What did the Reuters Institute actually ask, and of whom?
The Digital News Report is run annually by the Reuters Institute for the Study of Journalism at the University of Oxford. The 2026 edition, published on 16 June 2026, is based on fieldwork conducted by YouGov using an online questionnaire from the middle of January to the end of February 2026. The total sample was 97,520 across 48 markets, with roughly 2,000 respondents in each.
Samples were assembled using nationally representative quotas for age, gender and region in every market, with education quotas applied everywhere except Kenya, Nigeria, Morocco, Peru and Thailand, and political quotas based on vote choice at the most recent national election applied in thirteen markets including the United States. Data were weighted to census or industry-accepted targets.
The limits are stated by the Reuters Institute itself and should be carried through every number below. Online samples “tend to under-represent the news consumption habits of people who are older and less affluent”. Data from India, Kenya, Nigeria and South Africa represent younger English speakers rather than the national population, because an online survey cannot reach other groups representatively in those markets; the Kenyan and Nigerian samples were restricted to ages 18 to 50.
One more limit matters more than the rest for this audience, and it is a limit of scope rather than of method. The Digital News Report asks about news. It does not ask people whether they trust what an AI assistant tells them about a company, a product or a supplier. Nobody has published that survey. What follows is the closest available proxy, and it should be read as a proxy.
How much do people trust what a chatbot tells them?
Not much, and less than they trust anything else in the survey.
Trust in the answers AI chatbots give sits at 20% globally. For context, the same survey puts trust in news overall at 37% — described by the Reuters Institute as the lowest figure since it began measuring trust in 2015, with trust falling in 29 of the 48 markets covered. Trust in news on social media is 22%. In the United States, 25% say they trust the news most of the time.
% expressing trust, global average across 48 markets
Source: Reuters Institute for the Study of Journalism, Digital News Report 2026, published 16 June 2026. YouGov online survey, 97,520 respondents, 48 markets, mid-January to end February 2026.
Bars are scaled to the largest value shown. The 37% figure is described by the Reuters Institute as the lowest since measurement began in 2015.
The gap between 20% and 37% is smaller than it looks, because 37% is itself a record low. The useful reading is not that AI is distrusted and journalism is trusted; it is that both are distrusted, and the newest interface is the least trusted of the set while growing fastest. Weekly use of AI chatbots for news went from 7% in the 2025 report to 10% in 2026, reaching 17% in the youngest age group against 5% in the oldest. South Korea, Greece and Spain roughly doubled year on year.
What people use chatbots for is a more specific answer than the headline. Across 45 markets, the most cited use is the ability to ask follow-up questions at 42%, then getting the latest news at 35%, summaries at 34%, checking the reliability of a news source at 33% and simplifying something to make it easier to understand at 30%. A third of users are asking the machine whether a source can be believed. That is the survey's most uncomfortable finding for anyone who thinks of assistants as a retrieval layer rather than a judgment layer.
Does trust rise with use, or is that selection?
The single widest gap in the AI section of the report is between users and non-users. Among people who use AI chatbots for news, 44% say they trust what those chatbots tell them. Among people who do not, 17% do. That is a 27-point spread on the same question in the same survey.
Two explanations fit that pattern and the data cannot separate them. Either exposure builds warranted confidence — people try it, it works, they trust it — or the people disposed to trust AI are the people who adopted it first, and the gap is selection rather than learning. The Digital News Report is a cross-sectional survey taken once a year, not a panel following the same individuals, so it can show the association and cannot establish the direction.
The engagement figures point the same way without resolving it. Weekly use for news runs at 19% among those extremely interested in news against 7% among those not very interested.
For a company, the practical consequence of a 44-to-17 split is that the audience is bifurcating rather than converging. A growing minority takes the assistant's description of you at something close to face value. The majority does not use it for this at all, and would not believe it if they did. Both groups matter, and no single messaging posture serves both.
Does anyone click the source?
This is the number that should change how the channel is valued, and it is the one least often quoted.
Across 27 markets, the Reuters Institute found that just 4% of respondents say they always or often click through to a source from an AI chatbot. The comparable figures are 19% from search and 17% from social media. Whatever attribution an AI answer carries, it is being read rather than followed by all but a small minority.
Behavioural measurement finds the same thing, independently and more precisely. The Pew Research Center recruited 900 US adults from its KnowledgePanel who agreed to share browsing data, and analysed 68,879 unique Google searches made in March 2025, of which 12,593 produced an AI summary. Users who encountered an AI summary clicked a traditional search result on 8% of visits; those who did not encounter one clicked on 15% — nearly twice as often. Clicks on links inside the summary itself ran at 1% of visits. Users were also more likely to end their browsing session after a page with a summary, at 26%, than after one without, at 16%.
two separate studies, two different measures
Sources: Reuters Institute, Digital News Report 2026 (self-reported, 27 markets); Pew Research Center, 22 July 2025 (observed browsing, 900 US adults, 68,879 searches, March 2025).
The two studies measure different things — stated frequency against observed visits — and are shown together only because they point the same way. Bars are scaled to the largest value shown.
Put the two findings side by side and the shape of the problem is clear. A fifth of the audience trusts the answer. Almost none of them verify it. Low trust plus low verification is not scepticism in any useful sense; it is an unchecked impression, formed once and rarely tested. We have written separately on the difference between ranking in Google and being cited by an AI, and the click gap is the practical reason those two are not substitutes.
What do the other trust surveys measure, and what do they not?
Two other bodies of survey work get cited in this argument, and neither one asks the question in the headline.
The larger is the KPMG and University of Melbourne global study, led by Professor Nicole Gillespie and Dr Steve Lockey at Melbourne Business School. It surveyed 48,340 people across 47 countries and jurisdictions by online survey between November 2024 and mid-January 2025. It found that 46% of people are willing to trust AI systems, leaving 54% ambivalent or unwilling; that willingness to trust generative AI tools specifically runs at 42 to 45%; and that trust is contextual, rising to 52% for AI in healthcare, the most trusted application in 42 of the 47 countries.
Two of its findings bear directly on how AI descriptions of companies land. Seventy per cent agreed with the statement that they find it hard to trust information online because they do not know whether content is real or AI-generated, while only 47% were confident in their own ability to identify AI-generated misinformation. And among employees who use AI, two thirds — 66% — said they had relied on AI output without evaluating the information it provided. Stated distrust and actual verification behaviour are not the same thing, and the gap between them runs in the direction that should worry you.
The other is the Edelman Trust Barometer. Its 2026 edition, released on 18 January 2026, is built on 30-minute online interviews with nearly 34,000 respondents across 28 countries, conducted between 23 October and 18 November 2025. It is the most-quoted trust instrument in the communications industry and it is not measuring what this question needs. Edelman measures trust in institutions, and in the 2026 edition its AI findings are about displacement rather than credibility: 54% of low-income and 44% of middle-income respondents believe generative AI will leave them behind rather than benefit them. None of that tells you whether a buyer believes a chatbot's summary of your company.
Nobody has published a survey asking whether people trust what an AI assistant says about a specific business. Anyone quoting one to you should be asked for the questionnaire.
What does this mean for a company being described by an AI?
Three conclusions follow from the numbers above, and one of them cuts against the way this work is usually sold.
First, the audience for an AI description of your company is small, growing and disproportionately young, and a minority of it — 44% of users — treats the output as reliable. That is enough people to matter and not enough to build a channel strategy around. The honest framing is exposure management, not reach.
Second, the description is not being checked. With 4% clicking through from a chatbot and 1% clicking a link inside an AI summary, the citation under an AI answer is functioning as a credibility signal rather than as a route to your site. That has an odd implication: the value of having a good source cited about you is mostly in what the model reads, not in what the reader visits. It also means the thing an AI assembles its answer from — the coverage, reference entries and primary pages that exist about you — is doing the work. On which sources those actually are, the independent evidence is thin, and we set out what there is in our review of which sources AI assistants cite.
Third, low trust is not protection. A third of chatbot users say they use them to check whether a news source is reliable, and the KPMG data shows most people using AI output without evaluating it. An audience that says it is sceptical and behaves as if it is not will absorb an inaccurate description without ever surfacing the doubt. The counterweight is a record that holds up when someone does look: verifiable coverage, consistent entity information, and no gap between what you claim and what a third party has published. Our guides on the reputation signals that carry weight and on whether media coverage affects B2B buying deal with that record directly.
Media Morpheus sells media and podcast placement, and one of the arguments made for that work across this industry is that it improves what AI systems say about a client. The survey evidence above establishes that the surface exists, that it is lightly trusted and almost never verified. It does not establish that any particular placement changes what appears there, and we are not going to pretend that it does.
Methodology and limitations
Primary figures are from the Reuters Institute for the Study of Journalism's Digital News Report 2026, published 16 June 2026, based on a YouGov online questionnaire fielded from the middle of January to the end of February 2026, total sample 97,520 across 48 markets at approximately 2,000 per market, weighted to census or industry-accepted targets. Five limits apply. First, the survey asks about news, not about companies, products or suppliers; its trust figures are a proxy for the question in this report's title and not an answer to it. Second, online samples under-represent older and less affluent respondents, and the Reuters Institute states that data from India, Kenya, Nigeria and South Africa represent younger English speakers rather than the national population. Third, the trust figures are self-reported attitudes; the KPMG finding that 66% of employee AI users have relied on AI output without evaluating it shows how far stated caution can diverge from behaviour. Fourth, the 44%-against-17% user gap is cross-sectional and cannot distinguish trust caused by use from use caused by trust. Fifth, the click-through comparison sets a self-reported Reuters Institute measure across 27 markets against Pew's observed browsing of 900 US adults; the two are not methodologically equivalent and are presented together only because they agree in direction. Edelman Trust Barometer figures are quoted from its 18 January 2026 press release; the full global report was not retrievable, and no AI-credibility figure from it is quoted here.
Conclusion
Trust in what AI says is low, rising in use, and almost never tested. Twenty per cent of people globally trust the answers AI chatbots give, against 37% for news overall and 22% for news on social media, on a sample of 97,520 people across 48 markets. Among the minority who actually use chatbots for news, trust more than doubles to 44%. And across 27 markets only 4% say they often click through to check — a figure Pew's observed browsing data puts at 1% for links inside an AI summary. The practical position for a company is that a small, growing and youthful audience is forming an unverified impression of you from a machine, and the only thing you can influence is the published record that machine is reading. That is a reputational problem with a reputational remedy, and it is measured in accuracy rather than in clicks.
Frequently Asked Questions
What did the Reuters Institute actually ask, and of whom?
YouGov polled 97,520 people online across 48 markets, roughly 2,000 per market, from the middle of January to the end of February 2026, for the Digital News Report 2026. Questions covered news, not companies or products.
How much do people trust what a chatbot tells them?
Trust in the answers AI chatbots give sits at 20% globally, against 37% for news overall — the lowest since the Reuters Institute began measuring in 2015 — and 22% for news on social media.
Does trust rise with use, or is that selection?
44% of people who use AI chatbots for news trust the answers, against 17% of non-users. The survey is a single snapshot, so it cannot separate use causing trust from trusting people being the ones who use it.
Does anyone click the source?
Almost nobody. Across 27 markets, 4% say they always or often click through from an AI chatbot, against 19% from search and 17% from social media. Pew measured the same behaviour directly and found clicks on links inside AI summaries at 1% of visits.
What do the other trust surveys measure, and what do they not?
KPMG and the University of Melbourne polled 48,340 people in 47 countries and found 46% willing to trust AI systems. Edelman polled about 34,000 in 28 countries but measures trust in institutions, not in what an AI says. Neither asks about AI descriptions of specific companies.
What does this mean for a company being described by an AI?
You are being described to a sceptical audience that will not click the source to check. That makes the underlying record — coverage, reference entries, primary pages — the thing that matters, because it is what the answer is assembled from and what a doubting reader finds if they do go looking.
Sources
- survey research Reuters Institute for the Study of Journalism, 16 June 2026. Emerging uses of AI chatbots for news and what it means for journalism, Digital News Report 2026.
- survey research Reuters Institute for the Study of Journalism, 2026. Overview and key findings of the 2026 Digital News Report and its methodology.
- academic Gillespie, N., Lockey, S., et al., 2025. Trust, attitudes and use of artificial intelligence: A global study 2025. University of Melbourne and KPMG. 48,340 respondents, 47 countries, November 2024 to mid-January 2025.
- survey research Pew Research Center, 22 July 2025. Google users are less likely to click on links when an AI summary appears in the results.
- press release Edelman, 18 January 2026. 2026 Edelman Trust Barometer reveals trust is in peril. Nearly 34,000 respondents, 28 countries, 23 October to 18 November 2025.