AI summaries at the top of search results grab way more attention than the same info placed lower — why does position alone matter so much?

This explores why an AI Overview at the top of a search page has outsized effects, and whether position on its own changes what people look at, click, and trust. One caveat first: none of these notes compares an AI Overview at the top with the same Overview placed lower down. What they do show is what happens when an Overview sits in the spot people look at first, and why that spot is so powerful.

The clearest evidence comes from where people's eyes go. An eye-tracking study by RMIT and Microsoft found that AI Overviews have taken over the 'golden triangle', the top-left area that searchers scan first. People looked at the Overview for longer, and the share of attention going to the first-ranked link fell from 31% to 9% Where do searchers look when AI Overviews appear?. Searchers also rated the Overview and the ranked links as equally trustworthy. So readers didn't move to the summary because they trusted it more. It was simply the first thing they saw, and that was enough to claim most of their attention.

The click data follows the same pattern. Ahrefs found that the first organic result gets about 58% fewer clicks on searches that show an AI Overview How much are AI Overviews actually reducing organic search clicks?. That figure is a correlation, but a randomized field experiment by Agarwal and Sen points the same way. Overviews cut clicks to publisher sites by 39.8%, and users reported no better search quality in return Do AI Overviews actually reduce clicks to publisher sites?. Readers stopped where the Overview stopped them, even though they weren't more satisfied.

The less obvious point is that this kind of position bias shows up across the corpus, not just in search. Moving the same block of examples from the start of a prompt to the end can shift a language model's accuracy by up to 20% and flip nearly half its answers How much does demo position alone affect in-context learning accuracy?. Recommendation systems such as YouTube's have to correct for position in their models. Otherwise they mistake 'shown first' for 'liked best' and keep reinforcing their own earlier choices Why do ranking systems need to model selection bias explicitly?. Research on AI-assisted decisions finds that people anchor on whatever the machine offers first. Its proposed fix is to have the AI point out what to look at rather than hand over a ready-made answer Can AI guidance reduce anchoring bias better than AI decisions?. A summary at the top of the page is that ready-made answer, delivered at the moment people are most likely to anchor on it.

The top spot also changes who is speaking. A Munich court held Google directly liable for false statements in an AI Overview. Its reasoning was that the summary is Google's own content, not a neutral list of other people's pages Can courts hold Google liable for AI Overview false claims?. Put that together with the eye-tracking result and the picture is this: the first thing searchers read is now written by the search engine itself, and they trust it as much as the sources it replaced. Any mistakes in it reach people at the point where they are least likely to scroll down and check.