The Reuters Institute for the Study of Journalism at Oxford didn’t set out to measure trust in any particular news story. It set out to measure something broader: what people currently believe is happening, industry-wide, when they open a news app or a homepage, and whether they think that shift has made the experience better or worse. The answer split in a way that’s unusual even by the standards of AI skepticism research.
What the survey actually asked
The Generative AI and News Report 2025, produced by Felix Simon, Rasmus Kleis Nielsen and Richard Fletcher and published on October 7, 2025, surveyed 12,217 people across Argentina, Denmark, France, Japan, the UK and the US, fielded by YouGov between June 5 and July 15, 2025.
It’s a companion piece to the Institute’s better-known annual Digital News Report, built specifically to track public perception of AI’s role in journalism rather than news consumption habits generally.
The numbers
Fifty-one percent of respondents said they believe generative AI is used “always” or “often” by news media to produce content, according to the report’s full data. That’s a belief about the industry as a whole, not a claim any single respondent is making about a specific article they read that morning, but it puts a majority of the surveyed public in the position of assuming AI involvement is now the default rather than the exception. Asked whether that involvement has made their experience of news better, only 26% said yes, 8% “much better” and 18% “somewhat better.” News media was one of only three sectors in the entire survey, alongside government and politicians, where more people expected AI to make things worse than expected it to make things better.
What “better” was actually being asked about
The report frames this as a comfort gap rather than a knowledge gap: people aren’t confused about what generative AI can do for a newsroom, faster summarization, automated aggregation, translation at scale, they’re specifically unconvinced that any of it translates into a better experience for them as readers. Nieman Lab’s coverage of the same report, written by co-author Felix Simon, notes that this skepticism holds even as personal use of tools like ChatGPT has roughly doubled year over year among the same population. People are adopting the technology individually while remaining unconvinced that its adoption by the institutions producing their news is doing them any favors, which is a more specific and less resolvable finding than generic AI anxiety.
Why news sits with government and politicians
The company news media keeps in this ranking is worth sitting with. The other two sectors where pessimism outweighs optimism are government and politicians, institutions people already have structural reasons to distrust independent of AI. News media landing in that group, rather than with sectors like healthcare or education where the survey found more balanced or positive expectations, suggests the skepticism isn’t a generic technology worry. It’s specific to institutions the public already suspects of not fully acting in their interest, and AI adoption is being read through that pre-existing lens rather than evaluated on its own terms.
What the gap actually represents
A newsroom can adopt generative AI for entirely defensible reasons, cost pressure, speed, coverage of stories that would otherwise go unreported, and still run directly into a readership that has already decided the outcome will be worse regardless of the reasoning. That’s a harder problem than a disclosure label solves, because the report’s own data shows the belief that AI is already pervasive is not paired with any expectation that pervasiveness is good news.
Newsrooms weighing how visibly to adopt these tools are not choosing between an informed public and an uninformed one. They’re choosing how to operate in front of a public that already assumes the adoption happened and has already made up its mind about what that means.
A perception that may already be running ahead of the practice
Nothing in the report claims that generative AI actually is used “always or often” across the news industry, only that a slim majority of the public believes it is. Actual adoption rates vary enormously by newsroom, market, and beat, and plenty of reporting involves no generative AI at all.
That gap between perceived and actual usage is its own finding: distrust of a technology’s role in an institution can outpace the institution’s real adoption of it, particularly when the institution in question, like government and politicians, was already a low-trust category before the technology arrived.
Correcting the record on actual usage rates might do less to move opinion here than newsrooms would like, because the report suggests the skepticism isn’t purely a response to what AI is doing, but to who is doing it.
What this means for how disclosure gets read
That has a direct bearing on disclosure policy. A newsroom that clearly labels its AI-assisted content is answering the question “did AI touch this,” which the data suggests readers have mostly already decided the answer to regardless of the label. What the label doesn’t answer, and what the survey suggests actually drives the negative sentiment, is whether the reader trusts the institution’s judgment about when and how to use the tool at all. Disclosure changes what readers know. It doesn’t obviously change the more structural distrust the Reuters Institute data is picking up.
The one number in the report that complicates any tidy conclusion is the adoption rate among the same respondents: roughly double what it was the previous year. People are not opting out of generative AI in their own lives while condemning it in the news. They’re doing both at once, which suggests the objection was never to the technology in the abstract, only to trusting an institution’s judgment about where and how to deploy it on their behalf.
Whether that changes as the technology matures, or hardens further, is likely to depend less on newsroom disclosure policy than on whether the institutions using it can rebuild the baseline trust the survey suggests was already thin before AI entered the picture.
