AI use is rising fast—but only 12% are comfortable with fully AI-made news
A Reuters Institute survey of about 18,000 people across nine countries found widespread AI exposure alongside persistent caution about AI-produced journalism. The online quota sample measures reported behaviour and attitudes, not observed use or trustworthiness.
The short version
- What happened
- YouGov surveyed approximately 2,000 people in each of nine countries between 14 August and 28 September, using online quota samples weighted for demographic and political variables.
- Who it affects
- Readers increasingly encounter AI before deciding to use it. Newsrooms that deploy it still need visible human responsibility, clear sourcing and a way to correct errors rather than treating familiarity as consent.
- What remains unproven
- About 18,000 respondents
Keep in mind
About 18,000 respondents
Nine countries
Online non-probability panels
Weighted quota samples
Self-reported recall
Six-country trend comparison
No direct use logs
Industry-related funding disclosed
Source analysis · 1 linked source. Primary analysed source date: 29 October 2026.
Read the full report · 4 min The evidence, methods, implications and limitations
AI became routine faster than confidence in AI-made news
Generative AI has moved into ordinary information habits, but that does not mean people want it to replace journalists. A Reuters Institute report based on a YouGov survey of about 18,000 people across nine countries found that weekly use of standalone AI systems reached 51% in 2026 across the six countries followed since 2024, up from 34% a year earlier. Daily use in those recurring countries rose from 16% to 28%. Across all nine countries, 57% said they had seen an AI-generated search summary in the previous week.
Comfort with AI-produced news remained much lower. Twelve per cent said they were comfortable with news made entirely by AI, rising to 19% when some human oversight was added. Comfort reached 45% when a human did most of the work with some AI help and 67% for entirely human journalism. In every country, no more than one in six respondents accepted fully AI-made news.[1]
The survey measures recollection and attitudes, not what systems delivered
YouGov fielded the online questionnaire from 14 August to 28 September in Argentina, Australia, Denmark, Finland, France, Japan, Norway, the UK and the US. Each country contributed roughly 2,000 respondents. Quotas covered age, gender, region, education and political leaning, and results were weighted to population targets. Australia, Finland and Norway were new in 2026, so trend comparisons use only the six countries present in all three annual waves.
This was a non-probability online panel, which means a conventional margin of error cannot be calculated. People with limited internet access are likely to be under-represented, and answers depend on memory and personal interpretations of terms such as “AI agent”. The report therefore cannot verify whether a respondent actually used a system, whether an AI answer was accurate or whether greater familiarity caused trust. The authors say differences of two percentage points or less should be treated with particular caution.[1]
The distinction that matters
Exposure, use and acceptance are different measures
The report separates encountering AI, choosing to use it and accepting it as a news producer.
Exposure
57% saw AI search summaries in the last week
AI can reach people who did not choose a chatbot
Use
51% used standalone AI weekly in six recurring countries
Reported uptake rose year on year
News comfort
12% accepted fully AI-made news
Human responsibility remained strongly preferred
Publishers have an adoption problem and an accountability opportunity
For readers, the findings suggest that AI is increasingly an information intermediary before it is a trusted news author. For publishers, human responsibility remains the clearest source of comfort: translation, editing and summaries may be more acceptable than handing editorial judgement to an automated system. A label alone does not establish accuracy, independence or accountability.
The report itself discloses an unusually relevant set of interests. Reuters News supplied seed funding, with additional foundation and university support and core funding from the Thomson Reuters Foundation. It also says ChatGPT 6, Claude Opus 5.5 and Claude Fable 5.1 assisted data analysis, copy-editing and text suggestions, while the authors reviewed the manuscript line by line. Confidence would change with passive measures of actual use, repeated interviews with the same people and experiments testing whether transparent human review changes trust after readers see accurate and inaccurate AI-assisted reporting.[1]