Where do searchers look when AI Overviews appear?

A within-subjects lab eye-tracking study of ten search tasks, run by researchers at RMIT University with Microsoft (SIGIR '26), finds that placing an "AI Overview" above Google's traditional ten-blue-links results changes where searchers look without overturning the known "golden triangle" or F-shaped scanning pattern — it just moves which element sits inside it. Using a Tobii Pro Fusion eye tracker, the study reports "AI Overviews received significantly longer eye fixation and saccade times than Traditionally Ranked Results." Comparing fixation-duration proportions across three SERP conditions, the authors find the share of attention on the first-ranked organic result drops from 0.31 (traditional-results-only SERP) to 0.20 (a single direct-answer condition from prior work) to just 0.09 when an AI Overview is present — a figure the authors say is "comparable to the proportions of the 4th and 5th Ranked Result positions (0.11 and 0.07)," concluding "the 1st Ranked Result is effectively shifted to the 4th or 5th position when AI Overviews are placed above."

The authors' explanation is attentional, not merely positional. Citing an earlier finding that users skipped past uninteresting sponsored ads straight to the first organic result, they argue the opposite is happening here: "we speculate that our users examined AI Overview content due to interest, not just because the content appeared at the top." The F-pattern itself held — "the top-left position continuing to attract the most attention" — but the content occupying that position changed. Despite the attention shift, the study's qualitative exit interviews found "neither AI Overviews nor Traditionally Ranked Results perceived as more trustworthy or useful than the other": the overview wins eyeballs without winning a corresponding trust judgment.

This decouples attention from trust in a milder way than Do AI writing tools improve online discussion or degrade it?, which found AI tools raising engagement while perceived quality and authenticity actively fell — here trust stays flat rather than dropping, even as attention redirects sharply toward the AI content. It also suggests a mechanism behind why Does disclosing AI assistance make readers trust articles less? found only a small trust penalty for AI-labeled content: if viewers do not rate AI Overviews as more or less trustworthy than ranked results to begin with, a disclosure has less of a prior trust gap to erode. And it extends Which clarifying questions actually improve user satisfaction?'s point that interface design, not just content quality, governs how search behavior unfolds — here the mere position and framing of an AI Overview is enough to re-route a scanning pattern long treated as stable.

The study is explicitly exploratory, run on simulated SERPs with prewritten queries carried over from an earlier 2024 study and backstories drawn from a fixed test collection that excluded health, medicine, and politics topics; the authors themselves flag unbalanced task-complexity counts as a confound for future work. It reports aggregated attention and trust comparisons rather than breaking results out by participant demographics or prior AI familiarity, and it measures gaze plus self-reported trust and usefulness, not downstream outcomes like comprehension, decision accuracy, or what users actually clicked through to afterward. So the finding establishes that attention reallocates toward AI Overviews without establishing that users learn more, decide better, or verify sources any differently once that reallocation happens — a gap the authors propose to close with grouped, demographically stratified follow-up studies.