Who Controls the Truth? AI, Journalism and the Future of Information
Artificial intelligence is changing how we find, interpret and trust information. As AI systems become intermediaries between the public and the news, who decides what we are told, which sources we see and what counts as a credible answer?
For journalists, verifying information is not simply part of the job. It is what distinguishes reporting from the reproduction of claims.
A statement, however convincingly written, does not become a fact until its origins and supporting evidence have been examined. Sources are important because they allow readers to question not only what is being reported, but also who is providing the information and why.
Artificial intelligence is beginning to change that relationship.
Instead of searching for information, comparing sources and drawing conclusions, people can now ask a chatbot a question and receive a coherent answer within seconds. The answer may sound convincing. It may be informative, accurate and supported by credible references. But it may also contain errors, omit essential context or cite sources that do not support its conclusions. Even when individual claims are correct, the conclusions drawn from them may be biased.
What makes this development particularly significant is that AI does not simply direct people towards information. It increasingly interprets and presents that information on their behalf, often with an air of authority that can make its conclusions appear definitive.
The question is no longer just whether artificial intelligence can provide reliable answers. It is also who decides which information becomes part of those answers, and who is accountable when the result is misleading.
When the Answer Replaces the Source
Traditional search engines have long influenced which information people encounter. Their ranking systems determine which websites are most visible, giving technology companies considerable influence over the circulation of news.
Generative AI introduces another layer of mediation.
A chatbot can summarise competing accounts, combine material from several publishers and deliver what appears to be a complete explanation. Readers may receive the substance of a reported story without ever visiting the publication responsible for producing it.
This is not necessarily a bad development. AI can help users navigate complex subjects and understand issues that might otherwise require considerable time or specialist knowledge.
But convenience becomes problematic when the distinction between original reporting and machine-generated interpretation is difficult to recognise.
According to the Reuters Institute’s Digital News Report 2026, 10% of respondents across its surveyed markets used AI chatbots for news in the previous week, up from 7% in 2025.
The figure remains relatively modest. Yet the direction of change is important.
People are beginning to obtain information through systems that do more than organise links. They produce the explanation itself.
For journalism, this changes the relationship between the reader and the source. The technology providing the answer can become more visible than the journalist who investigated the story.
The Problem With AI-Generated Sources
One of the most concerning characteristics of generative AI is its capacity to produce credible-sounding information without sufficient evidence.
AI systems can attribute statements to the wrong publications, misrepresent an article’s conclusions or attach genuine references to claims those references do not support.
In October 2025, the European Broadcasting Union published an investigation involving 22 public service media organisations across 18 countries.
Professional journalists assessed more than 3,000 responses from ChatGPT, Copilot, Gemini and Perplexity to questions about news and current affairs.
Almost half of the responses contained at least one significant issue. Approximately one-third had serious sourcing problems, while one-fifth contained major accuracy problems.
These findings do not demonstrate that every AI answer is unreliable. They do, however, challenge the assumption that a fluent response accompanied by citations can be trusted without further examination.
A 2023 academic study by Nelson Liu, Tianyi Zhang and Percy Liang, published in the Findings of the Association for Computational Linguistics, evaluated the verifiability of four generative search engines.
The researchers found that only 51.5% of generated sentences were fully supported by citations, while 74.5% of citations supported the statements to which they were attached.
These results concerned the systems tested in 2023, rather than every subsequent AI model. Nevertheless, they illustrate a distinction that remains central to journalism: providing a citation is not the same as providing evidence.
A hyperlink can lead to a reputable publication without supporting the statement attached to it.
For a journalist, the responsibility does not end when a source has been identified. The evidence must actually support the claim.
Who Decides What AI Is Allowed to Say?
The accuracy of AI-generated information is only part of the issue.
A more difficult question concerns the people and organisations determining how these systems behave.
AI companies make decisions about model training, safety policies and how their systems should respond to controversial subjects.
Some restrictions are necessary. Protecting people’s privacy, for instance, is a legitimate concern. The difficulty lies in determining where reasonable safeguards end and editorial control begins.
In January 2026, Anthropic published a new constitution for Claude, outlining the values and priorities intended to guide its model’s behaviour.
Similarly, OpenAI’s work on collective alignment recognises the difficulty of determining how AI systems should respond to questions involving competing values. The company sought public input on its Model Spec and reported changes informed by that feedback.
Publishing these policies allows outsiders to examine some of the principles guiding the systems.
Yet the underlying issue remains: private companies are helping establish the boundaries within which increasingly influential information services operate.
This does not mean that every refusal constitutes censorship or that AI providers should be required to answer every possible question.
The challenge is to distinguish legitimate safety measures from restrictions that unnecessarily limit public debate.
A model may also give an incomplete answer for reasons unrelated to deliberate restrictions. It may lack reliable information, misinterpret a question or fail to represent the sources accurately. These problems should not be confused with one another.
Still, decisions about how controversial subjects are handled and when information is withheld have consequences for public knowledge.
Unlike a newspaper, whose editorial responsibility is ordinarily associated with identifiable editors and publishers, AI systems can make the chain of responsibility considerably more difficult to follow.
A New Form of Editorial Power?
The idea that technology shapes public information predates generative AI.
In a 2014 academic article, media scholar Philip M. Napoli examined algorithmic media production and consumption through institutional theory, arguing for closer attention to the role of algorithms within media systems.
His work provides a useful framework for understanding how technological systems influence established institutions, including those responsible for producing and distributing news.
Writing in Digital Journalism in 2019, Nicholas Diakopoulos argued for bringing journalistic values into the design and evaluation of automated news technologies.
Generative AI takes these debates into new territory.
A search engine can influence which publication a reader encounters. A generative system can go further, influencing which claims and interpretations appear in the answer itself.
It is important not to confuse this with conventional editorial judgement.
Journalists can be held accountable for their reporting. AI systems generate outputs through technical processes that do not involve the same professional responsibilities.
But the effects of those outputs can nevertheless resemble editorial decisions.
Which sources are given prominence, and which perspectives disappear from the answer? More importantly, can users understand why?
These are no longer exclusively questions for newsrooms.
What Happens to Independent Journalism?
Original reporting is expensive. An investigation can take weeks of examining documents and speaking to sources before a journalist has enough evidence to publish a single story.
AI services can then use publicly available information from that reporting to produce summaries and answers.
This creates a potential imbalance between those who invest in producing reliable information and those who benefit from distributing its substance through automated systems.
A 2025 investigation by Columbia University’s Tow Center for Digital Journalism examined how generative search tools retrieve and cite news material. The research found substantial problems with source attribution, raising concerns about how original journalism is represented in AI-generated answers.
The issue also extends to website traffic.
Cloudflare’s 2025 analysis examined the relationship between AI crawling and referrals to content providers. It highlighted an imbalance between the frequency with which some AI systems access online material and the traffic they subsequently direct back to the websites producing it.
However, the consequences for journalism are not entirely straightforward.
The Reuters Institute’s 2026 research suggests that some people using AI chatbots for news actively seek original sources. Indeed, 42% of chatbot users for news said they always or often click through from chatbot answers to the original news providers.
That figure is based on respondents’ reported behaviour, rather than a direct measurement of website referrals. It nevertheless suggests that AI may also become another route through which audiences discover journalism.
But that outcome cannot be assumed.
If AI systems continue to rely on publishers’ reporting while directing fewer readers to their websites, the economic foundations of independent journalism could weaken.
There is a longer-term risk here. If the institutions responsible for producing verifiable information become less sustainable, AI services themselves may have fewer reliable sources on which to draw.
The technology could end up undermining part of the information ecosystem on which its usefulness depends.
Can AI Strengthen Journalism?
The argument should not be reduced to a conflict between journalists and technology companies.
AI offers genuine opportunities for journalism.
For an investigative journalist, AI can help examine large collections of documents considerably faster than would otherwise be practical. It can also make reporting more accessible across languages.
These capabilities may be particularly useful for smaller newsrooms with limited resources, although their effectiveness depends on how the technology is implemented and checked.
There is an important difference, however, between using AI to support an investigation and allowing AI-generated material to substitute for independently verified reporting.
An AI tool might help a reporter locate a relevant document among thousands of files. It cannot remove the journalist’s responsibility to read that document, understand its context and establish whether it supports the story.
The same principle should apply to AI services presenting news directly to the public.
Readers should be able to trace information back to its original source. When AI systems make mistakes, there should also be a clear mechanism for correcting them and explaining what went wrong.
AI companies and news organisations should both be subject to scrutiny rather than relying exclusively on their own assurances of reliability.
Why Independent Journalism Still Matters
Journalism has never operated outside structures of power. Publishers and technology platforms have long influenced how information reaches the public.
Artificial intelligence does not create this problem from nothing, but it changes the relationship between the source and the reader.
When a chatbot produces an apparently authoritative answer, the reader may never see the original reporting or understand how the system arrived at its conclusions.
This is where the debate goes beyond fabricated citations. If a small number of technology companies increasingly mediate access to public knowledge, their decisions about which information to present deserve scrutiny.
As a journalist, I recognise the possibilities AI offers. It can help reporters examine material that would otherwise require weeks of work. It can also make complex reporting more accessible to audiences who might never encounter it through conventional news channels.
Yet those advantages depend on maintaining the distinction between generating an answer and establishing the truth.
Independent journalism matters because it allows claims to be investigated and powerful institutions to be questioned. Its value rests on the evidence behind the story, not simply on how convincingly the story is told.
The future of journalism will depend partly on how successfully it adapts to artificial intelligence. But the future of trustworthy AI may depend just as much on the survival of independent journalism.
References
- Liu, N., Zhang, T., & Liang, P. (2023). Evaluating Verifiability in Generative Search Engines. Findings of the Association for Computational Linguistics: EMNLP 2023, 7001–7025. https://doi.org/10.18653/v1/2023.findings-emnlp.467
- Napoli, P. M. (2014). Automated Media: An Institutional Theory Perspective on Algorithmic Media Production and Consumption. Communication Theory, 24(3), 340–360. https://doi.org/10.1111/comt.12039
- Diakopoulos, N. (2019). Towards a Design Orientation on Algorithms and Automation in News Production. Digital Journalism, 7(8), 1180–1184. https://doi.org/10.1080/21670811.2019.1682938
- Egan, J. (2026). Overview and Key Findings of the 2026 Digital News Report. Reuters Institute for the Study of Journalism, University of Oxford. Original report
- European Broadcasting Union (2025). News Integrity in AI Assistants. Original research
- Jaźwińska, K., & Chandrasekar, A. (2025). Tow Center's Latest Report on AI Search Engines. Columbia Journalism School. Original research
- Cloudflare (2025). The Crawl Before the Fall… of Referrals: Understanding AI's Impact on Content Providers. Original analysis
- Anthropic (2026). Claude's New Constitution. Official publication
- OpenAI (2025). Collective Alignment: Public Input on Our Model Spec. Official research
The Daily Five