People are handing more of their questions to AI assistants, and new research puts numbers on how much of what comes back gets taken on faith. The Artificial Intelligence Publisher Impact Study, conducted by Ipsos for the UK's Association of Online Publishers on a nationally representative sample of 1,000 people, finds that 37% of respondents do not know AI tools can fabricate information, Press Gazette reports. Among 45-to-54-year-olds the share rises to 45%.

The behavior data points the same way. When respondents completely trust the material an answer draws on, their confidence in the AI answer runs above 90%, and 32% of those most trusting users show what the study calls "stop here" behavior: they take the answer as delivered and click nothing further. The research was sponsored by a publisher trade body with its own stake in the findings, which is worth noting, but the sample is nationally representative and the pattern holds across the study.

The direction of travel is the story. AI assistants are moving from novelty to advisor: people ask, get an answer, and act on it. A user who does not know the tool can fabricate has no reason to verify anything, and a user who trusts what they are shown clicks nothing. Both groups experience the AI's answer as the information itself, not as a pointer to it.

The takeaway for brands: for a growing share of your buyers, what an AI model says about you is the final word, not the starting point for research. Nobody in that 32% is clicking through to your site to double-check. The model's opinion of your brand, formed from everything it has read, is the version of you those buyers meet, which makes knowing what AI believes about your brand, and correcting it where it is wrong, a first-order marketing problem rather than a curiosity.