025 ︎
25. Wellbrock, 2024 ︎
26. Kasica et al., 2023; Cifliku & Heuer, 2025 ︎
27. Wellbrock, 2024 ︎
28. Møller et al., 2025 ︎
29. Bradshaw, 2025 ︎
30. Helmond, 2015 ︎
31. Poell et al., 2022; Simon, 2022 ︎
32. Mohamedy & Wandwi, 2025 ︎
33. Dalgali & Crowston, 2020 ︎
34. De Cooker et al., 2025 ︎
35. De Cooker et al., 2025 ︎
36. Fridman et al., 2025; Mohamedy & Wandwi, 2025 ︎
37. Posetti 2018; see also Broussard et al. 2019; Kueng 2017; Thurman et al., 2019 ︎
38. De Cooker et al., 2025 ︎
39. For example, see Örnebring, 2010; Grimme, 2021. See also Møller et al. 2025 for AI specifically. ︎
40. De Cooker et al., 2025; Hollanek et al., 2025; Thurman et al., 2025 ︎
41. Dierickx et al., 2024 ︎
42. Bradshaw, 2025; Mohamedy & Wandwi, 2025 ︎
43. Hofeditz et al., 2025 ︎
44. Becker, 2023; de-Lima-Santos et al., 2024 ︎
45. Bradshaw, 2025 ︎
46. Bradshaw, 2025 ︎
47. Stray, 2019 ︎
48. Manolescu & Vaudano, 2024 ︎
49. CNTI, 2025 ︎
50. Ncube et al., 2025 ︎
51. Ncube et al., 2025 ︎
52. Gondwe, 2025 ︎
53. Ncube et al., 2025 ︎
54. Wellbrock, 2024 ︎
55. Wellbrock, 2024 ︎
56. Stray, 2019 ︎
57. Stray, 2019; Veerbeek, 2025 ︎
58. Umeora, 2025 ︎
59. de-Lima-Santos & Ceron, 2022 ︎
60. Radcliffe, 2025 ︎
61. Gondwe, 2025 ︎
62. Chimoio, 2026 ︎

The post AI Applications in Investigative Journalism appeared first on Center for News, Technology & Innovation.

The inclusion of students and young professionals in large numbers was heartening as many are already habitual users of AI tools and attended less as spectators and more as participants trying to understand where the technology is headed — and how to position themselves alongside it. Their presence underscored how quickly AI has shifted from a specialist policy conversation to an everyday technology shaping learning, work, and access to information.

One of the more substantive discussions focused on AI and journalism, where leaders from major Indian news organizations examined the structural pressures generative AI is placing on the news economy. Speakers highlighted the rise of AI-generated summaries and “zero-click” search experiences that reduce referral traffic to publisher websites, raising questions about how original reporting will continue to inform people and be funded. Editors emphasized that journalism depends on sustained investment in reporting and editorial oversight, even as AI systems increasingly rely on high-quality news content in their training and outputs. Panelists argued that AI systems participating in the distribution and summarization of news should meet higher standards of accountability and fair compensation. At the same time, they recognized AI’s potential to strengthen archives, improve workflows, and support subscription growth.

All in all, the @IndiaAI Summit was an important forum to move the conversation on how AI will be integrated into journalism and the broader economy — and what governance frameworks that integration will require.

The post Takeaways from the India AI Impact Summit appeared first on Center for News, Technology & Innovation.

AI governance is a complex ecosystem, incorporating policy instruments ranging from global compacts and legally binding domestic regulation to best practice standards and industry guidelines.¹ In December 2025, CNTI published a review of 188 governmental policy instruments and their impact on journalism, with a primary focus on legally binding legislation in various states of approval.²

In parallel, CNTI’s AI and Journalism Research Working Group reviewed the state of research on AI governance within newsrooms, including research on ethical implications and newsroom policy development for other emerging technologies. This briefing synthesizes 30 recent research papers.

Newsroom Policies Impacting AI in Journalism

Several studies show that while AI is being used in newsrooms, formal codification within newsrooms and professional societies is not yet universal, and there are still barriers to implementing AI policies.⁴ The working group reviewed 30 research articles addressing AI governance impacts on journalism, including policies developed by newsrooms and press associations, technology companies and governments. 

The speed of research is slower than the speed of policy development, which in turn lags behind technological development: it may well be that newsrooms have updated, added or advanced internal policies since these papers were published. Nonetheless, the takeaways and cautions from this research are still valuable, especially as journalists continue to develop policies.

Newsroom and professional guidance about the use of AI are not yet ubiquitous. As of late 2024, about 80% of the 221 Global South journalists surveyed by Thomson Reuters Foundation said their newsrooms have no AI policy.⁶ This number has almost certainly changed since then. However, what remains relevant is the barriers to developing and implementing policies identified here and in additional studies from around the world — including Germany, Greece, the Netherlands and Kenya.⁷ Barriers include a lack of access to technical expertise, difficulty in getting input and buy-in across organizations, the speed of technological change, the absence of regulatory frameworks in some places and the difficulty of complying with existing regulation in others. All the same, across contexts, journalists express the desire for guidelines and oversight. 

The newsrooms that do have AI policies share a similar approach, prioritizing transparency about the use of AI, human supervision of AI tools and human verification of outputs.⁸ However, few of these guidelines operationalize these priorities concretely or include clear oversight mechanisms. For example, some guidelines reference “proper” or “appropriate” uses without defining these terms. To date, three peer-reviewed papers explore newsroom policies, stylebooks and standards on the use of AI tools in journalism.⁹ Between them, these papers included 97 distinct policies from the European Union, Latin America as a region and 22 individual countries. 

These researchers also find that newsroom AI policies are ill equipped to address subtle biases that may be built into third-party tools. The guidelines focus more on AI outputs than on the systems themselves,¹⁰ are more concerned with generative than analytic AI¹¹ and rarely, if ever, provide practical guidance for working with third-party technologies.¹² For example, as outlined in the working group’s Transcription and Translation briefing, AI translation tools can introduce biases that may be difficult for non-experts to detect — like assuming doctors are men and nurses are women. These types of subtle biases exist beyond translation. By not addressing these concerns, newsroom AI policies fail to recognize that AI tools can harm journalistic integrity — and potentially journalistic independence — in ways that are difficult to detect. ¹³

In particular, few policies have clearly articulated when relying on third-party tools is appropriate and inappropriate. In theory, these concerns can be addressed through organizational procurement policies and guidelines that clearly identify the risks of different uses, particularly regarding data privacy and confidentiality. Concerns include the possibility that technology companies become indirectly involved in the newsroom, specifically, in the development of content.¹⁴ The major AI developers are primarily platform companies — including Google, Microsoft and Amazon¹⁵ — and newsrooms have long been at least somewhat dependent on them for distribution. Interviews with newsworkers suggest AI adoption is increasing dependence on platform companies, especially on the news production side.¹⁶ These concerns may be exacerbated in the Global South, because nearly all the early newsroom AI policies come from the Global North, while later policies borrow from them without necessarily addressing context-specific concerns,¹⁷ such as transcription quality issues. 

Procurement is an area where there has been little research to date. A 2025 study that considered 16 AI tools’ terms of service alongside interviews with decision-makers in newsrooms identified an ongoing challenge; most contracts granted developers the right to change the terms of service and the conditions without notice.¹⁸ This study also highlighted relative asymmetries between news organizations and AI developers as a barrier to managing risk contractually, a concern which may not even rise to awareness among individual journalists. While proprietary and local tools may have lower (or at least more customizable) risks than off-the-shelf ones, only a small number of the largest and wealthiest news organizations can practically build their own tools. (One promising recent conference paper explored how collaboratively governed and built LLMs could support the journalism field and address precisely this problem — but much more work is needed in this area.¹⁹)

Earlier guidelines addressing other new technologies in the newsroom — such as photo editing and social media — provide some useful parallels. As with AI policies, these policies typically start by articulating what journalism is and should be before highlighting how technology can support it and what uses are unacceptable. Like AI policies, photo policies did not always operationalize their values clearly; journalists and editors might disagree about what constitutes “excessive” retouching.²⁰

It is also common for social media policies to emphasize that journalists’ social media use must be consistent with existing journalistic values, ethics and procedures — including transparency and verification.²¹ Several researchers have also found that social media policies often protect news organizations — sometimes at the expense of individual journalists.²² Several studies of social media policies have found that differences in lived experience between editors developing the policies and reporters following them likely contributed to gaps in policies. In general, research analyses of various newsroom policies have concluded that including more stakeholders with varying job responsibilities and life experiences contributes to stronger policies.²³ A study that analyzed journalists’ tweets found that even when social media policies restrict their speech, journalists generally follow them.²⁴ Both the value of including more stakeholders and journalists’ general willingness to follow organizational policies are also likely to apply to AI policies.

Where More Research Would Be Helpful

• More research is needed on newsroom policies outside of the EU. There is some research, but much of it relies on data collected before the public release of ChatGPT and thus the widespread use of generative AI tools.
• There is also very little research on procurement, platform dependency, or relationships between newsrooms and technology companies outside the European context. 
• As technology and its use matures in news organizations, there is a need for more empirical and descriptive research, in addition to the early theoretical work.
• There is a lack of specific guidance for journalists and media organizations, especially regarding the use of third-party tools that may not be transparent. This is a particularly important gap since the research shows these tools may impact editorial decisions inconspicuously. 
• Given the distinct scopes, contexts and resources of different newsrooms, it is also important to examine how AI guidelines are being operationalized in daily workflows, as well as who participates in policy creation within the newsroom.

Footnotes

1. Ifayemi et al., 2024 ︎
2. Goodson et al., 2025 ︎
3. Mari, 2024 ︎
4. Porlezza 2024; Lefèvre et al., 2025 ︎
5. Lefèvre et al., 2025; van Drunen, 2025 ︎
6. Radcliffe, 2025 ︎
7. Dodds et al., 2025; Hofeditz et al., 2025; Kalfeli & Angeli, 2025; Mbaabu, 2025; Umejei et al., 2025 ︎
8. Becker et al., 2025; de-Lima-Santos et al., 2024; Sánchez-García et al., 2025 ︎
9. Becker et al., 2025; de-Lima-Santos et al., 2024; Sánchez-García et al., 2025, See also Cools & Diakopoulos, 2023 and Peck, 2023 for shorter pieces that have not been peer-reviewed. ︎
10. Becker et al., 2025, Porlezza, 2024 ︎
11. Sánchez-García et al., 2025 ︎
12. de-Lima-Santos et al., 2024; see also van Drunen, 2025 ︎
13. Becker et al., 2025; de-Lima-Santos et al., 2024; Sánchez-García et al., 2025 ︎
14. Simon, 2024; Simon, 2025; van Drunen, 2025. See also Timcke & Schroeder, 2025. ︎
15. Simon, 2022 ︎
16. Simon, 2024; see Seipp et al., 2024 for a policy perspective ︎
17. de-Lima-Santos et al., 2024 ︎
18. Piasecki & Helberger, 2025 ︎
19. Tseng et al., 2025 ︎
20. Solaroli, 2017 ︎
21. Duffy & Knight, 2019; Herrera-Damas 2014; Harlow, 2023; Sacco & Bossio, 2017 ︎
22. Harlow, 2023; Miller & Nelson, 2022; Molyneux & Nelson, 2024 ︎
23. Miller & Nelson, 2022; Molyneux & Nelson, 2024 ︎
24. Lu et al., 2025 ︎

The post Newsroom Policies for AI in Journalism appeared first on Center for News, Technology & Innovation.

AI chatbots — such as ChatGPT, Gemini, Grok, Google AI Mode or the Washington Post’s “Ask the Post AI” — are software products which simulate human-like conversations and generate responses across a wide range of topics. While even their creators caution against using them as arbiters of fact, research consistently demonstrates that people increasingly rely on them for information about the world. Since the line between “information” and “news” is hardly clear-cut and information-seeking is also likely to include many topics where users of AI chatbots might previously have turned to news sources, news providers need an understanding of those broader informational habits.

This report provides a detailed snapshot of relatively early adopters and their use of AI chatbots to get information, including news. In October 2025, CNTI interviewed 53 people in India and the U.S. who use AI chatbots at least once a week and said they “keep informed about issues and events of the day” at least “somewhat closely.” 

Top-level findings:

CNTI found that most AI chatbot users currently use them to supplement their existing information repertoires, not to replace them entirely. They toggle back and forth between AI chatbots, news sites, search engines, official sources and more. We saw:

• Actionable information: Interviewees use AI chatbots to act on what’s happening and understand it, more than simply to know about it or to feel something about it.
• Personalization: AI chatbot users see these tools as fast, easy, personalized, customizable and friendly ways of getting information.
• Low understanding but higher trust: Few AI chatbot users have deep knowledge about the processes behind either journalism or AI chatbots. At the same time, interviewees express a general trust of AI chatbots alongside a general distrust of news media (outside of some users’ trust in their specific sources).

More specifically:

Interviewees use AI chatbots to act on what’s happening and to understand it, more than simply to know about it or to feel something about it.

What most drives AI chatbot use is the desire for information that helps people act. This is the most common use case for AI chatbots in both countries. Few interviewees want information for its own sake. Instead, they are looking to inform their choices and actions. AI chatbots stand out, especially in the U.S., for proactively helping people meet the need to act.

When it comes to knowledge for knowledge’s sake, interviewees in both countries use AI chatbots to supplement existing news habits, not replace them. No interviewee in either country said they rely solely on AI chatbots when they want to know what is happening. They do turn specifically to AI chatbots to learn what’s new about a story they are familiar with or to check a story that they don’t think is accurate (the latter now offered on some social media platforms, such as @grok on Twitter or Meta AI on Facebook).

Both U.S. and Indian AI chatbot users turn to them for context that is often absent in traditional news media. Many interviewees said that news stories often start in the middle, with background information buried or difficult to find. They find AI chatbots helpful for understanding who a public figure is, how another country’s legal system works and the history of Israeli-Palestinian relations — all contexts that news stories sometimes take for granted.

Emotional stimulation and entertainment are low information priorities. The “need to feel” does not seem to be a focus for our interviewees in either country, and did not come up much in our conversations. Interviewees do engage emotionally with AI chatbots, but only rarely with the information itself.

Interviewees see AI chatbots as a fast, easy, personalized, customizable and friendly way of getting information.

Few interviewees have deep knowledge about the processes behind either journalism or AI chatbots. At the same time, they expressed generally positive attitudes towards AI chatbots alongside generally negative ones towards news media.

Acknowledgements

The authors would like to thank Amy Mitchell for ongoing intellectual mentorship, partnership and leadership; Nupur Chowdhury, Rahul Dass, Kyong Mazzaro, Amy Mitchell and Nikita Roy for their thoughtful feedback on this report; Jonathon Berlin and Kurt Cunningham for web and graphic design; Angelica Ruzanova for support with transcription and data processing; and Greta Alquist for editing this report.

Continue reading

The post Action, Ease & Personalization: AI Chatbot News Experiences appeared first on Center for News, Technology & Innovation.

Whether newsrooms are paying enough attention or not, however, many governments are adopting AI strategies, policies and laws. These regulatory attempts rarely directly address journalism and vary dramatically in their frameworks, enforcement capacity and level of international coordination, but they will almost certainly have an impact on journalism and the digital information environment. To understand exactly how regulation is, or could be, impacting journalism, we reviewed 188 national and regional AI strategies, laws and policies that collectively cover more than 99 countries. 

The analysis is structured around the seven regions of the world, as identified by the World Bank — North America, Latin America and the Caribbean, Europe and Central Asia, the Middle East and North Africa, Sub-Saharan Africa, East Asia and the Pacific, and South Asia — and around how regulatory activity in each of these regions addresses seven policy components that, based on the research of CNTI and others, are particularly relevant to and likely to impact journalism:

Explore the proposals

CNTI reviewed 188 national and regional AI strategies, laws and policies around the world to determine how they impact journalism. We focused our analysis around how regulatory activity addresses seven policy components that are particularly likely to impact journalism. Use the below map to explore each of the policy documents we reviewed and to see what journalism-related topics they include. Use the column on the left to see which documents mention a specific topic, or click on a country to see the list of its documents and the topics covered in each. 

Key findings: The landscape

The governance of AI is increasingly complex and varies dramatically by country and industry. 

The broad range of governance instruments in play include legally-binding regulation at every level of government, corporate policies, industry standards and many other types of documents. AI applications are being introduced across sectors: some of them, such as banking and health, are already heavily regulated, while others are much more lightly regulated. Even among national-level legislation, there is tremendous diversity in scope: some bills are sweeping and comprehensive, while others comprise small edits to existing laws or address AI in narrowly-defined use cases like online shopping, mammograms or road accidents. Others create new agencies, committees or councils to operationalize vaguely laid out principles. The law-making process varies by country, as do the stakeholders involved. There is also variability in countries’ power to impact technology companies. In some smaller markets, companies may consider pulling out rather than comply with regulations, especially if those regulations are burdensome or do not align with the requirements of larger markets. Furthermore, some countries — the United States among them — have dozens or even hundreds of bills in the pipeline, making it impossible to conduct an exhaustive review. While we offer numbers wherever possible to give a sense of scale and proportion, our research examines the breadth of policies, strategies and laws rather than providing a full audit.

As with all policy proposals, it is important to consider the AI documents we reviewed in their individual country contexts and in relation to pre-existing laws in those countries and regions. Even if journalism or data privacy is not explicitly mentioned in an AI document, for example, existing protections for these rights may supersede new AI proposals. It is also important for policymakers to coordinate with experts to determine whether their AI policy proposals are technologically feasible as they consider replicating the documents reviewed in this paper or originating new AI policy proposals. 

Policies don’t have to mention journalism by name to impact it. 

Twenty of the 188 documents explicitly mentioned “journalist,” “journalism,” “news,” “media” or “news media.”

When these terms do appear, they range from passing references to the field to laws whose primary concern is the practice of journalism. This by no means is a suggestion that more documents should directly name journalism. Calling out the journalism field in policy documents frequently has its own pitfalls: Once governments define “journalism” or “news,” those definitions can be weaponized against the news media. Instead, it is important that policymakers are aware of and think through potential impacts of directly naming or not naming journalism and news. 

• Four documents (from Bahrain, Chile, Costa Rica and Kenya) identify journalists or media workers as stakeholders, but go no further. Three others (from the African Union, Panama and Serbia) specifically identify the field of journalism as an important audience for educational materials and reskilling. 
• Four documents (from Algeria, Egypt, Lesotho and Sri Lanka) emphasize the importance of news media as a communications channel for public awareness campaigns, with varying assumptions about editorial independence from government. 
• One proposed law (from Ecuador) attempts to intervene in polarization and fragmentation by requiring providers of content recommendation algorithms to include content from a broad range of media outlets.
• Five different bills and laws exempt journalism from specific provisions: China’s Interim Measures on Generative AI say that other journalism laws supersede it; three U.S. bills would exempt news media from specific restrictions on deepfakes in the context of both reporting and advertisements (Illinois, New Hampshire and South Dakota); and Brazil’s proposed law grants exemptions to some forms of copyright violation for journalistic or research purposes..
• Two proposed laws — both from the U.S., one in New Jersey and one in New York — would specifically regulate the use of AI in journalism.

Transparency and data protection are the two topic areas (among the seven studied) that come up most frequently.

Of the 188 AI strategies, laws and policies we reviewed, 124 addressed transparency and accountability; 107 addressed data protection and privacy; 92 addressed algorithmic discrimination and bias; 76 addressed public information and awareness about AI; 64 addressed manipulated or synthetic content; 49 addressed intellectual property and copyright; and 19 addressed freedom of speech and expression. Each document could be counted in multiple categories. In all seven regions, either transparency or data protection was the most common topic. In every region, more than half of the documents we reviewed addressed transparency and accountability. Likewise, in every region, more than half of the documents we reviewed addressed privacy and data protection, with the exception of North America, which only featured the topic in eight of the 29 documents we examined. The emphasis on transparency likely responds to a key challenge for accountability: the opaqueness of many AI systems leaves even their creators without a full understanding of how they work, let alone policymakers. Meanwhile, the salience of data privacy is to be expected, since it has been a key issue in technology governance over the last ten years. At least four out of five people around the world are protected by a national-level privacy law. 

Freedom of speech and expression come up least often.

While policymakers regularly express concern about AI’s impacts on the information environment, references to freedom of speech and expression are infrequent across regions. In fact, these topics did not appear in any documents we reviewed from either the Middle East and North Africa or Sub-Saharan Africa. Moreover, the documents we reviewed addressed two very different concerns: Some, like Malaysia, addressed the possibility that using AI would threaten fundamental human freedoms, while others, like Venezuela, addressed the possibility that regulating AI would threaten human freedoms. The latter was more common in the United States than elsewhere, but both issues came up across regions. The EU’s AI Act highlights both concerns, emphasizing that AI systems can violate fundamental freedoms but also that labeling obligations do not restrict free speech. One resolution, New Jersey’s AR 158, instead focuses on freedom of speech about AI by urging technology companies to embrace stronger whistleblower protections. We note that the absence of explicit language about freedom of speech or expression does not necessarily indicate an absence of concern; in many countries, these freedoms are guaranteed in foundational legal documents, like constitutions, that supersede all other laws or policies. The certainty around the upholding of those foundational documents, however, cannot be guaranteed.

Key findings: The topics

Freedom of speech and expression

Out of the 188 documents we reviewed, 20 address freedom of speech and expression directly.

CNTI’s analysis finds that when freedom of speech and expression are recognized, it generally has positive implications for journalism. This demonstrates an understanding of the importance of this freedom and an acknowledgement of how AI can impact it. Because it is often only included in preambles or as a guiding principle, however, these policies often do not specify how they are going to protect these rights. Examining the range of existing levels of press freedom, both across and within regions, is foundational to understanding how these policies may be used.

Of the few concrete policies about freedom of speech and expression, most are likely to benefit journalism. Several countries ban AI systems that do not respect freedom of expression, but they do not specify how or what uses of AI systems would fail to respect fundamental freedom. For example, Argentina’s Bill 2573-S-2024 prohibits the use of AI “which violates fundamental human rights such as privacy, freedom of expression, equality or human dignity.” Other countries recognize trade-offs: In order to ensure that algorithmic content recommendation does not violate freedom of expression, Ecuador’s Organic Law for the Regulation & Promotion of AI in Ecuador requires clear terms and conditions, human supervision, accountability reports and an appeals process. These provisions would make it more difficult for journalists to be subject to censorship at the hands of an overly conservative algorithm. 

Manipulated or synthetic content

Of the 188 documents we reviewed, 64 address manipulated or synthetic content directly. At least two documents from each region contain provisions about this issue.

CNTI’s analysis finds that attempts to prevent the spread of false information and highlight the provenance of information are broadly positive for journalism and the information space — but these efforts must be scrutinized to ensure they do not infringe on freedom of expression. 

For example, some proposals may protect journalists from technology-facilitated gender-based violence and other forms of AI-generated harassment. When a journalist is targeted with a deepfake, it can push them into silence, decrease their credibility and even drive them to leave the field. By banning explicit deepfakes, the United States’ Take It Down Act, Mexico’s Ley Olimpia and Bahrain’s draft AI Regulation Law would prohibit such attacks on journalists. 

However, some of these proposals may end up penalizing journalists. For example, the United States’ 2025 Take It Down Act prohibits the nonconsensual publication of both authentic and AI-generated intimate images and requires online platforms to remove such images. It does not, however, include safeguards against bad-faith or fraudulent takedown requests, which some advocates say could be used to wrongly censor journalism. 

Moreover, some provisions could potentially be used by the government to target journalists for reporting it does not like. Bahrain’s draft AI Regulation Law would prohibit using AI to “upload or install personal images that damage an individual’s reputation or dignity; [… or …] modify, edit, or tamper with textual, audio, or visual content related to individuals without their explicit consent.” These provisions are sufficiently vague that they could potentially be weaponized against journalists for innocuous changes, like editing the levels of an audio recording to make it clearer.

Finally, some proposals could impact journalists’ ability to use AI to protect their sources. The Dominican Republic’s proposed Bill 563, for example, would ban the use of deepfakes to alter videos and punish violators with prison terms and hefty fines. Because it does not include an exception for journalism, even when it is labeled, this could prevent journalists from using AI for legitimate purposes, such as to create an avatar of a source who wishes to remain anonymous for their safety.

Many laws also treat synthetic audio, images and video differently than synthetic text, increasing complexity in this space.

Algorithmic discrimination and bias

Out of the 188 documents we reviewed, 92 address algorithmic discrimination and bias directly. 

Some laws would limit the tools that news organizations can use for decision-making, but these laws would not uniquely impact this sector — they typically focus on the use of AI tools for hiring and other consequential decisions, and would apply universally. News organizations would simply have to comply with these regulations.

However, policies that focus on bias in content recommendation systems could potentially have strong impacts on the reach of journalistic content. Content recommendation systems impact the content people see online, especially on social media platforms. Some news organizations also use them to suggest content to audiences. If social media platforms are required to recommend a diversity of sources and opinions, this would impact the reach of journalistic content, but it’s hard to tell if it will extend or limit that reach. Some AI policies — particularly in the EU — would limit the recommender tools used by social media platforms, but not by news organizations themselves, making the impacts unpredictable. Elsewhere, such as in Ecuador, provisions would require news organizations to scrutinize their personalization or content delivery tools, such as AI-powered, customizable homepages.

Other documents, such as those in Bangladesh and Lesotho, would push journalists and newsrooms to be more cognizant of biases in tools they use to analyze data or create content, especially if these provisions require representative or diverse training data. Because journalists already prioritize objectivity, such awareness falls under good journalistic practice and is unlikely to penalize journalists.

Looking beyond journalistic uses of AI, some proposals create registries and audits, which may facilitate journalists’ ability to conduct accountability reporting on AI more broadly. For example, Lesotho’s Draft Artificial Intelligence Policy and Implementation Plan calls on the country’s AI regulator and policymakers to develop “bias mitigation programs,” by mandating bias audits, providing open-source tools for bias mitigation and detection, and training developers and other stakeholders on bias prevention. This could grant journalists greater insight into how these technologies work, thus improving their reporting capability and increasing public accountability. It could also make it easier for them to select appropriate tools for their professional use.

Intellectual property and copyright

Out of the 188 documents we reviewed, 49 address intellectual property and copyright directly. At least three documents from each region included this policy component.

Copyright and intellectual property regulations are particularly complicated because different countries’ laws are incompatible with each other, and few have reckoned with major shifts in distribution made possible by the internet. 

We found that some bills and laws require licensing and compensation for the use of copyrighted material to train AI models. Colombia’s Bill 293 states that beyond an exception for scientific uses, developers cannot use copyrighted content to train AI without prior and explicit consent. On the other hand, the Digital Single Market Directive of the EU AI Act, for example, does allow copyright holders to ‘opt-out’ of their content being used as training data for commercial purposes. The bill in Colombia also gives copyright collectives the right to authorize, prohibit, or restrict the use of works under their management, and they can demand just and equitable remuneration in order to license that use. These legal provisions will likely have positive financial impacts for journalism producers if they can be enforced.

At the same time, a number of documents mention the importance of protecting existing copyright and intellectual property laws without necessarily specifying how AI systems — particularly those using data in training models — fit into these legal foundations, let alone offering a comprehensive assessment of the value of these interactions with digital content. 

For example, China’s generative AI regulation stipulates that both deployers and developers need to comply with intellectual property laws in the country, but it does not specify how AI systems and AI-generated content fit into those laws. 

Japan, on the other hand, takes one of the most permissive stances, allowing the use of copyrighted works for AI training regardless of purpose, so long as it does not unreasonably prejudice the rights-holder’s interests. These divergent approaches raise the issue of interoperability when models are trained globally and deployed across borders. 

The matter of intellectual property and AI is far from settled, though. In November 2025, a court in the United Kingdom found that AI models are subject to copyright infringement claims; however, the lawsuit did not answer the question of whether using copyrighted materials to train AI models falls within the U.K.’s “fair dealing” provisions. Courts in the United States have taken differing views, with two judges in California ruling that AI companies’ use of copyrighted materials to train LLMs constitutes “fair use” because the work is “transformative,” albeit with caveats and for different reasons, while a judge in Delaware ruled that a different AI company’s use of copyrighted materials was not “transformative” and, thus, did not meet “fair use.” This area will continue to evolve as ongoing lawsuits determine whether copyrighted material can be used to train AI models around the world, including in the U.S., India and Japan. 

What is becoming clear is the need to holistically address the value of the various kinds of uses of and interactions with digital content in building AI models and beyond. Current policy approaches tend to dilate between narrow licensing regimes and permissive exceptions, but few adequately balance creators’ rights, developers’ needs and the public interest, let alone the vast array of content itself. 

Transparency and accountability

Out of the 188 documents we reviewed, 124 address transparency and accountability directly. At least three documents from each region included this policy component.

CNTI’s analysis finds that their impact on journalism will likely vary depending on whether transparency obligations fall more heavily on developers or deployers. If they fall on developers, such as in Malaysia, it will be easier for journalists to assess whether third-party tools are valuable to their work; if they fall on deployers, like in this Argentine proposal, journalists, as deployers, would be held responsible for choices made by third-party companies which could lead to greater attentiveness among journalists, and, if not, to unanticipated lawsuits. Several documents, however, fall somewhere in between. The United Arab Emirates’ AI Ethics Guide, for example, states that “accountability for the outcomes of an AI system lies not with the system itself but is apportioned between those who design, develop and deploy it,” meaning journalists could be held accountable at any stage of the process.

We also found that, in general, transparency requirements can make it easier for journalists to report across a range of topics and sectors. For example, California’s AI Transparency Act requires transparency on the data used to train AI models, which would increase public understanding of the tools and, thus, allow journalists to better assess these models.

Some proposals would also require news organizations to label AI-generated content to ensure the public recognizes when it is being used. New York’s proposed Senate Bill S6748, for example, would require publications to “conspicuously” identify at the top of a page or webpage when AI is used to either partially or wholly create an article, image, video or other piece of content. Such requirements could potentially be a form of “compelled speech” or could reduce trust in the content. 

Data protection and privacy 

Out of the 188 documents we reviewed, 107 address data protection and privacy directly.

AI systems and tools are trained on vast quantities of data, and many countries have comprehensive privacy legislation in place that will interact with AI regulation in complex ways.

AI legislation, policies and strategies that address data protection and privacy typically do not consider the specific needs of journalists, who require access to sensitive data for investigative reporting. Bangladesh’s National Artificial Intelligence Policy, for example, states that “personal data usage will require valid consent, notice, and the option to revoke.” Requiring journalists to receive consent before using personal details, however, could inhibit investigative reporting on corruption, human rights abuses and more, given that the politicians, business leaders and others in question would likely not grant consent for their data to be used.

On the other hand, CNTI’s analysis finds that proposals and laws that ban the use of AI for surveillance, as some countries have contemplated, would likely improve the safety of journalists and their sources. For example, the EU AI Act bans AI systems that “create or expand facial recognition databases through the untargeted scraping of facial images from the internet or CCTV footage.” While this represents a step in the right direction, the law does have exemptions for national security. Some countries have taken advantage of this; Hungary, for example, allowed the police to use biometric surveillance to identify participants in LGBTQIA+ public events. 

Public information and awareness

Out of the 188 documents we reviewed, 76 address this issue directly. At least two documents from each region directly touch on this issue.

Many provisions for broad public awareness include journalists among their audiences, providing up-to-date sociotechnical knowledge that could inform stronger reporting. For example, Serbia’s AI strategy includes plans to organize seminars on AI, information security and big data specifically for journalists. Other policies offer similar information to all adults. 

Documents that recognize the importance of journalism as a vehicle for public awareness are varied in their strategies. Lesotho’s draft policy encourages stakeholders to do outreach with a diverse spectrum of media outlets, acknowledging the importance of news without raising concerns about press independence. On the other hand, Egypt’s strategy expressly calls on the media to share “positive news of AI,” perhaps suggesting a bid to influence coverage. No document we reviewed calls for governments to spend advertising dollars to place public awareness campaigns in news media.

Recommendations

The inclusion or exclusion of these seven topics does not necessarily make a proposal “bad” or “good” overall or for journalism and the digital information space. There are many important considerations for AI policy proposals, of which journalism is just one. It is one, though, that CNTI finds incredibly important to functioning societies.

If an independent, diverse news media and open internet are not protected in the AI era, these new regulations could potentially criminalize journalism, threaten news business models, contribute to information disorder and prevent the public from accessing a diversity of fact-based news. The solution is not to call out journalism by name in every AI proposal, as that can have unintended consequences for the field, but it is essential that policymakers see journalists as important stakeholders in AI discussions and consider journalism’s viability as they develop future proposals. Likewise, it is important that news organizations and journalists see themselves as key stakeholders and thoroughly engage in thinking through and discussing the future of AI regulation.

The analysis surfaces a few key areas in need of specific policy attention: 

1. 1. In AI proposals that address manipulated content, it is important that policymakers work towards methods that protect certain journalistic uses in ways that do not enable government censorship or determination of who is or is not a journalist. This is far from an easy task and may mean the best path is no legal policy at all. Either way, it is critical to fully think through. Legislative efforts to prevent the spread of disinformation can sometimes have unintended consequences for journalism. For example, legislation that regulates the provision and sharing of manipulated content but does not include exceptions for journalism can lead to journalists being targeted for allegedly “spreading disinformation” for reporting on the existence of false information, such as a deepfake of an elected official, for using AI-manipulated content (like AI-generated avatars or voice-altered audio) to protect a source, or for covering information that those in power do not want covered. While not legally feasible in the U.S., carve outs for journalists in other countries could protect an independent news media, as well as a source’s right to privacy and willingness to share. But such carve outs, if not crafted extremely carefully, could also easily lead to greater censorship and criminalization. This takes coordinated, thoughtful deliberation. Alongside these discussions, journalists and news organizations should determine how best to convey their use of AI to the public, whether through labeling, watermarking or something else. 

2. 2. Bias audits and transparency measures are best implemented before a tool is deployed. To date, bias audits and transparency measures have often been reactions to issues that arise after deployment rather than components required prior to public launch. While not all issues can be identified in advance, more can be done before a tool is deployed, especially as these systems and our understanding of them matures. Policy can help lay out foundational requirements for audits, as well as for transparency, around how they are conducted and how issues are addressed, particularly for technology companies whose algorithms can have a social impact (i.e. if the algorithm uses personal data and makes or influences decisions that may have a significant impact on society or an individual). 

3. 3. Policymakers should ensure that AI working groups include journalism producers, product teams and engineers, alongside AI technologists, researchers, civil society and other relevant stakeholders. Policymakers face the challenge of attempting to legislate a technology that is constantly evolving and has wide-reaching consequences. As such, it is important that they meet with diverse stakeholders to ensure that they properly understand the technology they are trying to govern, as well as the consequences that both the technology and potential legislation could have on society more broadly. It is also critical that all at the table, including journalism leaders, are fully read-in and share the goal of collaboration, as such cooperation leads to the most optimal societal solutions. Expanded regional and international collaboration would also be valuable as AI regulation is a global challenge with cross-border implications. 

Acknowledgements

The authors would like to thank Monica Attard, Charlie Beckett, Niamh Burns, Claudia Del Pozo, Mohamed Farahat, Megan Gray, Assane Gueye, Jhalak Kakkar, Ashkhen Kazaryan, Tanit Koch, Prabhat Mishra, Amy Mitchell and Daniela Rojas for their thoughtful feedback on this report. 

The authors also extend their gratitude to Greta Alquist for editing this report, Jonathon Berlin and Ryan Marx for designing the graphics and maps, Kurt Cunningham for creating the web design and the team at CETRA for their translation work. 

The post Journalism’s New Frontier: An Analysis of Global AI Policy Proposals and Their Impacts on Journalism appeared first on Center for News, Technology & Innovation.

On October 28 and 29, 2025, the Center for News, Technology and Innovation and Thomson Media brought together more than 35 journalists, product specialists and newsroom leaders from the Western Balkans and Central Europe in Sarajevo, Bosnia and Herzegovina, for “Journalism and AI: Building Resilient Newsrooms for the Future” — the second collaboration between the organizations. Over the course of two days, participants shared their struggles and successes integrating AI into their work, took master classes on AI prompt engineering and contemplated the question, “How do we harness AI’s power to build a strong future for our work?”

The answer that emerged was both multifaceted and promising. The journalists in this region are leaning in and figuring out what works best and what does not work for them and their audiences, and what they can learn from each other — and they often are managing with only small budgets and staff sizes.  

At one point during the workshop, CNTI Chair Craig Forman explained, “I wrote for the Wall Street Journal when I was a foreign correspondent, 30-ish years ago on the eve of the breakup of Yugoslavia.” He continued, “I bring that up, not to bring any ill will or bad memories, but to say that when I [was reporting in the region], it was unthinkable that we might be here today. Unthinkable. And as you all know, we have to think about the unthinkable, not only in the bad way, but in the good way.”

Forman’s words resonated deeply in a city once synonymous with war reporting but now hosting a discussion about how technology could safeguard truth and foster resilience in newsrooms.

AI Will Not Take Journalists’ Jobs, But Someone Who Uses AI Will

David Caswell, drawing on years of experience developing AI storytelling systems, kicked off the workshop with a powerful and straightforward insight: “We should take AI seriously because the trajectory of improvements has been radical.” He continued, “The effect of [AI investment] will be significant.”

This isn’t about jumping on every technological bandwagon or tacking the latest tool onto one’s current product. It’s about recognizing that structured storytelling models and AI-assisted workflows represent a practical evolution in how newsrooms operate. The question isn’t whether AI will change journalism, but whether journalism will adapt intelligently.

As Niamh Burns from Enders Analysis said, “AI will undoubtedly change how news is created, distributed, noticed and funded. This is really a moment where you [journalists] should all be thinking about what your value add is.”

These aren’t abstract problems for newsrooms in Sarajevo, Belgrade or Pristina. They’re immediate challenges that require practical solutions.

Sami Kçiku, a project manager at the independent news company Koha Group in Kosovo, shared that his newsroom was initially fearful of AI. Several other participants voiced similar stories of journalists worried they would be replaced by AI and, therefore, reluctant to try out new technology, with generational divides also at play. 

Kçiku offered the simple yet disarming advice that he gave to his team: “You won’t be replaced by AI, but you will be replaced by someone who uses AI.” Koha has since implemented a custom GPT to improve SEO and social media practices, and they use an external AI tool for transcription.

Journalists Must Begin to See Themselves as Innovators

As newsrooms attempt to address the tension between embracing AI and fearing it, Caswell recommended a calm but intentional approach of “innovation, adoption, diffusion.” 

It is important that innovation comes from collaboration between the editorial and product teams, and, as Burns pointed out, innovation should prioritize measurable success instead of bandwagon adoption. Nikola Bačić, editor-in-chief at Hercegovina Info, shared that his team has one AI meeting every week, where the editorial staff meets with the IT staff to discuss needs and potential solutions. 

Once a tool is created, newsrooms must take the time to properly train the entire staff on a technology to ensure they have the skills needed to adopt the AI tool, if they so choose. Tatjana Sekulic, an executive multimedia producer at N1 in Bosnia and Herzegovina, told the group that uptake at her outlet is mixed; some journalists are completely against AI, while others are overly reliant on it. As a result, the newsroom has implemented training for everyone to ensure they all have the skills to use AI responsibly. 

“It’s very important to teach them how to use AI in the proper way,” Sekulic said. “We’re investing in our knowledge and our people.”

Finally, following implementation, it is important to continue conversations to diffuse broader adoption of the technology. Veronika Munk, director of innovations at Denník N in Slovakia, explained that while full newsroom training sessions did not always achieve the desired impact for her team, they still provided useful insights. Over time, the organization found that complementing these larger trainings with a more targeted approach worked better. When Denník N introduces a new tool, they now focus on training smaller groups within the team or on one-on-one micro-trainings. These trained editors and reporters can then gradually share their knowledge with other colleagues. She also emphasized that the organization’s use of AI always involves human oversight.

While participants responded to this discussion with enthusiasm, some relayed concerns about the cost of implementing AI tools in newsrooms with already thin budgets. 

Damjan Dano, a tech entrepreneur from North Macedonia, explained that creating customized AI tools is not the only option; instead, there are a plethora of existing, inexpensive AI tools that newsrooms can use to support their work. Dano led participants through an exercise where they laid out their AI wish lists, and he shared a variety of existing tools that could be a solution to some of their needs. Frase, for example, can help with SEO optimization, and Asana AI can help manage newsroom workflows.  

“AI is a great tool, not your substitution,” he reminded participants. “Your job is safe, but you have to use the tools that exist today.”

These discussions came with a crucial caveat: Journalism’s future depends on “adaptability, but not at the expense of ethics,” Caro Kriel, chief executive of Thomson Foundation, said.

Ethics and Audience Relations Must Be a Core Part of Newsrooms’ AI Strategies

Another key theme over the two days of discussions was that AI should augment human judgment, not replace it. Newsrooms need to ensure their use of AI is driven by the values and ethics they espouse. It should enable and empower journalists to do their best work, to report relevant, important news and to deliver it to audiences effectively.

Marius Dragomir, director of the Media and Journalism Research Center, reminded participants that technology questions are never just technical. They’re about values, accountability and the social contract between journalists and their audiences. Every AI implementation carries an ethical weight that newsrooms must acknowledge and address.

“The trust of our audience is our highest value. To preserve this trust, we must be honest and transparent,” Vesna Ivanovska-Ilievska, co-founder and editor-in-chief of Umno.mk, said. 

In a session on AI, ethics and trust, Branislava Lovre, co-founder of AImpactful, noted that newsrooms can increase audience trust by communicating about their uses of AI clearly and effectively. 

As CNTI’s global AI Research Working Group has written, there is not one rule book or exact labeling technique that newsrooms should follow. Instead, audiences, like journalists, are still getting used to AI. What matters most is that journalists and their organizations are transparent about their use of AI and carry out a dialogue with audiences about what that means.

It is not enough to simply have an AI policy or guideline, Erjon Curraj, a digital transformation expert from Albania, cautioned. Newsrooms must implement these policies consistently and ensure their staff is aware of them, too.

“This isn’t just about disclosure,” Lovre concluded. “This is our unique chance to lead by example. Social media networks are overwhelmed by AI-generated content, and if we don’t try to explain to our audience what is happening, we won’t be in a good place in one or two years.” 

Journalism Producers Need to Pay Close Attention to Public Policy Related to AI

Participants also discussed how policies and government regulations can impact media freedom and ethics. 

Ana Toskić from Partners Serbia and Emily Wright from CNTI highlighted the potential impacts of EU digital laws, especially the AI Act and the Digital Services Act, on journalism in the Western Balkans. They discussed how in closed or partially closed media environments, where journalists rely on social media platforms to share stories, laws like the Digital Services Act will have a significant impact, especially when regulatory bodies do not operate separately from the government and can misuse the laws to stifle independent reporting. 

The pair further explained that media organizations located within the EU, and those in EU candidate states, will have to adhere to the General Data Protection Regulation (GDPR) and the AI Act when using AI in their operations. 

Gábor Kardos, CEO of the Hungarian publisher Magyar Jeti Zrt., emphasized the importance of staying on top of regulatory activity, especially in autocratic-leaning countries where independent media is the minority amongst state-controlled media. 

“Even if these regulations were created perfectly…. the state will always have the power to abuse them. And that’s the reason I’m advocating for better regulation,” Kardos said. “But we, as publishers, need to be aware that that’s not the thing that’s going to protect us. We as a community have to protect each other and ourselves, and be innovative and be faster than regulation can ever be.” 

Journalists Can Use AI to Help Address Challenges in the Information Environment 

In an era where press freedom is at risk, synthetic content floods our feeds and deepfakes grow more convincing by the day, journalists face an unprecedented challenge. Journalists are not just competing for attention anymore; they’re fighting for the very concept of verifiable truth.

The twist is that while AI can exacerbate some of the problems, it can also be part of the solution. Workshop participants recognized that AI tools could help newsrooms verify information faster and more thoroughly than ever before, allow them to reach new audiences and help them create new forms of content. The key lies in thinking ahead. 

“If we want to think strategically, we cannot focus only on the shortest term,” Kardos said. “It [the short term] does not matter if in five or seven years, it’s not journalism or AI, it’s humanity that will be in question. We need to focus on what happens in the midterm, within a few years.” 

Participants from several countries offered case studies of creative ways they have implemented AI into their workstreams, even on small teams with limited budgets that face pressure from their governments. 

The Center for Investigative Journalism of Serbia, which has a team of only 10 people, created a custom large language model (LLM) to analyze nearly 10 million pieces of data on wait times in Serbia’s healthcare system. The AI system allowed Ivana Milosavljević and her fellow reporters to analyze large amounts of data in record time. She noted, importantly, that a human reviewed all outputs to ensure accuracy, which was time-consuming, but she shared that AI enabled the team to reach and visualize conclusions in new and efficient ways. Initially tested on publicly available data, Milosavljević said the LLM system will be especially valuable for analyzing confidential data.

Ján Trangel, AI implementation lead at Ringier Slovakia, explained that the outlet has created an AI-driven hate speech moderation system after finding third-party tools ineffective for their language and regional needs. Ringier built custom hardware and an offline LLM — drawing on open-source options like Mistral and Google models — to evaluate messages in real time, classify their severity and support fully customizable moderation policies. The system processes more than 300,000 messages daily, reducing manual review while offering an admin panel for managing labels and decisions. Its offline architecture lets the team experiment freely, compare models and tailor features without depending on external cloud services. 

Vidi Vaka, a Skopje-based outlet with only three full-time reporters, created “KiberFlow,” an AI coworker that transforms the outlet’s reported stories into rap-style videos. KiberFlow uses character-animated performance and social satire to highlight everyday societal problems and has more than 1,000 followers on Instagram. The project is allowing Vidi Vaka to reach a younger audience that usually avoids traditional media in a new and engaging way. 

“For us, AI is not a shortcut; it’s a collaboration,” journalist Angela Petrovska said. “People follow KiberFlow not because it’s AI, but because it tells real stories made by good journalists.”

At Dennik N in Slovakia, Munk and her team explore AI uses on a smaller hiking website they manage. Every Thursday, the team publishes a set of recommended weekend hikes, using AI to assist with suggestions based on weather forecasts and difficulty levels. They have also developed tools for automated image cropping and social media posting. While not all types of tools are tested there, the hiking site provides an environment for experimenting, learning and identifying what might be useful before expanding these solutions to larger platforms at the different outlets in Denník N network.

Participants explained that AI tools are often easy to learn and improve creative flow, but they noted that human editing, cultural context and emotional nuance remain essential to keep the work authentic. Overall, they expressed optimism about the opportunities that AI tools can provide their newsrooms and left the workshop feeling motivated to see themselves as innovators in the news industry.

Conclusion

There’s something fitting about having this conversation in Sarajevo, a city that knows something about resilience in the face of existential challenges. The participants didn’t offer easy answers or technological determinism. Instead, they charted a middle path, one that takes AI seriously without surrendering the core values that make journalism essential.

The future won’t be built by those who reject AI wholesale or embrace it uncritically. It will be built by newsrooms that approach these tools with clear eyes, strong ethics and an unwavering commitment to serving their audiences with verified, trustworthy information.

The post AI for Sustainability: Building Journalism’s Future appeared first on Center for News, Technology & Innovation.

CNTI’s AI and Journalism Research Working Group looked at 55 research studies and other articles from computer science, social science and linguistics disciplines to better understand how AI is shaping transcription and translation and what these developments mean for journalism. These studies include data representing a range of geographic contexts and languages.

Artificial intelligence (AI) systems are increasingly being used for transcription — the process of converting audio to written form — and translation — the process of converting content in a source language to a target language. Journalist-facing transcription tools like Otter and Trint launched nearly ten years ago,¹⁰ and automated translation has been available to everyday users — at least for a limited set of languages — since the release of Google Translate in 2006.¹¹ Advances in AI systems mean that Google Translate and similar programs continue to improve rapidly.¹² However, journalists around the world do not uniformly experience the benefits of AI technologies in assisting with transcription and/or translation. Inconsistent access to technology and varying availability of high-quality and verified training data remain major challenges.

Overall, AI models are improving. However, transcription tools remain most accurate for a relatively narrow range of standard American English dialects and accents,¹³ and translation tools remain most accurate for only a few language pairs.¹⁴ There are significant gaps in performance for “low-resource” languages (i.e., languages with relatively little textual data online that can be used to train AI models) as well as concerns about accuracy in those languages. It should be noted that “low-resource” languages include a number of languages spoken by hundreds of millions of people. In fact, English is the language of about 50% of online content; the largest proportion of online content that any other language represents is 6%.¹⁵ Thousands of languages make only a minuscule imprint on the internet.¹⁶ Some of the ongoing efforts to close the divide between “low resource” and “high resource” languages focus on creating and improving training data. For example, Nigerian start-up Goloka is working with Meta to collect data in five Nigerian languages¹⁷ as part of a larger Meta-UNESCO partnership.

AI translation and transcription are rapidly developing areas of research. Nearly all the research articles cited in this briefing are from the last five years, and machine translation and transcription are steadily improving. The working group expects to see continued progress in building AI tools for these uses, especially for languages that have not received as much attention. However, there is also evidence that unverified AI translations are creating misinformation on sites like Wikipedia, and those low-quality translations are being used to train the next generation AI translation tools, leading to worse performance.¹⁸

Journalism Use Cases

Journalists regularly use AI tools for transcription and translation to assist in the production of news content, generally with human oversight over the process.¹⁹ Journalists also use transcription tools, such as the Houston Chronicle’s Meeting Monitor, to share summaries of government meetings²⁰ with the public. Large-scale automated transcription and translation tools, including A European Perspective and Dubawa, are assisting journalists to report on topics that would otherwise be difficult and time consuming to monitor. A European Perspective consists of 10 broadcasters across nine European countries that exchange content using AI transcription and translation.²¹ The automatic transcription of audio-visual content and translation between languages encourages greater coverage of European news topics while also using editorial oversight to correct cultural or linguistic details in outputs. Dubawa, an AI fact-checking system in Ghana and Nigeria, was specifically trained using local dialects and accents. The tool transcribes radio broadcasts and checks for mis- and disinformation in several local languages.²² In a similar vein, Paraguayan news outlet El Surti is building a community-based Guaraní language dataset and AI tools.²³

The degree to which AI systems have been implemented for transcription and translation depends on newsroom resources and languages in use. Through a series of semi-structured interviews in South Africa, one study finds that AI tools are mostly being integrated in larger newsrooms and used to aid in public-facing translation, particularly at public media outlets with mandates to publish in multiple official languages.²⁴ Still, journalists — especially those in the Global South — are concerned about the utility of AI tools for transcription and translation, given reported challenges with accents and lower accuracy for local languages.²⁵

Newsrooms find that using AI tools for translation saves time but can result in inaccuracies, necessitating human review, especially in audience-facing content. One study found that AI translations of international news in Tanzania were mostly accurate, but about 13% of the sentences included mistranslations, minor ambiguities or inaccuracies that missed cultural details, such as incorrectly translating the English phrase “street food” word-for-word as “food of the road” in Kiswahili, rather than using an idiomatic phrase.²⁶ Similarly, a 2023 study benchmarking AI transcription and translation tools for journalists also found that they can save considerable time even with human review, but translations into English perform better than other languages.²⁷ In general, hybrid translation, in which machine translations are reviewed by human experts, is a promising approach.²⁸

Translation and transcription tools can also be used to personalize news content for different audiences. For example, publishers can add widgets to their websites or apps that would allow audiences to access an automatically generated transcript of video or audio content, or to access all content in the language of their choice. Publishers are increasingly interested in providing these features.²⁹ However, publisher interest is currently far outpacing audience interest: according to the 2025 Reuters Digital News Report, 65% of publishers were actively exploring AI translation content but only 24% of audiences said they were interested in using AI translation.³⁰ Similarly, 75% of publishers were exploring making audio available in text format (and vice versa), but only 15% of audiences expressed an interest. The reasons for these gaps are beyond the scope of this report, but researchers have suggested they may be linked to a lack of awareness of what the features might look like, or to a broader audience distrust of AI in journalism.³¹ However, audiences are more comfortable with AI translation than with many other newsroom uses of AI.³²

What Level of Accuracy is Good Enough?

One question that has not been addressed by research, is what level of accuracy is good enough? For example, is it appropriate to use AI tools if translation and transcription reach a certain level of accuracy? Even a tool with 95% accuracy may still miss crucial cultural and language nuances. Thus, human review and revision are necessary to ensure accuracy and appropriateness for most public-facing content. Future research examining the value of different strategies for improving accuracy and appropriateness, such as diversifying sources of training data, metadata, language-specific models or algorithms, can help determine how to address accuracy standards. However, research can only inform what is, at its core, a value judgment.

Technical Evaluations of AI Transcription and Translation

Models for these types of tasks are rapidly improving,³³ and translations between certain language pairs — like Spanish and English — generally perform well.³⁴ Researchers find that while AI tools cannot currently handle cultural details and ambiguity as well as human experts can, they are promising in a number of other contexts, such as (1) translations of medical terms from English to German³⁵ and (2) translations of legal documents across Arabic and English.³⁶ Researchers also find that AI translations from Indonesian to English often rival those of students in translation educational programs when it comes to implementing techniques like paraphrasing and structural transposition.³⁷ Meanwhile, AI transcription is being used to increase coverage of government meetings.³⁸ Summarizing these meetings may be a particularly fruitful use of AI transcription because participation follows a consistent structure and because figurative language and wordplay are rare. Progress in this field continues as research and development of novel techniques to build training datasets advances and as the development of specific translation and transcription models receive more attention.³⁹

While AI transcription and translation technologies are improving, recent research also highlights limitations and shortcomings of these AI tools⁴⁰ — including those particularly relevant to professional fields like journalism.⁴¹ These limitations include the tools’ (1) inability to fully handle language ambiguity and cultural nuance, (2) struggle to perform tasks at the level of human experts and (3) biases in outputs based on personal characteristics and attributes present in speech and text data. Evaluating the quality of these tools is challenging in itself: some metrics are overly simplistic, while more holistic methods for evaluating quality or accuracy are opaque and hard to interpret.⁴²

AI translation tends to focus on words rather than meaning, but languages have a great deal of typological variation and do not necessarily have parallel sentence structure. Moreover, word-level translation often focuses overly on referential meaning (i.e., what something is about) and lacks attention to indexical meaning (i.e., the social functions of language). Languages do not align one-to-one, with words carrying different formality and/or emotional meanings; AI tools may use the incorrect word to reflect the perspective of the speaker.⁴³ For example, professional speech is less formal in English than in Korean⁴⁴ or Japanese,⁴⁵ and a more literal translation into those two languages will often be socially inappropriate. The outputs these systems produce also change depending on how the translation is described through prompting, such as using specific requests to retain key themes from the source language versus merely asking for a translation into a given language.⁴⁶ 

There is also evidence that AI translation models produce results with gender biases⁴⁷ (though these biases are diminishing as models improve) by consistently assigning gender to professions (e.g., assuming doctors are men and nurses are women).⁴⁸ A review of 133 studies finds that much research on this topic treats gender bias as a purely technical or linguistic problem, rather than examining its social dimensions; these authors also note that most of the studies used machines rather than people to evaluate bias.⁴⁹ Given the social nature of bias, the authors raise concerns about this evaluation method. In practice, these studies suggest that journalistic content that is translated without careful review may inadvertently produce outputs that include biased pronouns, occupations or perspectives stemming from the AI tool’s training data.

AI transcription tools carry their own limitations, such as a tendency to add content that was never said by the source. This type of error appears more commonly when transcribing speech with longer gaps between words and phrases.⁵⁰ Strikingly, there are also critical deficiencies in these tools when transcribing audio from people who speak any form of English besides a fairly narrowly defined set of standard American accents, such as both World Englishes⁵¹ and African American Vernacular English.⁵² 

Overall, a recurring theme in the literature reviewed by the working group was that AI tools for transcription and translation of “low-resource” languages are severely lacking.⁵³ There are significant gaps between human-written and LLM outputs in languages other than English,⁵⁴ with journalists in the Global South reporting less confidence in these tools than those in the Global North.⁵⁵ Among “low-resource” languages, machine translation for signed languages lags even further behind spoken ones.⁵⁶ The most used signed language data sets are small and frequently rely on interpreted data, which is likely to include considerable interference from spoken languages.⁵⁷ It is not yet clear how best to represent signed languages computationally, nor how best to evaluate translation between signed and spoken languages.⁵⁸

Although there are many challenges for AI transcription and translation, greater attention is being focused on “low-resource” languages than before. Two prominent examples, Masakhane and Dataphyte, seek to improve natural language processing (NLP) research across Africa. Yet more needs to be done, including designing tools for local settings — particularly in locations that have thus far received less attention from technology companies and AI developers.

Where More Research Would Be Helpful

• Day-to-day use: The research does not yet include a deep understanding of where and when journalists are using AI tools for transcription and translation, nor does it explain where and when they would like to once they feel adequate tools are available. What specific issues are the journalists finding? How might these issues be addressed? 
• Good-enough accuracy: What level of accuracy is good enough for journalistic content?⁵⁹ For example, is it acceptable if AI translation tools achieve 95% accuracy when handling text between two languages? The incorrect 5% may provide essential cultural context for audiences in the target language. Research can help inform news organizations and journalism providers to decide what they are comfortable with as current tools are unable to achieve 100% accuracy in every context. Journalism-specific benchmarks for assessing AI transcription and translation tools are also worth developing to address industry-specific needs.
• Languages studied: The research on bias in machine translation has been limited to relatively few languages — with a bias toward written texts — and has focused primarily on translation into English.⁶⁰ It also primarily focuses on gender bias in isolation from other social identities and contexts. Expanding research in this area would help news workers make informed decisions about when AI translation is appropriate and when the risks are too high.
• Downstream effects: There is little research about the downstream effects that bias in the tools might have on journalism. For example, if journalists are working under time pressure, are they less likely to interview sources whose voice automated tools do not transcribe as well? Are journalists identifying and editing gender bias in translation tools, or is it impacting their reporting, their audiences’ understanding or both?
• Validated data in more languages: If the goal is to use large language models (LLMs) for transcription and translation, considerable attention and effort needs to be placed on building robust LLM training data in non-English and low-resource languages.⁶¹ We need more high-quality translation training data⁶² that has been validated in both the source and target languages. Further research should examine how to do this most effectively and efficiently to create inclusive AI models.
• Third-party tools: Larger news organizations are developing in-house AI models, particularly adaptations of smaller language models that can be run entirely locally.⁶³ Meanwhile, many smaller newsrooms are relying on pre-existing models from technology developers. More research needs to consider the potential impacts of relying on third-party tools,⁶⁴ and explore how smaller newsrooms can (1) adapt (fine-tune) existing models to better fit their needs and/or (2) build custom AI tools that are financially viable for them.

Footnotes

1. Mari, 2024 ︎
2. Ananny & Pearce, 2025; Beckett & Yaseen, 2023; Gondwe, 2025; Kahn, 2025; Munoriyarwa et al., 2023; Simon & Isaza-Ibarra, 2023 ︎
3. Qingliang; 2024; Shahmerdanova, 2025; Tokalac, 2023 ︎
4. Spencer, 2025 ︎
5. Court & Elsner, 2024; Kocmi et al., 2024; Kocmi et al. 2025; Pava et al., 2025; Moghe et al., 2025 ︎
6. Chan et al., 2022; Chang et al., 2024 ︎
7. Ghosh & Caliskan, 2023; Prates et al., 2020; Savoldi et al., 2025; Ullmann, 2022 ︎
8. Fredrikzon, 2025 ︎
9. Dubois et al., 2024 ︎
10. Jarnow, 2017 ︎
11. Frey & Llanos-Paredes, 2025 ︎
12. Caswell & Liang, 2020 ︎
13. Chan et al., 2022; Chang et al., 2024 ︎
14. Court & Elsner, 2024; Kocmi et al., 2024; Kocmi et al. 2025 ︎
15. W3Techs, n.d. ︎
16. Brandom, 2023 ︎
17. Orimemi, 2025 ︎
18. Judah, 2025 ︎
19. Ananny & Pearce, 2025; Beckett & Yaseen, 2023; Simon & Isaza-Ibarra, 2023; Canavilhas, 2022; Ohumu, 2025; Simon, 2025 ︎
20. Langer, 2025 ︎
21. Canavilhas, 2022 ︎
22. Ohumu, 2025. See Vo 2025 for another example. ︎
23. Valdez Sanabria & Auyanet, 2025 ︎
24. Munoriyarwa et al., 2023 ︎
25. Beckett & Yaseen, 2023; Munoriyarwa et al., 2023; Kahn, 2025 ︎
26. Gondwe, 2025 ︎
27. Tokalac, 2023 ︎
28. Shahmerdanova, 2025 ︎
29. Newman & Cherubini, 2025; Ross Arguedas, 2025 ︎
30. Ross Arguedas, 2025 ︎
31. Ross Arguedas, 2025 ︎
32. Simon et al., 2025 ︎
33. Breaking Language Barriers with AI: Maximizing Accuracy and Efficiency with Machine Translation Technology, 2023 ︎
34. Alonso Jiménez & Rosado, 2024 ︎
35. Noll et al., 2025 ︎
36. Moneus & Sahari, 2024 ︎
37. Asi et al., 2024 ︎
38. Langer, 2025; Schellmann, 2025 ︎
39. Guo et al., 2025; Kocmi et al., 2024; Kocmi et al. 2025; Levit et al., 2017; Moghe et al. 2025; Wang, 2022 ︎
40. Lee, 2024; Novytska et al., 2025; Yan et al., 2024 ︎
41. Schellmann, 2025; Song, 2020 ︎
42. Leiter et al., 2024 ︎
43. Song, 2020 ︎
44. Song, 2020 ︎
45. Lee, 2024 ︎
46. Lee, 2024 ︎
47. Savoldi et al., 2025; Ullmann, 2022 ︎
48. Ghosh & Caliskan, 2023; Prates et al., 2020 ︎
49. Savoldi et al., 2025 ︎
50. Koenecke et al., 2024; Schellmann, 2025 ︎
51. Chan et al., 2022 ︎
52. Chang et al., 2024 ︎
53. Ojewale et al., 2025; Ojo et al., 2025; Pava et al., 2025 ︎
54. Guo et al., 2025 ︎
55. Beckett & Yaseen, 2023 ︎
56. Wolfe et al., 2025 ︎
57. De Coster et al., 2024 ︎
58. De Coster et al., 2024 ︎
59. How best to evaluate translation is largely out of scope, but see Leiter 2024. ︎
60. Savoldi et al., 2025 ︎
61. Howcroft & Gkatzia, 2022 ︎
62. Qingliang, 2024 ︎
63. Hagar et al., 2025 ︎
64. Simon, 2024 ︎

The post AI Transcription and Translation in Journalism appeared first on Center for News, Technology & Innovation.

Newsgeist is an unconference: the agenda is not pre-planned. On the first night, the participants shared session ideas — ranging from journalistic neutrality in wartime to what journalism can learn from K-Pop to investigative journalism with TikTok DNA — which were then organized into an always adaptable mix of choices. Newsgeist operated under the Chatham House Rule. There were no “speakers,” only session moderators to help manage the flow of conversations.

The takeaways below don’t represent the full range of everything discussed at Newsgeist, nor do they suggest a consensus among participants; instead, we’ve identified broad themes that arose across the weekend.

Innovation in Journalism

Over the weekend, a recurring theme was the need to push the industry’s thinking around new approaches and products for public relevance and business success. Many participants also shared in the acknowledgement that journalism has not historically taken well to adaptation and technological ideation. It will take real effort for the industry to become more innovative — and it may require traditional newsrooms to adopt some characteristics that are more natural to newer, creator-type journalists.

Several participants remarked that the news industry can learn from the technology industry when it comes to rapid experimentation, shared learning, mentorship and data testing — skills that “traditional” journalism is not known for but that can help foster new voices and approaches that better serve today’s consumers.

The news industry can also better adopt and internalize new technology to enhance and empower the work of journalists in ways not before possible. This may be even more pertinent in the age of AI: “In the age of advertisements, the metric of value exchange was the impression. In the era of traffic, the metric of value exchange was the click. In the era of generative AI, the metric of value exchange is the fact,” one participant said. Participants did not agree on the extent to which the news industry should use AI, noting that the industry needs to better understand both the value and limitations of AI and other emerging technologies in order to be more strategic, and less reactive, to innovation. There was agreement, however, that journalism must act and figure out how to be part of the change.

Other innovative ideas focused on things that bring communities together, such as ways to crowd-source local needs, needs-based audience surveys and more. And some wondered whether we need to rethink the structure of journalism itself, asking “How do we support journalists who operate outside any kind of institution?”

Journalistic Independence and Safety

Journalists are facing increased threats around the world — and newly so in the United States — including deportation, violence, government-imposed restrictions and harassment from people in power, bringing journalistic freedom and safety top of mind for many participants. Several participants emphasized that these threats are not a future issue; they are happening now, and journalists need to be prepared to respond.

The group shared many ideas for ways to respond, such as maintaining decentralized networks to protect journalists, sources and content; creating more solidarity and public awareness around journalism’s work; building relationships throughout a community in order to reach all members when reporting on threats to health and safety, such as pandemics; emphasizing journalism’s role as giving a voice to the voiceless; and consistently covering attacks on free expression and press freedom. Several people also made the point that journalists in the United States have much to learn from journalists in other countries who have or are currently experiencing threats against press freedom.

We were also reminded that media independence goes beyond independence from government interference and control. Participants shared instances where ownership structures and funding reliance hampered journalistic independence — highlighting the need for journalism producers to build strategies for self-sustainability.

Trust, Truth and Activating the Audience

Newsgeist featured powerful discussions around truth and facts, what they each mean and how they fit together with trust. Information is always filtered through personal experience; what feels true to one person may not feel true for someone else. As one participant said, “We need to ask ourselves who we’re willing to hear the truth from, and who our audience is willing to hear it from, and why.”

Questions about trust and truth came up across our sessions:

• Can a slate of more diverse viewpoints lead to greater trust in news organizations? What does doing that authentically require?
• Did Jimmy Kimmel’s experience show that the public will activate when corporations or the government attempt to censor or ban content?
• How do journalists demonstrate that they can respect their audiences who might be skeptical or negative towards them? How best to hear and engage with those communities?
• Is it possible for corporate owners and publishers to support their newsrooms in disputes over editorial control?
• How can news organizations do a better job of supporting one another when one among them is treated unfairly?

Participants broadly agreed that reporting the truth is the answer to regaining public trust — but as one person noted, “Truth by itself is not a culture. If we want to create kinder communities, how would we report differently?” This openness to rethinking how the audience receives information will be key to rebuilding trust, but it will require a rethinking of how journalism is practiced.

Articulating Journalism’s Value

Does journalism need a marketing campaign? A theme across several sessions was the need to do a better job of articulating journalism’s value, but to be able to do that, journalists need to first articulate for themselves what that value is and how it is changing. This is not a new problem, as CNTI’s research has shown, but it has become even more urgent today.

While the general consensus was that journalism should still mean verifiable, fact-based reporting, participants acknowledged that such journalism is now happening across a variety of platforms and from a larger number of players than ever before. As one participant put it, “Journalism with a capital J isn’t intrinsically important anymore,” explaining that a small number of traditional outlets that were seen as authoritative sources are no longer the only way people get news. Instead, many people are turning to non-traditional outlets and news producers with a specialized focus or who target local and underserved audiences. In this changing environment, journalism’s value must go beyond just updating audiences, several participants said; news practitioners must be able to demonstrate how journalism is helping news consumers live their daily lives.

Some participants also argued that rather than try to prevent people from going to social media, individual brands or AI to get informed, journalists could instead explore what makes these tools attractive, valued choices: What role do these tools serve, and what can journalism producers learn from that? The public will continue to turn to a mix of tools and sources for their information, and, in many cases, they will prioritize sources that give them some kind of shared experience instead of a top-down rendering of the day’s news.

CNTI gives our thanks to the financial sponsors of this Newsgeist, Google and The Knight Foundation, and to everyone who came and created such a rich 40 hours. We hope the conversations continue, both in and outside of Newsgeist.

The post CNTI Newsgeist 2025: Trust, Truth and Innovation in a Shifting Industry appeared first on Center for News, Technology & Innovation.

This is the first in a series of reports from the AI and Journalism Research Working Group convened by the Center for News, Technology & Innovation (CNTI). The working group currently consists of more than 15 cross-industry members from around the world, bringing research, journalism and technology experience to the discussions. Each quarter, we’ll synthesize the state of global research across two or three questions or topics at the intersection of journalism and AI.

The goal of the working group is to summarize research for an audience of journalism practitioners and researchers around the world. That means we focus on what’s actionable for journalism, not other fields that are concerned with AI.

What do we mean by “AI”?

This report uses the OECD definition: “An AI system is a machine-based system that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments. Different AI systems vary in their levels of autonomy and adaptiveness after deployment.”

Wherever possible, we try to use specific terms rather than “AI” to avoid conflation or confusion. Journalism has been adopting forms of automation for more than 50 years,¹ but widespread use of the term “AI” is more recent — and may include both newer technologies and those that have been in use for quite some time.

The Focus of This Report

This briefing focuses broadly on journalistic communication about AI. We first summarize research and frameworks that address the following questions about societal AI literacy:

• What does the public currently know about AI, and what does the public need to know about AI?

We then move on to a synthesis of current research on journalists’ communication about AI in two different contexts:

• How should journalists and news organizations communicate about journalistic uses of AI?
• How do journalists report on AI and automation more broadly, and what are the consequences for their audiences?

In each section, we lay out the general findings of the research to date, any suggested considerations or actions for practitioners that emerge and areas where more or new research is needed. A clear takeaway across the board is that journalists and news organizations need to communicate more clearly about journalism itself — not only about AI.

This report was prepared by the research and professional staff of CNTI in partnership with several external contributors who collectively authored this briefing.

If you have ideas or research findings that are important for CNTI and the working group to include, please email them to info@cnti.org.

What the public knows, and needs to know, about AI

There is increasing consensus across fields that the public requires more knowledge about AI — including about different types of systems, what these systems can and can’t do and how to critically assess their output.² Despite the publication of several systematic reviews, there is not yet a consensus on exactly what so-called “AI literacy” should include, other than that it should cover both computational and ethical dimensions.³ Nor is it clear how to include “AI literacy” in broader conceptions of media and information literacy.

While the field of journalism can’t be expected to address AI literacy alone, news reports are one of the primary ways that adults learn about new technologies.⁴ As a consensus begins to emerge about what people should know, the role of journalism in AI literacy may also become clearer.

Global Perspectives

The way AI development, use and literacy are playing out in the Global North and the Global South varies widely. What journalists know and need to know — before we even consider the broader public — simply does not look the same in different political and economic contexts with variable infrastructure and educational systems. 

Working group member Oluwapelumi Oginni shares her perspective about AI and journalism in Africa:

“While AI adoption in journalism is becoming a well-explored topic in the Global North, the conversation in the Global South still requires more exploration, especially within the context of the real on-the-ground barriers around skills and resources. The discussion is not just about whether journalists want to use AI, but whether they can. Effective use of AI in newsrooms depends heavily on two things: first, journalists and editors need to understand what these technologies can do (AI literacy), and second, they need the financial capacity to implement them. In many African countries, and indeed, in the Global South, that second part is a serious roadblock…⁵

[W]hile the idea of AI in journalism is gaining ground, the infrastructure, funding, and training needed to make it work at scale are still very much missing.⁶ What this means, ultimately, is that journalists and media houses in the Global South, particularly in Africa, are operating at a double disadvantage. Though they are expected to keep pace with global innovation cycles, many lack the financial means and technical training needed to truly compete or drive innovation. This makes it all the more important for research to explore, in depth, the uneven realities of media systems operating in low-resource environments.”

The education field offers a popular framework⁷ that breaks down AI literacy into four components: 

1. Literacy practices (the ability to understand and evaluate AI, which people can demonstrate directly) 
2. Core values (which support learners to use the tools safely and effectively) 
3. Modes of engagement with the tools (understanding, evaluating and using AI tools) 
4. Types of use (the purposes for which learners use AI tools) 

Under this framework, journalism may be well suited to help adults build literacy practices (rather than core values or specific types of use). Literacy practices include holistic and fundamental knowledge that crosses social and technical domains. However, it is unlikely that journalism can — or should — directly teach adults how to use AI tools.

Within the realm of literacy practices, some concepts are more fundamental than others. A 2024 synthesis⁸ highlights three key facets that differentiate AI literacy practices from other technological literacies:

1. First, AI is (relatively) autonomous. It can act without human intervention and sometimes have material impacts on the world without humans knowing it was used. 
2. Second, AI can learn. Its input data can enable automated improvement.
3. Third, AI is “inscrutable” – deeply challenging to understand or interpret. AI inference is not analogous to human reasoning. For the most part, its models and decision-making processes are opaque or “black-boxed.” They may only be understandable to a point, or only to people with advanced technical knowledge. That, in turn, can also make it hard to interpret outputs.

The authors of this synthesis also note that AI presents two additional qualities that violate long-held assumptions of human-technology interaction: 

1. AI has inconsistent outputs. Because models are increasingly probabilistic rather than deterministic, the same input will not always lead to the same output. For example, a calculator will give the same results to the same query every time; an algorithmic search engine may not. And ongoing changes and improvements to models may lead to inconsistencies over time.
2. It is opaque. Because AI is “under the hood” of so many technologies and has no specific interface, users may not even know when they’re interacting with it.

These five aspects of AI sit at the intersection of understanding and awareness, where journalism is best poised to intervene. For example, these high-level points could be addressed through evergreen explainers and sidebars that outlets reuse across stories on AI.

The field of journalism may also be well positioned to help people understand how content delivery algorithms work. Understanding these algorithms can help people increase their agency over information, inform citizenship and strengthen community building, among other outcomes.⁹

Research we reviewed suggests

It is important for anyone explaining AI to focus on the humans behind AI systems and remind audiences that computers and humans have different and complementary strengths.¹⁰ In particular, “AI is a tool that uses existing data to make predictions or generate content. In other words, it is creating a best approximation … based on what has already been created by humans.”¹¹

Aspects of AI literacy that may be most relevant for journalism

Journalism plays a part in supporting adults in learning about new technologies, including “AI literacy” and “algorithmic literacy.” It is important that the approach to coverage fully realizes and thinks holistically about this role.

Any given outlet’s audience will have different background knowledge and different information needs. Journalists and outlets will need to ensure they understand their particular audience, rather than developing a universal blueprint to explain emerging technologies.

Journalism may be well suited to connect the dots between technical explanations and social and ethical impacts,¹² since they frequently draw similar connections in their work. On the other hand, journalists may not be as well positioned to train the public in how to use AI tools, since experiential learning is likely to be more effective.¹³ 

Journalists are also in a strong place to report on funding, business models, government relationships and other political factors behind the development of AI and other new technologies. 

Where more research would would be helpful

Before research can help journalism support public AI literacy, we need better answers to three big questions:

• What do people need to know about AI? (Relatedly, researchers should also consider what people currently know about AI so that efforts can meet the public where they are at this time.)
• What can journalism realistically teach people about AI?
• What are the impacts current reporting on AI has on AI literacy?

Frameworks and validated measures for AI literacy — which will likely come from educators and psychometricians — would make it easier to measure the efficacy of particular reporting strategies on public AI literacy.¹⁴

Communicating About AI Use In Journalism

“Transparency” and “disclosure” are buzzwords in the AI and journalism space these days: dozens of toolkits and guidelines advise news organizations on their importance vis-à-vis the use of AI, and how to best communicate about its use. But there is much less agreement on the goals behind being transparent about AI use (including where and when it is needed), the value to audiences and what works best to improve audience understanding and awareness. Members of the public consistently say that they want transparency about AI use,¹⁵ but current data offers, at best, a mixed sense of what that actually means in practice. Given both the range of use cases for journalism and the need for better public understanding of both AI and journalism, we need more data on how the public uses and interprets disclosures in context. 

Global Perspectives

Several surveys of journalists suggest that Global South journalists are highly likely to adopt AI tools and other new technologies.¹⁶ One recent survey found that more than eight in ten Global South journalists say they use AI for work, but less than one in five say their organization has policies about AI use.¹⁷ 

Working group member Zara Schroeder shares her perspective: 

This gap highlights a pressing need for institutional guidance, training, and ethical frameworks to support responsible adoption.¹⁸ Without such policies, journalists may inadvertently compromise accuracy, transparency or privacy, especially in high-stakes reporting environments.¹⁹ 

Moreover, disparities in access to resources, infrastructure and AI literacy between Global South and Global North newsrooms may exacerbate existing inequalities in news production and distribution.²⁰ Journalists in under-resourced contexts may rely on freely available generative AI tools without fully understanding their limitations, biases or risks of misinformation. ²¹

For example, in countries like Zimbabwe, Uganda, Kenya and South Africa journalists are using AI tools to automate time-consuming tasks such as generating basic news reports, transcribing interviews or summarizing press releases.²² Off-the-shelf tools like ChatGPT, Otter.ai (for transcription) and Google’s AI-based-speech-to-text services are being used to speed up workflows, especially in fast-paced newsrooms with limited staff.²³

There is also a lot that the Global North can learn from homegrown African AI projects.

Journalists in Ghana and Nigeria have used machine learning models to help analyze large public datasets, such as government budgets, procurement records or COVID-19 case numbers to identify patterns, anomalies or potential corruption.²⁴ AI-driven data analysis tools help uncover insights that would be difficult to detect manually, enabling journalists to produce in-depth, evidence-based stories. 

Dataphyte, a Nigerian media and data analytics organization, launched Nubia, an open-source AI platform in 2022 designed to analyze large datasets and generate first-draft news stories.²⁵ Users input raw data (e.g. government statistics), and Nubia produces narrative reports, templates, data breakdowns and visuals which journalists then refine and contextualize. This significantly speeds up complex data-driven reporting. 

Another example is Dubawa, a West African fact-checking initiative, which has developed two AI-powered services; one is a WhatsApp chatbot that responds to user queries by drawing on previously verified fact-checks from Dubawa and the International Fact-Checking Network.²⁶ The second AI service is an audio platform that monitors live radio broadcasts, transcribes audio in Nigerian and Ghanaian English dialects, identifies suspicious claims and helps journalists verify or debunk them.²⁷ 

Overall trust in journalism and in the broader information environment also varies widely by region, as does knowledge about journalism. That makes it particularly difficult to generalize from studies that focus on a small number of countries. Local political pressures, media ownership structures and digital access all influence how AI tools are used and received by both journalists and audiences. 

To build a more comprehensive understanding, future research should incorporate regionally diverse case studies and prioritise collaborations with local media organizations. This can help ensure that the global AI transition in journalism is inclusive, context-sensitive and responsive to the needs of journalists working in a wide range of environments. 

In assessing this area of research, we considered over 20 studies related to AI transparency and disclosure, the full list of which can be found in the References section. Importantly, the discussion here focuses on transparency within the journalism industry, rather than transparency on platforms (e.g., Apple News) and social media sites (e.g., TikTok). There has been considerable research on transparency, especially compared with other questions related to AI and journalism, allowing for a more conclusive set of takeaways. However, it is important to note that the analytical methods, geographical ranges and specific research questions vary widely, from surveys and experiments to in-depth interviews with journalists or news audiences. 

A number of studies have explored whether labelling could further harm societal trust in information, but the results have varied with even small differences in label text. For example, one experiment found that labeling content “AI-generated” reduces audience ratings of its accuracy but not their trust in news or journalists.²⁸ Another found that a longer label specifying autonomous AI reduces perceived trustworthiness but not accuracy.²⁹ A third study suggests that some of these impacts may decrease as people become more familiar with these technologies.³⁰ Moreover, negative consequences of disclosure could align with other structured forms of bias. One recent study found that while both humans and LLMs rated identical news articles less favorably when an AI disclosure was present, the LLMs also penalized authors from particular racial or gender groups disproportionately for their transparency.³¹

Several studies (using different methods) find that audiences tend to overestimate the prevalence and autonomy of AI in journalism.³² Additional studies find audiences do not consistently or accurately interpret the intended meaning of AI attribution in bylines (e.g. “written by staff writer with artificial intelligence (AI) tool”) or simple labels.³³ Several studies also suggest that people with deeper knowledge about journalism have more concerns about AI use, perhaps because they better understand verification practices.³⁴ But offering excessive information can have negative impacts; while there is not much research on this aspect of AI transparency, a number of studies on privacy disclosure policies have explored “privacy fatigue” and disengagement.³⁵

Almost across the board, the research we reviewed suggests that many people lack a nuanced understanding of journalistic processes and principles. And recent data CNTI collected shows that journalists themselves recognize the need to do a better job communicating the field’s value in general.³⁶ It is nearly impossible to envision successful communication about technology tools that does not reckon with this larger challenge. After all, a major goal of journalism is to provide the public with reliable information in order to support civic participation, public life and democracy. And the norms of production that seek to make journalism reliable, such as independence and verification, also underlie journalists’ use of new technologies — these norms need to be better communicated to achieve greater levels of public trust in information. 

A 2024 synthesis paper argues that briefly explaining both human and machine contributions has shown some promise.³⁷ A recent meta-analysis found that people interpret AI-labeled content as slightly less credible but equally readable, accurate and fair.³⁸ Perceived AI authorship was also found to have a bigger impact than actual AI authorship on most measures³⁹ — all the more reason to provide information about the role of human journalists, highlighting the importance of verification.

Research we reviewed suggests

Before communicating about uses of AI, journalism outlets should consider communicating about the methods and value of journalism more broadly. 

Some methods, both human and technological, may best be explained as policies on a stand-alone page rather than repeated with every individual story or segment. For example, some frequent internal AI uses — like transcription and spellcheck — are widely accepted by audiences and may not require constant disclosure.⁴⁰

Rather than using the label “AI,” which many people interpret narrowly to refer only to generative AI, it may be clearer to answer the following three questions: What tool did I use? What did I use it to do? Do I stand by my work? In other words, it’s better to explain briefly how AI was used than to label that AI was used.⁴¹

At least at this point in time, the process of generating or manipulating images and video should be explained at every use, because research shows there is distrust and concern about these uses.⁴² However, what that explanation should contain remains less clear. Providing sources for AI-generated text may help mitigate distrust.⁴³

Journalism organizations should collaborate to build consensus around how to best operationalize shared values such as transparency and verification.⁴⁴

Where more research would be helpful

There has been far more research on transparency about AI use in synthetic text and images than audio — what information would be most appropriate?

How substantive must changes to photos or video be to require explanation? Many types of photo editing have been a common practice for years and are not typically explained. Existing photo manipulation policies, many of which date back to the widespread adoption of software such as Adobe Photoshop, may be helpful in developing guidelines and thresholds for communication.

How do people interact with transparency information in the context of existing news habits and routines?

What is the impact of transparency strategies on people’s understanding of AI? …on people’s understanding of journalism?

What is the relationship between reactions to transparency and (mis)trust in particular outlets and journalists?

Covering AI in Journalism

Journalism serves an important and powerful role in explaining how new technologies work to the public, impacting people’s understanding of and attitudes towards technology.⁴⁵ Recent developments in AI technologies have sparked research interest in coverage of AI, especially (1) when and how coverage has increased over time and (2) what sentiment and frames journalists are using to explain AI.⁴⁶ 

Global Perspectives

Of the topics explored in this briefing, the research about journalism’s coverage of AI is probably the most geographically diverse. Even so, considerations around issues like sourcing are likely to vary considerably by region. For example, journalists in countries with fewer technology companies may have less access to company spokespeople — but better access to outsourced workers. Infrastructure and regulatory contexts also vary in ways that may impact what’s most relevant to a particular audience.

Working group member Zara Schroeder shares examples:

In many African countries, journalists often report on the downstream effects of global tech decisions rather than direct interactions with Big Tech firms. This shapes how sources are selected and what stories get told. 

For instance, Ghanaian journalists have investigated the local impact of global content moderation work outsourced by companies like Facebook.⁴⁷ In 2023, Ghanaian outlets helped amplify stories from Nairobi, Kenya, where whistleblowers exposed poor working conditions and psychological harm among content moderators employed by third-party firms on behalf of Meta.⁴⁸ Without direct access to Silicon Valley executives and spokespeople, reporters in Ghana and across the region instead sourced their stories from affected workers, labor rights advocates and leaked internal communications demonstrating how sourcing adapts when direct access is limited. 

In Nigeria, journalists have covered growing concerns over AI-powered surveillance technologies, such as facial recognition systems used by law enforcement.⁴⁹ With little transparency from tech vendors or government bodies, reporters often rely on investigative methods like freedom of information requests, leaked documents or interviews with civil society watchdogs.⁵⁰ A notable case involved uncovering the quiet deployment of Chinese-made surveillance systems in public infrastructure. Here, sourcing is shaped more by investigative persistence and NGO collaboration than corporate access. 

These examples demonstrate that sourcing practices are nuanced globally, and that concerns about overreliance on sources that are not independent may be less salient in lower-access environments. Instead, sourcing in these environments is often more grassroots, labor-intensive and dependent on alternative networks of information. Understanding these differences is critical for accurately interpreting journalism produced in diverse media ecosystems. 

We looked at two dozen studies which examine (1) the tone of AI coverage, (2) the language journalists use to frame AI (and potential benefits and risks), (3) interviews with journalists and (4) surveys and experiments. These studies reveal important findings but they are rarely conclusive due to gaps in the types of news organizations studied (large vs. small), the geographic contexts included and the unique research methods used. Thus, we need more data and research to form a comprehensive understanding of AI coverage in contexts around the world. 

When covering AI topics — including technology developments and regulations — several researchers find that coverage may not include a full range of sources and perspectives. This can pose challenges for covering AI because journalists may (1) rely too heavily on sources that are have a financial interest in the outcomes,⁵¹ (2) lack the technical expertise needed to verify claims or ask important questions of first-person reports⁵² and thus (3) be overly swayed by the positions or frames offered by their sources. Research also finds that coverage of AI follows one of two overarching narratives:⁵³ (1) AI-in-general, which orients towards expected future implications, treats AI as inevitable and foregrounds economic competition;⁵⁴ and (2) AI-in-particular, which emphasizes specific, current uses of the technology, foregrounds impacts to particular communities and stakeholders, and highlights both continuity and change. That is, stories that focus on concrete technologies tend to take a more measured and balanced approach.

Researchers have also explored whether coverage of AI portrays the technology positively or negatively (or both, depending on topic). These studies examine the sentiment and/or frames used to explain the technology to audiences. Findings to date suggest that the topic of the story matters a great deal as to whether the take is positive or negative.⁵⁵ For example, coverage of AI may lean positive regarding healthcare, economic topics and innovation,⁵⁶ neutral-to-negative in the context of political topics⁵⁷ and negative when discussing topics like data bias and cyber crime.⁵⁸ As AI systems continue to develop, it is important to know both how journalists are explaining these developments and also how the topic of the story impacts how journalists cover it.. 

The differences described above are likely due to unique research methods and settings. Researchers are exploring how AI is covered (1) in different countries (i.e., information environments), (2) by different news organizations, (3) across different time periods (e.g., 2010-2021, 2021-2024, etc.), (4) with different measures for sentiment, topics and framing and (5) using unique keywords (e.g., “artificial intelligence” and “big data” or “artificial intelligence,” “robots/robotics,” “algorithms” and “automation”) to select articles for inclusion in the study. 

Another factor that can shape AI coverage is the ideological make-up of a news organization’s audience or overt ideology of the news organization itself. Indeed, one study found that organizations with either left or right ideological biases (based on ratings from Media Bias/Fact Check,⁵⁹ an independent effort in which fact checkers grade organizations) report on AI risks more than center/low ideological bias organizations.⁶⁰ Yet, it remains to be seen if these findings will be corroborated by further research.

Most studies consistently find that newsroom coverage of AI dramatically increased during the mid-2010s.⁶¹ Much of this research, however, does not include coverage after late 2022, when ChatGPT was released to the public. Recent analyses, though, find steady increases in the number of articles about AI through 2023.⁶² Coverage of AI remains a major area of focus in newsrooms and forthcoming research will shed light on how the amount of coverage has changed over time.

Research to date has explored a range of research questions — including the tone of AI coverage, the frames used to explain AI technologies and differences in AI coverage over time — but research does not necessarily yield clear assessments of the quality or thoroughness of AI coverage (especially outside of English-speaking locations). While there are some valuable learnings from the research to date, it is critical that we (1) have more research and (2) consistently apply that research to understanding AI coverage at a deeper level.

Research we reviewed suggests

Journalism should continue to track not just how much they cover AI but how they are covering it. To do so, newsrooms need to have a deep understanding of the technology itself and the range of stakeholders involved. Journalists can then ask deeper questions, seek input from a range of voices and monitor their own framing more effectively. It is critical that they are intentional and aware of the frames they use because the public learns about technologies from this coverage.⁶³

Coverage of AI should consistently include a comprehensive selection of perspectives, including affected workers, cross-industry experts, non-industry technical experts, members of the general public and critics — in addition to technology company and government sources — to ensure the public has access to a full range of information.⁶⁴

News organizations should prioritize coverage of specific, current uses, which can help audiences form a clearer understanding of what the technologies can and can’t do. However, broad, future-focused reporting that treats the technology as inevitable may lead to hype and unrealistic expectations of both benefits and harms.⁶⁵

As AI technologies continue to develop and improve, journalists and newsrooms should continue to clarify that AI systems (1) are built and maintained by humans and (2) do not have the ability to think, understand, talk, etc. the way humans do.⁶⁶

Where more research would be helpful

We need to continue building knowledge — across a variety of settings — about how specific topics, sources and perspectives relate to how AI is covered. While there has been a lot of research in this area, there are few conclusive findings that can be applied globally. 

Learning how audiences around the world react to different coverage language and frames — likely through experimental designs⁶⁷ — is important for understanding how coverage affects audiences’ beliefs about AI. 

Most research has examined national-level news coverage, whereas regional and local news organizations have received much less attention. How does coverage of AI differ when looking at local and regional news organizations?

There are also gaps when it comes to geographic location. Research to date has focused on the U.S. and Western Europe. Does coverage elsewhere show similar patterns? What is unique to each region of the world?

Similarly, most of the research uses English-language (Anglophone) news coverage. Further expanding research to non-English sources will provide a more comprehensive understanding of AI coverage.

What constitutes ‘comprehensive’ or ‘high-quality’ reporting on technologies and their political and economic contexts?

Current Working Group Members

A list of current working group members and their affiliations is shown here:

• Akintunde Babatunde
  Executive Director, Centre for Journalism Innovation and Development
• Jay Barchas-Lichtenstein
  Senior Research Manager, Center for News, Technology & Innovation
• Madhav Chinnappa
  Independent Media Consultant
• Utsav Gandhi
  Research Intern, Center for News, Technology & Innovation; PhD Student, University of Illinois Chicago
• Samuel Jens
  Research Associate, Center for News, Technology & Innovation
• Amy Mitchell
  Executive Director, Center for News, Technology & Innovation
• Sophie Morosoli
  Postdoctoral Researcher at the AI, Media & Democracy Lab, University of Amsterdam
• Gary Mundy
  Director Research, Policy and Impact, Thomson Foundation
• Oluwapelumi Oginni
  Project Manager, AI Initiatives, Centre for Journalism Innovation and Development
• Joshua Olufemi
  Executive Director, Dataphyte Foundation
• Amy Ross Arguedas
  Research Fellow at the Reuters Institute for the Study of Journalism
• Zara Schroeder
  Researcher, Research ICT Africa
• Felix M. Simon
  Research Fellow in AI and News, Reuters Institute for the Study of Journalism & Research Associate, Oxford Internet Institute, University of Oxford
• Scott Timcke
  Senior Research Associate, Research ICT Africa
• Jaemark Tordecilla
  Independent Media Advisor, Philippines

References

Adefioye, S. (2024). The adoption of artificial intelligence into journalism practice: perspectives from the Ghanaian media industry. MediAsia Official Conference Proceedings, 405–419. https://doi.org/10.22492/issn.2186-5906.2024.33 

Aikulola, S. (2025). Press freedom day: UN, IPC, MRA call for responsible use of AI. https://guardian.ng/features/media/press-freedom-day-un-ipc-mra-call-for-responsible-use-of-ai/

Alayande, A & Olufemi, J. (2023). Generating buzz: How can local newsrooms in Africa use generative AI to stay ahead of the curve? Research ICT Africa. https://researchictafrica.net/2023/06/27/generating-buzz-how-can-local-newsrooms-in-africa-use-generative-ai-to-stay-ahead-of-the-curve/

Allaham, M., Kieslich, K., & Diakopoulos, N. (2025). Global perspectives of AI risks and harms: analyzing the negative impacts of AI technologies as prioritized by news media. Preprint. https://arxiv.org/abs/2501.14040

Altay, S., & Gilardi, F. (2024). People are skeptical of headlines labeled as AI-generated, even if true or human-made, because they assume full AI automation. PNAS Nexus, 3(10), 403.

Ananny, M. (2024). Making generative artificial intelligence a public problem: seeing publics and sociotechnical problem-making in three scenes of AI failure. Javnost – The Public, 31(1), 89–105.

Anderson, A. A., Scheufele, D. A., Brossard, D., & Corley, E. A. (2012). The role of media and deference to scientific authority in cultivating trust in sources of information about emerging technologies. International Journal of Public Opinion Research, 24(2), 225–237.

Associated Press. (2024). The Associated Press stylebook, 2024-2026. Basic Books.

Barchas-Lichtenstein, J., Mitchell, A., Wright, E., LeCompte, C., Jens, S., & Beed, N. (2025). What it means to do journalism in the age of AI: journalist views on safety, technology and government. Center for News, Technology & Innovation. https://cnti.org/2024-journalist-survey/

Bartholomew, J. & Mehta, D. (2023). How the media is covering ChatGPT. Columbia Journalism Review. https://www.cjr.org/tow_center/media-coverage-chatgpt.php

Beckett, C. & Yaseen, M. (2023). Generating change: a global survey of what news organisations are doing with ai. JournalismAI. https://www.journalismai.info/research/2023-generating-change

Berente, N., Gu, B., Recker, J., & Santhanam, R. (2021). Managing artificial intelligence. Management Information Systems Quarterly, 45(3), 1433–1450.

Biagini, G. (2025). Towards an AI-literate future: A systematic literature review exploring education, ethics, and applications. International Journal of Artificial Intelligence in Education.

Bien-Aimé, S., Wu, M., Appelman, A., & Jia, H. (2025). Who wrote it? News readers’ sensemaking of AI/human bylines. Communication Reports, 38(1), 46–58.

Brantner, C., & Saurwein, F. (2021). Covering technology risks and responsibility: automation, artificial intelligence, robotics, and algorithms in the media. International Journal of Communication, 15(0), Article 0.

Brause, S. R., Zeng, J., Schäfer, M. S., & Katzenbach, C. (2023). Chapter 24: Media representations of artificial intelligence: surveying the field. In S. Lindgren (Ed.), Handbook of Critical Studies of Artificial Intelligence. Edward Elgar Publishing. 

Brennen, J. S., Howard, P. N., & Nielsen, R. K. (2022). What to expect when you’re expecting robots: futures, expectations, and pseudo-artificial general intelligence in U.K. news. Journalism, 23(1), 22–38.

Bunz, M., & Braghieri, M. (2022). The AI doctor will see you now: Assessing the framing of AI in news coverage. AI & SOCIETY, 37(1), 9–22.

The Cable. (2023a). How digital surveillance threatens press freedom in Nigeria, West African countries. https://www.thecable.ng/how-digital-surveillance-threatens-press-freedom-in-west-africa/

The Cable. (2023b). Heightened surveillance by security operatives puts Nigerian journalists under climate of fear. https://www.thecable.ng/special-report-heightened-surveillance-by-security-operatives-puts-nigerian-journalists-under-climate-of-fear/

Canavilhas, J., & Essenfelder, R. (2022). Apocalypse or redemption: How the Portuguese media cover artificial intelligence. In J. Vázquez-Herrero, A. Silva-Rodríguez, M.-C. Negreira-Rey, C. Toural-Bran, & X. López-García (Eds.). Total journalism: Models, techniques and challenges (pp. 255–270). Springer International Publishing.

Cheong, I., Guo, A., Lee, M., Liao, Z., Kadoma, K., Go, D., Chang, J. C., Henderson, P., Naaman, M., & Zhang, A. X. (2025). Penalizing transparency? How AI disclosure and author demographics shape human and AI judgments about writing (No. arXiv:2507.01418). arXiv. https://doi.org/10.48550/arXiv.2507.01418

Choi, S. (2024). Temporal framing in balanced news coverage of artificial intelligence and public attitudes. Mass Communication and Society, 27(2), 384–405.

Cools, H., B. Van Gorp, and M. Opgenhaffen. (2022). Where exactly between utopia and dystopia? A framing analysis of AI and automation in U.S. newspapers. Journalism, 25(1), 3–21.

Cools, H., de Vreese, C., El Ali, A., Helberger, N., Prajod, P., Mattis, N., Morosoli, S., Naudts, L., & Weikmann, T. (2025, April 22). Tackling the transparency puzzle: Five perspectives from AI disclosure research in news. Medium. https://generative-ai-newsroom.com/tackling-the-transparency-puzzle-0969b3bcc489

Egwu, P. & Saint, E. (2024). From debunking disinformation to turning datasets into stories, AI is changing newsrooms in Nigeria. International Journalists Network. https://ijnet.org/en/story/debunking-disinformation-turning-datasets-stories-ai-changing-newsrooms-nigeria

Epstein, Z., Fang, M., Arechar, A., & Rand, D. (2023). What label should be applied to content produced by generative AI? Preprint. https://osf.io/v4mfz

Fletcher, R. & Kleis Nielsen, R. (2024). What does the public in six countries think of generative AI in news? Reuters Institute. https://reutersinstitute.politics.ox.ac.uk/what-does-public-six-countries-think-generative-ai-news

Frau-Meigs, D. (2024). Algorithm Literacy as a subset of media and information literacy: Competences and design considerations. Digital, 4(2), 512–28.

Gagrčin, E., Naab, T. K., and Grub, M.F. (2024). Algorithmic media use and algorithm literacy: An integrative literature review. New Media & Society. 

Gilardi, F., Lorenzo, S. D., Ezzaini, J., Santa, B., Streiff, B., Zurfluh, E., & Hoes, E. (2025). Willingness to read AI-generated news is not driven by their perceived quality. Preprint. https://arxiv.org/abs/2409.03500

Gondwe, G. (2025). Perceptions of AI-driven news among contemporary audiences: A study of trust, engagement, and impact. AI & SOCIETY. doi.org/10.1007/s00146-025-02294-x

Hall, R. & Wilmot, C. (2025a). Meta faces Ghana lawsuits over impact of extreme content on moderators. The Guardian. https://www.theguardian.com/technology/2025/apr/27/meta-faces-ghana-lawsuits-over-impact-of-extreme-content-on-moderators

Hall, R. & Wilmot. C. (2025b). ‘I didn’t eat or sleep’: a Meta moderator on his breakdown after seeing beheadings and child abuse. The Guardian. https://www.theguardian.com/technology/2025/apr/27/meta-moderator-on-the-cost-of-viewing-beheadings-child-abuse-and-suicide

Hokkanen, J. (2025). AI adoption in South African newsrooms: Exploring journalists’ perceptions. Journalism Research News. https://journalismresearchnews.org/article-ai-adoption-in-south-african-newsrooms-exploring-journalists-perceptions/

Höppner, S. (2025). Africa’s content moderators want compensation for job trauma. DW. https://www.dw.com/en/africas-content-moderators-want-compensation-for-job-trauma/a-72401025 

Institute of Development Studies. (2023). Nigeria spending billions of dollars on harmful surveillance of citizens. https://www.ids.ac.uk/press-releases/nigeria-spending-billions-of-dollars-on-harmful-surveillance-of-citizens/

Ishengoma, D. J., & Magolanga, E. M. (2025). Adaptations of artificial intelligence (AI) in Tanzanian newsrooms: Opportunities, benefits and threats to journalism professionals. Journal of Applied Journalism; Media Studies. https://doi.org/10.1386/ajms_00178_1 

Jamil, S. (2020). Artificial intelligence and journalistic practice: The crossroads of obstacles and opportunities for the Pakistani journalists. Journalism Practice, 15(10), 1400–1422. https://doi.org/10.1080/17512786.2020.1788412 

Ji, X., Kuai, J., & Zamith, R. (2024). Scrutinizing algorithms: Assessing journalistic role performance in Chinese news media’s coverage of artificial intelligence. Journalism Practice, 18(9), 2396–2413.

Jia, H., Appelman, A., Wu, M., & Bien-Aimé, S. (2024). News bylines and perceived AI authorship: Effects on source and message credibility. Computers in Human Behavior: Artificial Humans, 2(2), 100093.

Kabir, A. & Adebajo, A. (2023). How digital surveillance threatens press freedom in West Africa. HumAngle. https://humanglemedia.com/how-digital-surveillance-threatens-press-freedom-in-west-africa/

Korneeva, E., Salge, T. O., Teubner, T., & Antons, D. (2023). Tracing the legitimacy of artificial intelligence: A longitudinal analysis of media discourse. Technological Forecasting and Social Change, 192, 122467.

Köstler, L., & Ossewaarde, R. (2022). The making of AI society: AI futures frames in German political and media discourses. AI & Society, 37(1), 249–263.

Kuai, J. (2025). Navigating the AI hype: Chinese journalists’ algorithmic imaginaries and role perceptions in reporting emerging technologies. Digital Journalism, 1–20.

Kumar, A., & Sangwan, S. R. (2024). Conceptualizing AI literacy: Educational and policy initiatives for a future-ready society. International Journal of All Research Education and Scientific Methods, 12(4).

Lammar, D., Horst , M., & Müller, R. (2025). AI in the German media: Narratives of AI-in-Particular and AI-in-General in German media reporting about artificial intelligence. Digital Journalism, 1–19.

LeCompte, C., Mitchell, A., & Jens, S. (2024). Focus group insights #2: perceptions of artificial intelligence use in news and journalism. Center for News, Technology & Innovation. https://innovating.news/article/focus-group-insights-2-perceptions-of-artificial-intelligence-use-in-news-and-journalism/

Lush, D. (2022). How African newsrooms are using AI to analyse data and produce good journalism. IMS. https://www.mediasupport.org/blogpost/how-african-newsrooms-are-using-ai-to-analyse-data-and-produce-good-journalism/

Lyu, T., Guo, Y., & Chen, H. (2024). Understanding the privacy protection disengagement behaviour of contactless digital service users: The roles of privacy fatigue and privacy literacy. Behaviour & Information Technology, 43(10), 2007–2023. 

Magalhães, J. C., & Smit, R. (2025). Less hype, more drama: open-ended technological inevitability in journalistic discourses about AI in the US, the Netherlands, and Brazil. Digital Journalism, 1–18. https://doi.org/10.1080/21670811.2025.2522281

Malik, N. (2025). The decline of print advertising and its impact on journalism. International Journal of Advance Research and Innovative Ideas in Education, 11(1). https://ijariie.com/AdminUploadPdf/The_Decline_of_Print_Advertising_and_its_Impact_on_Journalism_ijariie25794.pdf 

Mari, W. (2024). The pre-history of news-industry discourse around artificial intelligence. Emerging Media, 2(3), 499–522.

Mattis, N. M., Kieslich, K., & de Vreese, C. (2025). Feeling iffy about generative AI: When journalists disclose AI use, trust in news is lower. Preprint. https://osf.io/preprints/osf/tmzq4

Media Bias/Fact Check. (2025). Methodology. Media Bias/Fact Check. https://mediabiasfactcheck.com/methodology/ 

Mills, K., Ruiz, P., Lee, K., Coenraad, M., Fusco, J., Roschelle, J., & Weisgrau, J. (2024). AI literacy: a framework to understand, evaluate, and use emerging technology. Digital Promise.

Misri, A., Blanchett, N., & Lindgren, A. (n.d.). “there’s a rule book in my head”: Journalism ethics meet AI in the newsroom. Digital Journalism, 1–19.

Mitchell, A., Jens, S., Moon Sehat, C., LeCompte, C., Beed, N., Wright, E. and Barchas-Lichtenstein, J. (2025). What the public wants from journalism in the age of AI: a four country survey. Center for News, Technology & Innovation. https://innovating.news/2024-public-survey/

Mohammed, A., Elega, A. A., Ahmad, M. B., & Oloyede, F. (2024). Friends or foes? Exploring the framing of artificial intelligence innovations in Africa-focused journalism. Journalism and Media, 5(4), Article 4.

Montag, C., Nakov, P., & Ali, R. (2024). On the need to develop nuanced measures assessing attitudes towards AI and AI literacy in representative large-scale samples. AI & Society 40, 1129-1130.

Moran, R. E., & Shaikh, S. J. (2022). Robots in the news and newsrooms: unpacking meta-journalistic discourse on the use of artificial intelligence in journalism. Digital Journalism, 10(10), 1756–1774.

Moriniello, F., Martí-Testón, A., Muñoz, A., Silva Jasaui, D., Gracia, L., & Solanes, J. E. (2024). Exploring the relationship between the coverage of AI in WIRED Magazine and public opinion using sentiment analysis. Applied Sciences, 14(5), Article 5.

Morosoli, S., Naudts, L., Cools, H., Venkatraj, K. P., Helberger, N., & Vreese, C. de. (n.d.). Individuals’ need for rights and their sense of accountability connected to AI disclosures. Qualitative evidence from group interviews. Preprint. http://osf.io/768zc

Mugadzaweta, M.S. (2025). The adoption of AI in Zimbabwe’s newsrooms: A case of Zimpapers and Alpha Media Holdings. (Unpublished master’s thesis). Aga Khan University. https://ecommons.aku.edu/cgi/viewcontent.cgi?article=1016&context=etd_ke_gsmc_ma-digjour

Mukasa, R. (2024). Examining the role of artificial intelligence (AI) in transforming print journalism in Uganda. (Unpublished master’s dissertation). Aga Khan University, East Africa. Retrieved from https://ecommons.aku.edu/theses_dissertations/2322/ 

Munoriyarwa, A., Chiumbu, S., & Motsaathebe, G. (2021). Artificial intelligence practices in everyday news production: the case of South Africa’s mainstream newsrooms. Journalism Practice, 17(7), 1374–1392. https://doi.org/10.1080/17512786.2021.1984976 

Ncube, L., Mofokeng, R. W., Chibuwe, A., Munoriyarwa, A., & Murangi, A. K. (2025). ‘Mind the gap’: artificial intelligence and journalism training in Southern African journalism schools. Media Practice and Education, 0(0), 1–17. https://doi.org/10.1080/25741136.2025.2464483

Nguyen, D. (2023). How news media frame data risks in their coverage of big data and AI. Internet Policy Review, 12(2).

Nguyen, D., & Hekman, E. (2024). The news framing of artificial intelligence: A critical exploration of how media discourses make sense of automation. AI & SOCIETY, 39(2), 437–451.

Oeldorf-Hirsch, A., & Neubaum, G. (2025). What do we know about algorithmic literacy? The status quo and a research agenda for a growing field. New Media & Society, 27(2), 681-701.

Olanipekun, S. O., & Olakoyenikan, O. (2022). Ethical implications of generative AI in journalism: Balancing innovation, truth, and public communication trust. World Journal of Advanced Research and Reviews, 16(3), 1293–1311. https://doi.org/10.30574/wjarr.2022.16.3.1159

Olawuyi, E. A., & Enuwah, J. (2025). Framing the future: Media narratives on artificial intelligence and its societal impact. International Journal of Current Research in the Humanities, 28(1), 269–290. https://doi.org/10.4314/ijcrh.v28i1.19 

Parratt-Fernández, S., Chaparro-Domínguez, M.-Á., & Martín-Sánchez, I.-M. (2024). Spanish media coverage of journalistic artificial intelligence: Relevance, topics and framing. Revista Mediterránea de Comunicación, e25169–e25169.

Piasecki, S., Morosoli, S., Helberger, N., & Naudts, L. (2024). AI-generated journalism: Do the transparency provisions in the AI Act give news readers what they hope for? Internet Policy Review, 13(4).

Pinski, M., & Benlian, A. (2024). AI literacy for users – A comprehensive review and future research directions of learning methods, components, and effects. Computers in Human Behavior: Artificial Humans, 2(1), 100062.

Radcliffe, D. (2025). Journalism in the AI era: Opportunities and challenges in the Global South and emerging economies. Thomson Reuters Foundation.

Ross Arguedas, A. (2024). OK computer? Understanding public attitudes towards the uses of generative AI in news. Reuters Institute. https://reutersinstitute.politics.ox.ac.uk/news/ok-computer-understanding-public-attitudes-towards-uses-generative-ai-news

Ruiz, P., & Glazer, K. (2024). Anthropomorphism of AI in learning environments: Risks of humanizing the machine. EdSurge. https://www.edsurge.com/news/2024-01-15-anthropomorphism-of-ai-in-learning-environments-risks-of-humanizing-the-machine

Sánchez-García, P., Diez-Gracia, A., Mayorga, I. R., & Jerónimo, P. (2025). Media self-regulation in the use of AI: limitation of multimodal generative content and ethical commitments to transparency and verification. Journalism and Media, 6(1), Article 1.

Schell, K. (2024). AI transparency in journalism: Labels for a hybrid era. Reuters Institute for the Study of Journalism. https://reutersinstitute.politics.ox.ac.uk/sites/default/files/2025-01/RISJ%20Fellows%20Paper_Katja%20Schell_MT24_Final.pdf

Schüller, K. (2022). Data and AI literacy for everyone. Statistical Journal of the IAOS, 38(2), 477–490.

Sofiullahi, A. (2024). How journalism groups in Africa are building AI tools to aid investigations and fact-checking. Global Investigative Journalism Network. https://gijn.org/stories/africa-journalism-building-ai-investigations-fact-checking/

Solomons, S., & Ndlovu, M. W. (2024). AI adoption in South African newsrooms: exploring journalists’ perceptions. Communication, 50(2), 122–143. https://doi.org/10.1080/02500167.2024.2439971 

Tadimalla, S. Y., & Maher, M. L. (2024). AI literacy for all: adjustable interdisciplinary socio-technical curriculum. 2024 IEEE Frontiers in Education Conference (FIE), 1–9. https://doi.org/10.1109/FIE61694.2024.10893159.

Takahashi, B., & Tandoc Jr., E. C. (2016). Media sources, credibility, and perceptions of science: learning about how people learn about science. Public Understanding of Science, 25(6), 674–690.

Thomson, T. J., Thomas, R. J., & Matich, P. (n.d.). Generative visual AI in news organizations: challenges, opportunities, perceptions, and policies. Digital Journalism, 0(0), 1–22. https://doi.org/10.1080/21670811.2024.2331769

Toff, B., & Simon, F. M. (2024). “Or they could just not use it?”: The dilemma of ai disclosure for audience trust in news. The International Journal of Press/Politics.

Umeora, C. C. (2025). Artificial intelligence and journalistic practices in Nigeria: navigating awareness, adoption, and structural challenges. Multidisciplinary Research and Development Journals Int’l, 7(1), 136–152. https://mdrdji.org/index.php/mdj/article/view/125

Valderrama Barragán, M., Tironi, M., Cotoras, D., Correa, T., Humeres, M., & López, C. (2025). From industry hype to emerging criticism: analysing Chilean news media coverage of artificial intelligence. Digital Journalism, 1–23.

van der Schyff, K., Foster, G., Renaud, K., & Flowerday, S. (2023). Online Privacy Fatigue: A scoping review and research agenda. Future Internet, 15(5), Article 5. https://doi.org/10.3390/fi15050164

Wang, C., Sturgis, P., & Kadt, D. de. (2025). AI labeling reduces the perceived accuracy of online content but has limited broader effects (No. arXiv:2506.16202). arXiv. https://doi.org/10.48550/arXiv.2506.16202

Wang, S., & Huang, G. (2024). The impact of machine authorship on news audience perceptions: a meta-analysis of experimental studies. Communication Research, 51(7), 815–842.

Wittenberg, C., Epstein, Z., Berinsky, A. J., & Rand, D. G. (2024). Labeling AI-generated content: promises, perils, and future directions. an MIT exploration of generative AI. Preprint. https://mit-genai.pubpub.org/pub/hu71se89/release/1

Yang, S., Krause, N. M., Bao, L., Calice, M. N., Newman, T. P., Scheufele, D. A., Xenos, M. A., & Brossard, D. (2023). In AI we trust: the interplay of media use, political ideology, and trust in shaping emerging ai attitudes. Journalism & Mass Communication Quarterly, 102(2), 382–406.

Yeste-Piquer, E., Suau-Martínez, J., Sintes-Olivella, M., & Xicoy-Comas, E. (2025). What if I prefer robot journalists? Trust and objectivity in the AI news ecosystem. Journalism and Media, 6(2), Article 2.

Zhu, H., & Zhang, M. (2024). “I never read it, but I always accept it”: unravelling social, individual, and policy design-induced influences on privacy policy acceptance. ICIS 2024 Proceedings. https://aisel.aisnet.org/icis2024/security/security/5

Appendix

The studies and their research methods for the “Communicating about AI use in journalism” section are presented below. 

The studies for the “Covering AI in journalism” section, including time frames and data sources, are provided in the table below.

Footnotes

1. Mari, 2024 ︎
2. Biagini, 2025; Frau-Meigs, 2024; Gagrčin et al., 2024; Kumar & Sangwan, 2024; Mills et al., 2024; Oeldorf-Hirsch & Neubaum, 2025; Pinski & Benlian, 2024; Schüller, 2022; Tadimalla & Maher, 2024 ︎
3. Biagini, 2025; Gagrčin et al., 2024; Mills et al., 2024; Pinski & Benlian, 2024 ︎
4. Anderson et al., 2012; Takahashi & Tandoc, 2016; Yang et al., 2023 ︎
5. Jamil, 2021; Malik, 2025; Munoriyarwa et al., 2023 ︎
6. Adefioye, 2024; Ishengoma & Magolanga, 2025; Mukasa, 2024; Umeora, 2025  ︎
7. Mills et al., 2024 ︎
8. Pinski & Benlian, 2024; see also Berente et al., 2021 ︎
9. Gagrčin et al., 2024 ︎
10. Associated Press, 2024; Mills et al., 2024 ︎
11. Ruiz & Glazer, 2004 ︎
12. Biagini, 2025 ︎
13. Pinski & Benlian, 2024 ︎
14. Montag et al., 2024 ︎
15. Beckett & Yaseen, 2023; Fletcher & Kleis Nielsen, 2024; Gondwe, 2025; LeCompte et al., 2024; Piasecki et al., 2024; Ross Arguedas, 2024; Toff & Simon, 2024 ︎
16. Barchas-Lichtenstein et al., 2025; Radcliffe, 2025 ︎
17.  Radcliffe, 2025 ︎
18. Ncube et al., 2025 ︎
19. Radcliffe, 2025 ︎
20. Munoriyarwa, 2024 ︎
21. Olanipekun & Olakoyenikan, 2022 ︎
22. Alayande & Olufemi, 2023; Sofiullahi, 2024 ︎
23. Mugadzaweta, 2025; Lush, 2022; Mukasa, 2024; Hokkanen, 2025 ︎
24. Egwu & Saint, 2024 ︎
25. Egwu & Saint, 2024 ︎
26. Egwu & Saint, 2024 ︎
27. Egwu & Saint, 2024 ︎
28. Altay & Gilardi, 2024 ︎
29. Toff & Simon, 2024 ︎
30. Wang et al., 2025 ︎
31. Cheong et al., 2025 ︎
32. Altay & Gilardi, 2024; Fletcher & Kleis Nielson, 2024; Ross Arguedas, 2024 ︎
33. Bien-Aimé et al., 2025; Jia et al., 2024 ︎
34. Mattis et al., 2025; Toff & Simon, 2024 ︎
35. Lyu et al., 2024; van der Schyff et al., 2023; Zhu & Zhang, 2024 ︎
36. Barchas-Lichtenstein et al., 2025 ︎
37. Schell, 2024 ︎
38. Wang & Huang, 2024 ︎
39. Gilardi et al., 2025; Jia et al., 2024 ︎
40. Fletcher & Kleis Nielsen, 2024; Mitchell et al., 2025 ︎
41. Bien-Aimé et al., 2025; Jia et al., 2024; Wang & Huang, 2024 ︎
42. Fletcher & Kleis Nielsen, 2024; Mitchell et al., 2025; Thomson et al., 2024 ︎
43. Toff & Simon, 2024 ︎
44. Misri et al., n.d.; Morosoli et al., n.d.; Sánchez García et al., 2025 ︎
45. Anderson et al., 2012; Takahashi & Tandoc, 2016; Yang et al., 2023 ︎
46. Annany, 2024; Brause et al., 2023; Choi, 2024; Moran & Shaikh, 2022; Parratt-Fernández et al., 2024 ︎
47. Hall & Wilmot, 2025 ︎
48. Hall & Wilmot, 2025b; Höppner, 2025 ︎
49. Aikulola, 2025; Institute of Development Studies, 2023 ︎
50. The Cable, 2023a; 2023b; Kabir & Adebajo, 2023 ︎
51. Bunz & Braghieri, 2022; Canavilhas & Essenfelder, 2022 ︎
52. Brennen et al. 2022; Ji et al., 2024; Kuai, 2025 ︎
53. Lammar et al., 2025 ︎
54. See also Magalhães & Smit, 2025 ︎
55. Brantner & Saurwein, 2021; Canavilhas & Essenfelder, 2022 ︎
56. Mohammed et al., 2024; Moriniello et al., 2024 ︎
57. Canavilhas & Essenfelder, 2022 ︎
58. Nguyen & Hekman, 2024 ︎
59. Media Bias/Fact Check, 2025 ︎
60. Allaham et al., 2025 ︎
61. Brantner & Saurwein, 2021; Cools et al., 2022; Korneeva et al., 2023; Nguyen, 2023; Nguyen & Hekman, 2024; Valderrama et al., 2025 ︎
62. Bartholomew & Mehta, 2023; Valderrama et al., 2025 ︎
63. Allaham et al., 2025; Brantner & Saurwein, 2001; Choi, 2024; Korneeva et al., 2023; Nguyen, 2023; Nguyen & Hekman, 2024; Parratt-Fernández et al., 2024; Radcliffe, 2025 ︎
64. Ananny, 2024; Brennen et al., 2022; Canavilhas & Essenfelder, 2022; Mohammed et al., 2024 ︎
65. Lammar et al., 2025; Magalhães & Smit, 2025 ︎
66. Associated Press, 2024 ︎
67. Epstein et al., 2023 ︎

The post AI Literacy & Communication appeared first on Center for News, Technology & Innovation.

Resumen

Hoy la cantidad de medios, canales y voces a través de los cuales el público accede a las noticias no deja de crecer, lo que hace que tengan más formas de enterarse de los asuntos y los sucesos que les importan como nunca antes. Como pudo comprobar el CNTI en una serie de grupos focales que se reunieron antes de esta encuesta, las personas dedican un gran esfuerzo a ponerse al día con esas noticias. Al mismo tiempo, muchas están eligiendo desconectarse de las noticias y manifiestan sentir una sensación de agobio a causa de ellas. El advenimiento de la inteligencia artificial (IA) solo ha venido a sumarse a la gran cantidad de maneras en las que se comparte y consume información.

¿Cómo navega el público por estas nuevas formas de informarse? ¿Qué relevancia le asigna al periodismo? ¿Y cómo puede el periodismo comunicar mejor su valor único? El CNTI profundizó en estas preguntas (y en algunas otras) en esta encuesta.

El CNTI encuestó a 4037 personas de cuatro países (Australia, Brasil, Sudáfrica y Estados Unidos) entre el 4 de septiembre y el 16 de octubre de 2024 para comprender sus perspectivas sobre quién puede producir periodismo, cómo acceden a las fuentes de información en las que confían y sus actitudes sobre los usos periodísticos de tecnologías como la IA. Al igual que en todas las investigaciones del CNTI, este informe fue elaborado por el personal investigador y profesional del CNTI. 

Principals Hallazgos

En lo que respecta a satisfacer las necesidades de información del público, las organizaciones de noticias desempeñan un papel clave, junto con otros proveedores valiosos, incluida la gente común. A pesar de que la mayoría de las personas dicen que las organizaciones de noticias desempeñan un papel importante en la sociedad y aún acuden con mayor frecuencia a ellas para informarse, una minoría considerable (20-30 %) considera a las personas como su principal fuente de noticias. Además, aproximadamente la mitad o más de las personas encuestadas en cada país dicen que creen que los periodistas no son los únicos individuos capaces de hacer periodismo y, de hecho, que la gente común puede hacerlo. (Lea esta sección del informe.)

La mayoría de las personas se sienten generalmente positivas sobre su capacidad para mantenerse informadas y consideran que las tecnologías digitales desempeñan un papel importante, pero la confianza es el mayor desafío: en tres de los cuatro países, la gran mayoría de las personas opinan que las tecnologías digitales son muy importantes para mantenerse informadas (67-85 %). Además, la mayoría de las personas tienen una actitud positiva sobre su capacidad para mantenerse informadas (59-67 %). Estados Unidos fue el caso atípico en ambas preguntas: una pequeña mayoría de los encuestados (57 %) dice que las tecnologías digitales son muy importantes para mantenerse informados y, aunque ninguna opinión recogió la mayoría de las respuestas sobre su capacidad para mantenerse informados, la mayoría simple tiene una opinión neutral (aunque mucho más positiva que negativa). A pesar de esta positividad, al menos el 70 % de las personas en cada país dicen que “saber en quién o en qué confiar” es al menos en cierta medida un desafío. (Lea esta sección del informe.)

En general, a las personas no les incomoda que los periodistas utilicen la tecnología con fines profesionales, pero las opiniones sobre la edición de imágenes y la IA son más variadas en Estados Unidos y Australia: una amplia mayoría (71-93 %) en los cuatro países considera que es gran medida aceptable usar la tecnología para verificar si algo es cierto, traducir contenido y resumir la información de múltiples documentos. Sin embargo, surgen diferencias en lo que respecta a la edición de imágenes. Los brasileños y los sudafricanos se sienten cómodos casi en igual medida con que los periodistas utilicen la tecnología de edición de imágenes (71-75 %), mientras que los estadounidenses y los australianos lo están mucho menos (35-49 %). Del mismo modo, los brasileños y los sudafricanos tienen una actitud más positiva que negativa sobre el impacto de la IA en el periodismo: el 46 % o más dice que tendrá un efecto mayormente positivo en la capacidad de los periodistas para informar sobre asuntos y sucesos. Por otro lado, los australianos y los estadounidenses son más negativos que positivos: entre el 28 % y el 41 % de los encuestados dicen que tendrá un efecto mayormente negativo, mientras que entre el 15 % y el 18 % dicen que el efecto será positivo. (Lea esta sección del informe.)

Más del 60 % en cada país confía en general en que Internet seguirá siendo un lugar para obtener y compartir noticias. Fuera de Estados Unidos, el 58 % o más de las personas en cada país también son optimistas sobre el impacto que los desarrollos en tecnología digital tendrá en su capacidad para mantenerse informadas. (Estados Unidos no se queda atrás, con el 46 %). La mayorías simples en Brasil y Sudáfrica también son positivas sobre el impacto de la IA en su capacidad para mantenerse informadas, mientras que en Estados Unidos y Australia son neutrales. En cada país, un número considerablemente mayor de personas tienen una actitud positiva con respecto a la tecnología en general en comparación con la IA en particular. (Lea esta sección del informe.)

Esta perspectiva, junto con la actitud abierta del público con respecto a que los periodistas utilicen diversas formas de tecnología, genera una oportunidad para quienes hacen periodismo: aprovechar las tecnologías disponibles de manera tal que conecten con el público y satisfagan sus necesidades de información, al tiempo que les garantiza su aceptación del producto final. Esto, a su vez, puede aumentar las propias capacidades de los periodistas en materia de cobertura de información, seguridad y, en última instancia, sostenibilidad.

Lea el informe complementario aquí.

The post Qué quiere el público del periodismo en la era de la IA: una encuesta en cuatro países appeared first on Center for News, Technology & Innovation.

View the report in English

En las últimas décadas, la adopción y el uso generalizados de las nuevas tecnologías de la comunicación han transformado rápidamente los ecosistemas mundiales de la información.

Para los periodistas y las organizaciones de noticias, estos cambios han potenciado su trabajo y, a su vez, han planteado nuevos desafíos. Para el público, estos cambios tecnológicos y sociales han aportado muchas más opciones, alterando la forma en que los medios de comunicación han interactuado desde siempre con la gente.

Al mismo tiempo, los Gobiernos de todo el mundo están vulnerando cada vez más la libertad de prensa y convirtiendo la ley en un arma contra los periodistas, mientras que los asuntos sobre los modelos de ingresos y la valoración del contenido digital siguen sin resolverse.

Las plataformas de redes sociales, en concreto, han ofrecido nuevas oportunidades para encontrar al público justo donde está, pero también han convertido a los periodistas en blanco de constantes amenazas legales y acoso. 

Últimamente, la conveniencia y la idoneidad de las herramientas de IA¹ pueden ayudar a personas y equipos a producir más contenido, pero estas herramientas requieren muchos recursos, han mostrado potencial de inexactitud y la opacidad de sus algoritmos deja a los periodistas con la incertidumbre acerca de qué nuevos propósitos se le está dando a su trabajo. 

Las encuestas son una instantánea de lo que la gente piensa en un momento concreto. Entre el 14 de octubre de 2024 y el 24 de noviembre de 2024, encuestamos a más de 430 periodistas en más de 60 países sobre el Gobierno, la tecnología, el ciberacoso y lo que implica ser periodista en estos días.

Al igual que todas las investigaciones del CNTI, este informe fue elaborado por nuestro equipo de investigadores y profesionales. Aquí se presentan algunos aspectos destacados de lo que descubrimos:

Los periodistas ven mucho valor en su sector, pero no están seguros de que el valor se esté comunicando bien, lo que genera confusión pública: los periodistas estadounidenses no creen que el público pueda diferenciar qué es y qué no es periodismo. Aproximadamente uno de cada cuatro periodistas estadounidenses (24 %) cree que el público puede distinguir el periodismo de otros tipos de noticias e información. Por su parte, cerca de la mitad de los periodistas mexicanos (48 %) creen que el público puede hacer esa distinción, al igual que el 70 % de los periodistas nigerianos. Y, si bien la formación profesional y las instituciones son importantes para la autopercepción de los periodistas, la mayoría de ellos está de acuerdo en que quienes no son periodistas pueden hacer periodismo. 

La mitad de los periodistas encuestados (50 %) han sufrido una extralimitación directa del Gobierno en el último año. Esta puede ser la razón por la cual la gran mayoría (más de tres cuartas partes) manifiesta que no corresponde que el Gobierno defina el periodismo o a los periodistas, y casi la mitad afirma que el Gobierno ejerce demasiado control sobre el periodismo.

Los periodistas creen que la tecnología mejora su trabajo, pero dudan de la IA: dos tercios afirman que la tecnología en general, y las redes sociales en particular, están teniendo un efecto positivo en su trabajo, aunque solo un tercio manifiesta lo mismo sobre el efecto de la IA en el panorama de la información. Los periodistas en el Sur Global son más positivos en general.

Hay riesgos graves que están muy extendidos: uno de cada tres los enfrenta con relativa frecuencia o más. De todos modos, los niveles de precaución varían: los periodistas cambian contraseñas y actualizan hardware y software en los dispositivos con frecuencia, pero no siempre se comunican con las fuentes a través de las plataformas más seguras. Alrededor del 40 % afirmó que hace ambas cosas una vez cada pocos meses en promedio, y el 30 % o menos dijo hacerlo una vez cada pocos años o cuando el dispositivo deja de funcionar. Además, los periodistas en el Norte Global hacen ambas cosas con más frecuencia que sus colegas de otros lugares. Mientras tanto, el 15 % de los periodistas afirman que utilizan la mensajería cifrada entre pares como su principal medio de comunicación con las fuentes. La gran mayoría no tiene problema en hablar sobre seguridad con colegas y gerentes, por ejemplo, sobre la censura por parte del Gobierno y experiencias personales de abuso.

Por último, para conectar las diferentes áreas temáticas, planteamos una pregunta general sobre siete problemas que enfrentan muchas organizaciones de noticias en la actualidad. Los resultados aportan información valiosa sobre la percepción del sector de los periodistas, y las prioridades de la sala de redacción, que pueden no coincidir con la perspectiva de los jefes de redacción. Según los encuestados, desde hace tiempo el compromiso del público continúa recibiendo la mayor atención dentro de las organizaciones de noticias, seguido por los flujos de ingresos y la información errónea. El tema al que se le presta menos atención es el abuso en línea.

En este informe se aborda el entorno informativo con énfasis en cuatro de las áreas temáticas más apremiantes de la actualidad, cada una de las cuales merece que se investigue y converse más al respecto: definiciones de noticia y periodismo, las relaciones entre las organizaciones de noticias y el Gobierno, la tecnología y la IA, y la seguridad y la protección. Si bien el informe desglosa estas áreas, hay una gran cantidad de información interconectada que se aborda en las distintas secciones.

Footnotes

1. Dada la falta de consenso sobre lo que abarca la “inteligencia artificial”, usamos el término en sentido amplio para referirnos a “ciencias, teorías y técnicas cuyo propósito es reproducir mediante una máquina las facultades cognitivas de un ser humano”. Si bien no hay una definición técnica comúnmente aceptada, es útil considerar ejemplos como los modelos de lenguaje de gran escala (LLM), que se “entrenan” con datos para reconocer tendencias estadísticas con las que generar texto plausible. Este tipo de modelos suelen tener demasiados parámetros para ser totalmente transparentes o para que sean explicados, incluso por sus creadores. ︎

The post Qué significa hacer periodismo en la era de la IA: Opiniones de los periodistas sobre la seguridad, la tecnología y el Gobierno appeared first on Center for News, Technology & Innovation.

This report is the first in a series of deep dives into the data from the Center for News, Technology & Innovation (CNTI)’s global surveys of journalists and the public.

In conjunction with our partners at the Centre for Journalism Innovation and Development (CJID), CNTI explored what the surveys reveal about artificial intelligence (AI)’s impact on journalism in Africa. We focused on two central questions:

1. How do journalists in Africa feel about the relationship between journalism and technology, including AI?
2. How does the South African public feel about the relationship between journalism and technology, including AI?

The emergence, continuous evolution and daily use of AI presents both opportunities and challenges for the news industry. As newsrooms adapt to new technological developments and habits, they must also stay true to journalism’s mission and fundamentals. While the rise of AI-powered tools is revolutionizing storytelling, fact-checking, security and audience engagement, the tools’ associated ethical dilemmas, increased accessibility of spyware and government surveillance that use AI and rampant misinformation are persistent issues. 

To answer our two questions, we pull from two separate studies which reflect distinct pools of people. Our global survey of journalists included more than 100 journalists across Africa. Our public survey included a representative sample of the South African public. In both studies, we asked about attitudes towards and uses of technology as a whole — including several questions specifically focused on AI. As with all CNTI research, this report was prepared by the research and professional staff of CNTI.

Part I: What Journalists in Africa Think about Technology

Between October 14, 2024 and December 1, 2024, CNTI surveyed 433 journalists from 63 different countries. This report focuses on respondents based in Africa, including 75 in Nigeria and 34 in other African countries (N = 109). 

This data finds journalists in Africa to be more optimistic than journalists in the rest of the world about the impacts of technological developments — including AI and social media — on enabling journalists’ work and an informed public. They are also more likely to have incorporated technological tools into their work to do things like summarize or analyze documents and enhance their writing. Compared to journalists in other regions, journalists in Africa are more likely to report that their news organizations are transparent with audiences about the use of technology, especially when it is used to check accuracy or improve their writing. Still, similar to journalists around the world, they believe news organizations should pay as much attention as they currently do — or more — to the opportunities and risks AI presents to the news industry.

Journalists in Africa are overwhelmingly using digital technology for information gathering, communicating with sources and reaching their audiences

Similar to their global counterparts, African journalists highlighted the importance of digital technology for reaching their audiences (96%), information gathering (90%) and communicating with sources (89%).

Journalists in Africa say their news organizations are paying at least some attention to tech issues, but a plurality want to see them pay more attention to AI

Journalists in Africa (28%) are more likely to say that relationships with technology companies are getting a great deal of attention from their organization (compared with 18% of journalists in the rest of the globe). They report AI in journalism is getting a similar amount of attention as do journalists elsewhere.

Journalists in Africa rank both of these topics near the bottom for how much attention they get in newsrooms, surpassing only “Online abuse, surveillance, and cyber threats.”

About half of journalists in Africa think their news organization pays too little attention to developing ways AI can help journalists and other newsroom employees do their work and to mitigating the ways it can harm their ability to do their work (49 and 54%, respectively).These responses are similar to those of journalists in other regions of the world; when we look at all journalists together, those numbers are 45 and 48% respectively.

Journalists in Africa are more positive than others about the impacts of social media and developments in technology and AI

Journalists in Africa are more positive than journalists elsewhere about the impact of both “developments in technology” and “social media” on their ability to deliver journalism. Compared with 69% and 64% of journalists overall — and thus even smaller numbers of journalists on other continents — 87% and 77% of journalists in Africa  were positive about developments in technology and social media, respectively.

Journalists in Africa are also more positive about AI’s effects on enabling an informed public about events and issues of the day: 55% of African journalists said they were positive versus 36% of all journalists (and fewer still on other continents).

Journalists in Africa are more likely than others to have used technology in a variety of ways in their profession

African journalists were more likely than journalists elsewhere to have used technology to make their writing better (74% vs. 56% overall), summarize or analyze a lot of documents or data (75% vs. 60% overall), check the accuracy of something (69% vs. 52% overall), develop story drafts (54% vs. 38% overall) and help determine how to best present a news story to the public (67% vs. 48% overall).

African journalists were similar to their global counterparts in their likelihood to use technology to edit an image (52% vs. 45% overall), search archives of their reporting or their organization’s (59% vs. 51% overall) and translate content from one language to another (both 65%). 

Journalists in Africa think their organizations are doing a fairly good job of communicating about how they use technology — and they are more likely than journalists elsewhere to say their organization communicates about specific technology uses

Journalists in Africa agree with journalists elsewhere on how well their news organization communicates different uses of technology — including AI — to the public. A plurality say they are doing a fairly good job.

Journalists in Africa were more likely than journalists elsewhere to say their organization usually communicates the use of technology to their audience when making their writing better (83% vs. 66% overall), checking the accuracy of something (91% vs. 77% overall), developing story drafts (80% vs. 66% overall) and helping determine the best way to present a news story to the public (76% vs. 63% overall). 

Journalists in Africa were similar to their global counterparts when saying their organization communicates the use of technology to their audience when searching archives of their reporting or their organizations (82% vs. 71%), summarizing or analyzing a lot of documents or data (80% vs. 69%), translating content from one language to another (78% vs. 68%) and editing an image (75% vs. 70%). 

Journalists in Africa have had their voice, image (or both) reproduced by technology at rates similar to journalists across the rest of the globe

In the past 12 months, 11% of respondents in Africa learned of someone using technology to reproduce their voice (compared to 10% of all respondents) and 17% of someone using technology to reproduce their image (compared to 15% of all respondents).

Conclusion

Journalists in Africa stand out with their optimism about the impact of technological developments — including AI and social media — on their ability to produce journalism and enable an informed public. They also report using these technologies to assist them in a more diverse range of tasks than their global counterparts and their news organizations are more likely to report this use of technology to news consumers. 

At the same time, journalists in Africa share many common experiences and concerns with journalists in the rest of the world. Like journalists around the world, they rate the importance of technology to their work as extremely high. They experience technology being used to replicate their voice or their image at similar rates. And both the African and overall respondents express a desire for their news organizations to pay at least as much attention as they currently do — or more — to the opportunities and challenges of AI. 

Overall, this research highlights that while AI is impacting newsrooms everywhere, journalists in Africa are actively embracing these tools and are notably positive about the role of technology in the future of journalism.

Part II: What the South African Public Thinks About Technology

We surveyed 1,012 South African adults between September 23, 2024 and October 16, 2024 to understand their attitudes about journalistic uses of technologies such as AI. 

The data shows that the South African public views digital technology as very important and believes that future developments will help them keep informed. Further, they are using technology for content related tasks and express openness to journalists doing the same. Generally, they are overwhelmingly more optimistic about the future of the open internet than people from Australia, Brazil and the U.S.

When asked more specifically about AI, the South African public was relatively positive about AI’s future impact, not only on journalists’ ability to report but also on their own ability to stay informed. However, they are split on how much they care about journalists’ use of AI in reporting and they use multiple factors to decide if its use is okay. Even though there is a general sense of optimism about the future of AI and journalism, South Africans have heard less about AI than people elsewhere and they are using generative AI at low rates.

The South African public feels more positive about their ability to keep informed than people in Australia, Brazil and the U.S.

About two-thirds of South Africans (67%) feel positive about their ability to keep informed.

Of South Africans that consume 0-4 pieces of journalism per week, 32% say they keep up very closely; however, among people in other countries who consume the same amount of news, less than 10% say the same (6% Australia, 7% Brazil and 9% U.S.). 

The South African public says digital technology is very important for keeping informed and is positive about its future impact

A strong majority of South Africans say that digital technology is important for keeping informed, with 84% saying it is very important and 9% somewhat important.

About three-quarters of people in South Africa (74%) are optimistic about the impact developments in technology will have on their ability to keep informed and 16% are neutral.

Although a strong majority of the South African public is positive about technology, they report more challenges with it than people in Australia or the U.S. Over half (56%) say that having technology that works is at least somewhat of a challenge in their ability to stay informed.

The South African public uses technology for content-related tasks and is comfortable with journalists doing the same

Strong majorities of South Africans report using technology “to check the accuracy of something” (67%) and “to make their writing better” (66%).

The South African public is generally comfortable with journalists using technology, with three-quarters or more reporting acceptance of each use asked about.

The South African public is much more optimistic about an open internet than Australia, Brazil or the U.S.

In South Africa, eight in ten (79%) are at least somewhat confident the internet will be a place to get and share news openly in the future.

The South African public is relatively positive about AI’s future impact on both journalists’ ability to report and their ability to stay informed

Strong majorities of people in South Africa (71%) think AI will either have a positive impact on their ability to keep informed or will have no effect one way or another.

Roughly half of people in South Africa are optimistic (49%) and about one-in-four are neutral (22%).

We also asked people to consider whether AI will have a mostly positive, negative or neutral impact on “journalists’ ability to report on issues and events.” We found that a plurality of South Africans express a sense of optimism, with 46% saying they think it will mostly have a positive impact.

The South African public is split on how much they care about AI use in reporting

About one-in-three (32%) in South Africa feel it matters a great deal and about half (52%) say it matters at least a fair amount if AI is used in the reporting process.

The South African public considers multiple factors when deciding if a journalists’ use of AI is okay

We asked about five factors people might use to decide if a journalist’s use of AI is okay. Slim majorities of people in South Africa say that each factor is at least somewhat important (55-62%).

The South African public has not heard as much about AI as people in Australia, Brazil or the U.S., and most of them have not tried to use generative AI

 In South Africa, 22% say they have heard a great deal about recent developments in AI, and 17% said they have heard a fair amount — a total of 39%. Majorities in all other countries have heard at least a fair amount (62-73%).

About a quarter of people in South Africa have tried to use generative AI (27%) in the last year while the same proportion (27%) do not know what it is. 

Conclusion

Our data shows that the South African public recognizes the growing significance of technology in the news landscape and is generally optimistic about its potential, including advancements in AI. Despite this, there is a gap between the importance placed on technology for staying informed and the challenges in having devices that work. Further, while the South African public uses technology and expresses comfort with journalists also using it, compared to people elsewhere they may be less aware of and experienced with recent innovations, including generative AI.

Overall, the data suggests that while the South African public values technology’s role in news consumption, there is a knowledge and usage gap between South Africa and other countries.

Part III: Larger Trends

The clearest common thread across both the South African public and the journalists in Africa is optimism about technology. Both of CNTI’s surveys showed a similar optimism across the Global South, in comparison with the Global North. 

Research focused on the Global South — and Africa in particular — adds further detail about this optimism. The Thomson Reuters Foundation’s 2024 survey of 221 journalists across the Global South found that about 80% of respondents report using AI tools regularly in their work and about half were doing so daily. A 2023 Africa-specific study from International Media Support showed uneven AI adoption across Africa’s public interest media: while Kenyan and South African media outlets were leading in implementation, overall adoption across the continent was relatively low. 

Challenges to implementation across Africa include concerns about transparency, accountability and ethical safeguards — particularly in the absence of official guidelines. Among the participants in the TRF study, 13% said their organizations had official policies about AI use. And most of them were navigating this space independently, as they were largely self-taught, learning by “playing with tools” or using “online courses or guides.”

The 2023 International Media Support study found similar barriers, including a gap in knowledge about AI, resource constraints and concern about algorithmic bias and harm. And a 2023 interview study of 17 journalists from five African countries who had used ChatGPT in their work by January 2023 (soon after its release) revealed an increasing skepticism at generative AI’s early stages. These journalists generally found that outputs were not particularly sophisticated or up-to-date in their contexts. While they saw potential for AI, they also saw serious limitations, especially in localized contexts where the availability of training data may be limited. Similarly, a piece on AI in Kenyan newsrooms highlighted that much of the AI technology is developed from the West, and often trained on data that reflects a global perspective that may not align with Kenya’s local context.

Turning to the African public, there is also a notable sense of cautious optimism about the use of AI in media — with variation across the continent. Both Gregory Gondwe’s 2025 study of 1,960 members of the public from 10 African countries (Burkina Faso, Cameroon, the Democratic Republic of Congo, Ghana, Kenya, Malawi, Nigeria, Uganda, Zambia and Zimbabwe) and Mphathisi Ndlovu’s 2024 study on public perceptions of an AI-powered newsreader in Zimbabwe found that while there is moderate trust in AI-generated news overall, it varies widely. Gondwe’s study found significant differences in trust based on age and also explored the role media polarization and exposure to AI-generated content plays in this variation. Ndlovu found that while some audience members liked the use of AI news anchors, others were critical due to their lack of human emotion and accents. Ndlovu highlighted the need for “decolonizing the AI tools in newsrooms,” calling for greater sensitivity to local cultures and languages to increase acceptance across diverse audience groups. 

Overall, studies on African journalists and the public show that implementation has been uneven across the continent. Widespread challenges include lack of policies, knowledge gaps and issues of AI transparency — both in newsrooms explaining its use to consumers and model developers offering insight into the data used to train these systems. From the audiences’ perspective, this transparency is a very important consideration: in general, as found in Gondwe’s study, those who thought newsroom transparency about AI use was important also had somewhat higher trust in AI-generated content.

Last but not least, there is a lot of promising work happening around transparency and localization right now, including AI initiatives being developed by, with and for African journalists and newsrooms. These tools include fact-checking, transcription and data analysis applications.

About These Studies

Like many parts of the world, newsrooms in Africa are harnessing the potential of AI in their work. However, there are numerous advantages and disadvantages to this quickly evolving technology. Together with the Centre for Journalism Innovation and Development (CJID), the Center for News, Technology and Innovation (CNTI), wanted to explore the perspectives of African journalists’ when it comes to the challenges and opportunities of AI in the news industry. 

This project is part of CNTI’s broader Defining News Initiative which seeks to understand how journalism is defined today. Access to information is not just important for its own sake; it makes democracy possible. In 2024, the United Nations outlined Global Principles for Information Integrity in response to growing challenges around misinformation, disinformation and hate speech. 

The journalist data in this article are a subset from a larger global survey, What It Means to Do Journalism in the Age of AI: Journalist Views on Safety, Technology and Government. The representative data of South Africans are from a series of country-level surveys, What the Public Wants from Journalism in the Age of AI: A Four County Survey.

How We Recruited Participants

Journalist survey: Surveys are a snapshot of what people think at a particular moment in time. CNTI collected responses from 433 journalists across 63 countries between October 14 and December 1, 2024.This data highlights the perspective of journalists around the world during that time frame.

CNTI partnered with journalism organizations in multiple continents. The questionnaire was crafted with input from partner organizations who knew the current situations and challenges of their members across various country contexts. These partner organizations also shared the survey with their membership and can be found here. 

The data representing Africa in this article reflects the responses of journalists across the continent, with 75 in Nigeria and another 34 in other African countries (N = 109). The overall data reflects the responses of 433 journalists across 63 countries with 256 of our respondents (59%) coming from three countries: Mexico, Nigeria, and the U.S. Because no global census of journalists exists, no survey can be “fully representative” of all journalists.

South African public survey: Data are from an Infinite Insight RDD CATI/cell phone sample conducted from September 23 to October 16, 2024. The total sample size was 1,012 respondents. The design effect was 1.48 and a margin of error of 3.7 points. The survey was available in English (n = 811), Zulu (n = 138), Sesotho (n = 24), Sepedi (n = 17), Setswana (n = 12) and Xhosa (n = 10).¹ All interviews were conducted by telephone, and the median interview length was 19 minutes and 17 seconds. There were 47 interviewers all of whom were trained. The sample collection age categories were: 18-24, 25-34, 35-49, 50-64 and 65+. This survey was part of a larger four-country project. 

For specific questions about the sample frames, weighting procedures and/or additional survey details, please send an email to the research team at info@innovating.news.

How We Addressed Attrition

Journalist survey: The survey was lengthy (with an estimated completion time of 20 minutes) and consisted of five distinct sections on different topics. The survey was broken into sections, and we treated each as a drop-off point: that is, if a respondent answered at least one question within a section, non-responses were treated as true non-responses. Respondents who answered no questions within a section were not included within the section’s N.

For transparency, topline tables include both percentages of the full survey N (Percent) as well as percentages of the section N (Valid Percent).

South African public survey: We recoded missing, refused and don’t know responses into a catch-all category to keep the sample sizes consistent across each question that did not explicitly have survey logic (i.e., questions that were asked to every respondent).

How We Tested for Statistical Significance

Journalist survey: This report presents findings for both (1) the overall set of journalists and (2) journalists who reported living in Africa. Thus, the African journalists are nested within the overall results. To test for statistical significance, we compared responses from African journalists with those from non-African journalists to ensure independence between the groups. Responses were compared using Chi-squared proportion tests. We used a standard threshold of p < 0.05 for assessing statistical significance. Differences mentioned in the report text are statistically significant.

South African public survey: We analyzed the results using Chi-squared proportion tests to assess differences in responses between countries. We used a standard threshold of p < 0.05 for assessing statistical significance. Differences mentioned in the report text are statistically significant.

How We Protected Our Data

CNTI did not collect any identifiable information that risked the privacy and confidentiality of participants. For the survey of journalists, data collection was supervised by CNTI staff only. The survey included individual-level information such as gender and the country where one worked and resided. It would be very difficult to identify study participants because CNTI did not collect their personal contact information or contact participants directly and the participants’ personal information was not shared by the partner organizations.

For the survey in South Africa, Infinite Insight, the country vendor, handled data collection. Demographics were collected for weighting purposes. CNTI did not receive specific locations of or contact information for survey participants.

The data are securely stored in an encrypted folder which is only authorized to the core research team at CNTI.
More information about how each survey was administered may be found here for the journalist survey and here for the public survey in South Africa.

Footnotes

1. These numbers are unweighted. ︎

The post A Window into AI and Journalism in Africa: Perspectives from Journalists and the South African Public appeared first on Center for News, Technology & Innovation.