The summary gets the first and longest look
When an AI summary sits above ordinary search results, it takes a large share of the reader's attention.
That is the central finding of a new eye-tracking study presented at SIGIR 2026. Researchers from RMIT University and Microsoft followed 36 adults as they completed ten information-search tasks on simulated pages modelled on Google's search interface.
Participants spent an average of 27.34 seconds looking at the AI summary and 14.82 seconds across the traditional ranked-results region. The latter contained roughly 2.8 times more words and occupied 2.7 times more space on the page.
After the researchers adjusted for word count, the difference grew sharper: 170.39 milliseconds of fixation per word for the AI summary, compared with 33.99 milliseconds across the traditional-results region.
Attention is not agreement, trust or a click. But it is the first scarce resource on a search page, and the study suggests that AI summaries are taking much of it.
The old F-pattern did not disappear
Search researchers have long described a 'golden triangle' or F-shaped pattern. People concentrate near the upper left, then scan less as they move down the page.
The new study found that pattern again. What changed was the content occupying the top-left space. The AI summary sat where the first ordinary result used to receive the strongest attention.
Across the experiment, the AI summary accounted for 59% of fixation duration. The first traditional result accounted for 9%. In an earlier study without a summary, the first result had received 31%.
The authors say this comparison suggests that a first-ranked link beneath an AI summary may receive attention closer to a fourth- or fifth-ranked link on an older results page. It is a comparison between different studies and interfaces, not a direct ranking law.
People did not stop using links. Once they reached the ranked results, they still favoured the first one and looked progressively less at results further down.
People liked the convenience without fully trusting it
The participants' own answers were less one-sided than their gaze.
Among the 33 people who completed the final qualitative questions, 15 said AI overviews were more useful, 12 preferred traditional results and six were neutral or mixed. Neither format emerged as clearly more trustworthy.
That gap matters. A concise answer can be easier to read and still leave a person uneasy about where it came from. Traditional links take more work, but they offer multiple pages that can be compared and inspected.
The recreated summaries in this experiment did not include the clickable source links found inside Google's live AI Overviews. The researchers removed them because the added interaction did not fit the eye-tracking setup.
So the study tells us how attention moved between a summary block and the links below. It does not tell us how often people inspect citations inside a current commercial AI answer.
A useful result with narrow boundaries
This was an exploratory lab study, not a measurement of billions of live searches.
The sample was small and highly educated: half held a master's degree and 22% held a doctorate. Most participants were frequent search users, and more than half reported high use of language models.
Everyone used a fixed query for each task on a controlled desktop setup. They could open result pages, but they could not rewrite a query, start a follow-up search or switch to a conversational system. The source pages were recreated from Google results collected in January 2025.
Those controls make gaze easier to compare. They also remove the messy choices that define ordinary searching on phones, at work and under time pressure.
The study is best read as evidence of a mechanism: put a large generated answer in the most visible place, and people give it attention. The size of that effect across different users, devices and live interfaces still needs larger field research.
What is confirmed, interpreted and still open
Confirmed: the peer-reviewed SIGIR paper reports a 36-person, ten-task lab study. AI summaries received significantly longer fixation and saccade times than traditional results, while the familiar F-shaped scan remained visible.
The researchers' interpretation: the first traditional result may be effectively pushed several attention positions lower when an AI summary appears above it. They also suggest that interest in the summary, not position alone, helped draw the gaze.
Still open: how the findings translate to live Google, Bing or other search products; how embedded citations change behaviour; what happens on mobile; and whether more attention produces better decisions or simply more exposure to a generated answer.
For publishers, the paper is evidence that visibility inside a summary can matter before a reader considers a blue link. It is not proof that conventional search optimisation no longer matters. People still followed the ranked list once they moved below the summary.
For users, the practical lesson is plain. The first thing that catches the eye is not automatically the best-supported answer. The extra second spent checking a source still belongs to the reader.
Sources
- Allawati et al. - An Eye Tracking Study: Are AI Overviews Changing Search Behavior?Primary peer-reviewed paper presented at SIGIR 2026. Source for the study design, participant sample, gaze metrics, qualitative findings, comparisons and limitations.
- Microsoft Research - publication recordPrimary co-author institution record used to verify the authors, venue, publication status and abstract.
- SIGIR 2026 - official conference programmeOfficial programme used to verify the paper's placement in the Search Behavior session and its conference presentation context.
- Sara Al-Lawati - reproducibility repositoryPrimary research artifact linked by the paper, containing the released experimental setup and analysis material.
