Search behaviour may not be changing as much as we think

Search behaviour may not be changing as much as we think. What may be changing is what occupies the most valuable space on the results page.

A recent eye-tracking study from researchers at RMIT University, Microsoft and the ARC Centre of Excellence for Automated Decision-Making and Society looked at how people interact with search pages that place AI-generated summaries above traditional ranked results.

The study explored a timely question: are AI Overviews changing the way people search?

The answer appears to be both yes and no.

One of the most interesting findings is that the classic “golden triangle” or F-pattern still appears to hold. In other words, users continue to focus their attention on the top-left area of the search results page, much like they have in traditional search behaviour studies for years.

What has changed is what now sits in that prime attention zone. When AI-generated summaries appear above the traditional ranked results, they receive substantially more visual attention. The study found that users still interact with traditional search results in familiar ways, but they spend significantly more time engaging with the AI Overview.

This matters because it suggests users may still behave like searchers, but the first thing they genuinely inspect is increasingly the answer layer.

The paper’s word-normalised fixation analysis is especially striking. Even after accounting for the amount of text in each area, AI Overviews received far more visual attention than traditional ranked results. The AI Overview had a mean fixation time normalised by word count of 170.39, compared with 99.50 for the first traditional result and 33.99 across the traditional ranked results overall.

That is a major shift.

It suggests that AI-generated answers are not just another feature on the page. They are becoming the primary object of attention.

The study also found that users fixated around four times longer above the fold than below it. This reinforces the importance of what appears immediately on the page before a user scrolls. If an AI answer appears there, it is likely to carry a disproportionate amount of influence over what the user reads, trusts and does next.

Another interesting result is that traditional search behaviour does not disappear after the AI Overview. Users still clicked and scanned traditional ranked results in patterns that were broadly consistent with earlier search behaviour research. The first traditional result still received the most attention and clicks among the blue links, with attention generally declining further down the list.

But the AI Overview appears to push the importance of those traditional results down the page.

The paper compares this with earlier research into direct answers in search results. In the AI Overview condition, the proportion of fixation on the first ranked result was much lower than in traditional search-only conditions. The authors suggest that when AI Overviews appear above traditional results, the first ranked result may effectively behave more like a lower-position result in terms of user attention.

For organisations, this has significant implications.

If users are spending more time with the generated answer than with the underlying links, then the quality of that answer becomes critical. It is no longer enough for the search results to be relevant. The AI-generated response needs to be accurate, grounded, current and easy to verify.

This is where trust becomes central.

The study’s qualitative findings showed mixed views on whether AI Overviews or traditional results were more trustworthy. Some participants valued AI Overviews because they were convenient, concise and easy to understand. Others were cautious because the answer was machine-generated, lacked clear sourcing, or could misrepresent the underlying material.

Several participants described using traditional results to cross-check or confirm the AI Overview.

That behaviour is important. It suggests users may appreciate the speed and clarity of AI-generated summaries, while still wanting the reassurance of visible sources and supporting evidence.

For Airgentic, this strongly reinforces the importance of bringing search and AI question-answering together in one governed experience.

Users should not have to choose between a conversational answer and a reliable search experience. They need both.

A well-designed answer layer can help users move faster, reduce friction and make complex information easier to understand. But it must be supported by strong grounding, citations, source visibility and human control.

This is especially important for organisations such as councils, universities, healthcare providers, manufacturers and enterprise knowledge portals, where users may be making decisions based on the information they receive.

If an AI answer becomes the first thing users inspect, then organisations need confidence that the answer is based on approved content, reflects the right source material and can be reviewed, improved or overridden when needed.

The paper also highlights a broader point: AI search is not simply replacing search behaviour. It is reshaping the attention hierarchy within search.

People may still scan pages in familiar ways. They may still value traditional results. They may still click through to sources. But when an AI-generated summary appears in the most visible part of the page, it becomes the first and most influential layer of the experience.

That makes answer quality a front-page issue.

For Airgentic, this is exactly why search and AI should be designed together. The answer layer needs the retrieval layer behind it. The retrieval layer needs governance, metadata, freshness, ranking and source control. The overall experience needs to help users either browse results or ask for a direct answer, depending on what they are trying to achieve.

The future of search may not be a complete break from the past. Users are still searchers. They still scan, compare and verify.

But the centre of gravity is moving.

The question is no longer just “What result ranks first?”

Increasingly, it is “What answer appears first, and can users trust it?”