AI answers are only as trusted as the news brands they cite
TL;DR:
- An Ipsos survey of 1,000 UK adults for the AOP found trust in an AI answer tracks trust in the cited source.
- A completely trusted source yields over 90% trust in the answer; a distrusted source yields around 10%.
- 37% of respondents did not know AI tools can fabricate information or sources.
New research from the Association of Online Publishers, based on an Ipsos survey of 1,000 UK adults, quantifies something AI platforms have been reluctant to concede: the credibility of an AI answer is largely borrowed from the publishers it cites.
Borrowed credibility, measured
The correlation is close to linear. A completely trusted source produces more than 90% trust in the AI answer; a completely distrusted one produces around 10%. A source rated neither trusted nor distrusted yields only about 25% — neutrality is not enough to carry an answer.
Trust also drives clicks. Almost half of users who trust a cited source will click at least one link, while distrusted sources are barely clicked. But the study finds a counterweight: “stop here” behaviour peaks at both extremes, reported by 32% of those who completely trust the sources and 35% of those who completely distrust them. Confidence in a citation can substitute for visiting it.
The awareness gap is the more troubling figure. Some 37% of respondents did not realise AI tools can fabricate information or sources, rising to around 45% among 45-to-54-year-olds. That sits alongside UK findings that chatbots have given unreliable advice on voting — errors reaching an audience that substantially does not expect them.
The AOP’s argument is that publishers hold more leverage than they exercise, weakened by confidential licensing deals that prevent collective bargaining. UK publishers did gain one lever in June, when the CMA ruled Google must let them opt out of AI Overviews.
Looking forward
For UK businesses beyond publishing, the finding generalises uncomfortably. If perceived authority transfers from citation to answer, then any AI interface citing your organisation is trading on your reputation — and any interface citing a weak source while sounding authoritative is a liability you do not control.