The Answer Box Did Not End Search
The question here is not whether AI Overviews attract attention; the eye-tracking evidence has already shown that they do. The more useful question for AI-mediated search is what happens to the ranked results after the generated answer has taken the first look. The evidence points to coexistence rather than replacement: traditional links still organize navigation and receive rank-dependent attention, while the overview changes when and why searchers reach them. That distinction helps analysts treat ranked pages as part of a changed search path instead of assuming that lower initial attention makes them irrelevant.
Rank still organizes the list
The eye-tracking study recorded more than where participants looked. It also logged interactions with the results page, including link selections, the time until a participant’s gaze first arrived at each ranked position, and how many pages they visited. Across the traditionally ranked results, the first result attracted the most attention and clicks, while attention and clicks generally declined farther down the list. Arrival time increased as participants scanned toward lower positions. Click rate and rank were strongly associated: the reported Spearman correlation was 0.94, with p = 0.0001.
That pattern matters because it shows that the generated block did not erase the ordinary ordering mechanism. Once a participant entered the ranked list, position still shaped the probability and timing of engagement. A result at position one remained more visible than a result at position ten. The AI Overview altered the conditions under which the list was encountered, but it did not turn the ten results into ten equally likely alternatives.
The two effects can therefore coexist. The overview can receive more total attention than the traditional-results region while the first ranked link remains the strongest individual link within that region. One statement concerns the competition between page regions; the other concerns the ordering of links inside the ranked list. Treating them as contradictory would confuse levels of comparison.
The difference is especially important for a brand or retailer trying to understand its place in search. A result may lose some early exploratory attention because the overview satisfies the broad information need. Yet if the searcher continues downward, rank still affects whether that result is noticed and selected. The result’s role may shift from being the first answer to being the first external source, a comparison point, or a place to verify a claim. Those functions are not interchangeable, and a single click count does not identify which one is occurring.
The interface can thus be understood as a sequence with two hierarchies. The first hierarchy places the generated overview before the ranked list. The second hierarchy orders the links within the list. The first hierarchy changes the starting condition; the second continues to govern what happens after a searcher enters the traditional-results layer. This is why “AI versus search” is a misleading binary. The observed page contains both, and the first layer changes the visibility and timing of the second.
A further implication follows for measurement. If an analyst sees less activity on the first ranked result, they should not immediately infer that the result has lost all influence. Its influence may have become conditional on an earlier interaction with the overview. A searcher who clicks after reading the generated answer may have a different purpose from one who clicked the first result in a list-only interface. The available evidence does not identify those purposes directly, but it shows why position and sequence must be retained when interpreting them.
Arrival time adds another piece of the relationship. In the ranked list, later positions are encountered later, so a result’s opportunity to be considered depends partly on whether the searcher continues after the overview. A lower-ranked page may never enter the user’s field of attention, while a first-ranked page may be reached only after the generated answer has set an initial frame. These are not simply impressions of “more” or “less” visibility; they are different routes through the same interface. An analytics model that collapses them into a total page-view count loses the sequence that gives the event its meaning.
The result also cautions against replacing one simplistic story with another. It would be wrong to say that rank no longer matters because a generated block appears first. It would be equally wrong to say that traditional ranking explains the whole page as it did before. The generated layer adds an upstream condition to the ranking system. The right comparison is therefore not a contest with one winner, but a changed path in which the first layer filters who reaches the second and why.
A summary can be useful without being trusted
Attention is only the first layer of the search decision. The study also asked participants how useful and trustworthy they found AI Overviews compared with traditional results. The answers were mixed rather than decisive.
Among 33 participants who responded in writing about trustworthiness, 15 considered the traditional search results more trustworthy, 12 considered the AI Overview more trustworthy, and 6 gave mixed responses or treated the two as equivalent. For usefulness, 15 preferred the AI Overview, 12 preferred traditional results, and 6 were neutral or mixed. The counts do not identify a universal winner. They show that usefulness and trust can point in different directions for the same person.
Participants who favored the overview’s trustworthiness or usefulness emphasized convenience, readability, concise synthesis, and the fact that one response could aggregate information from multiple sources. Those who distrusted it worried about the origin of the claims, possible hallucination, missing context, and the difficulty of knowing what had been included or left out. Several participants described a verification habit: they used the traditional results to check the summary even when they found the summary convenient.
This is a crucial distinction for interpreting the eye data. A participant may spend more time examining an overview because it is the most efficient way to understand the topic, because it raises doubts that require inspection, or because its position makes it hard to avoid. Fixation alone cannot tell these explanations apart. The self-reports do not resolve the ambiguity completely, but they make one conclusion harder to defend: that visual prominence automatically means reliable acceptance.
The overall User Engagement Scale result reinforces the mixed picture. Engagement was moderate overall, at 3.58 on the study’s five-point scale. Perceived usability averaged 4.19 and reward 3.90, above aesthetic appeal at 3.13 and focused attention at 3.09. Participants could experience the interface as usable and rewarding without reporting complete confidence in the generated content. Convenience can coexist with verification; a tool can save time at the first stage while creating a need to check its basis at the next.
That split is easy to miss when a team uses one broad satisfaction score. A person may judge an overview useful because it gives a fast orientation, then judge a traditional page more trustworthy because its author, evidence, or scope is easier to inspect. “Useful” answers whether the interface helps with the immediate task. “Trustworthy” answers whether the person is prepared to rely on what it says. The two judgments can converge, but the study’s responses show that they should not be assumed to do so.
The distinction between utility and trust is not merely psychological detail. It changes the role of traditional pages in the journey. If an overview is useful for orientation but not fully trusted, a traditional result becomes a verification surface. The user may not visit it to discover the topic from scratch. They may visit it to compare wording, inspect detail, find a source, or test whether the summary omitted an important qualification. A reduction in exploratory visits can therefore accompany a continued or even more deliberate role for selected external pages.
The diagram represents alternatives, not a claim that every searcher follows one fixed route. Some people may stop after the overview. Some may move to ranked results immediately. Others may return to the list only after noticing an uncertainty. The important point is that the generated answer creates a branching condition before the traditional results can play their familiar role.
Verification is a form of continued search
When a searcher uses a ranked page to cross-check an AI Overview, the external visit can look less like discovery and more like due diligence. That distinction is easy to lose in a channel report. A page view is usually counted as a page view whether it introduced a topic, supplied a missing detail, or challenged a generated statement. The eye-tracking study does not provide enough evidence to quantify those roles, but participant comments make verification a plausible interpretation of why traditional results remain in use despite lower first-glance attention.
This also explains why source visibility matters. The overview can synthesize content without requiring the searcher to open the pages that informed it. If the searcher trusts the synthesis, the traditional source may never receive an event. If the searcher doubts it, the source may receive a purposeful visit. In both cases, the generated layer changes the meaning of the downstream trace. No click can mean that the answer was sufficient, that the source was not needed, or that the user abandoned the search. A click can mean curiosity, comparison, or verification. The event alone does not tell us which explanation is correct.
The traditional-results layer retains value precisely because it exposes more variation than a single summary. Different pages may emphasize different years, methods, examples, or qualifications. A generated overview can be easier to read, but a searcher who needs to inspect disagreement may prefer the underlying pages. The study’s qualitative responses describe this tension: participants appreciated a concise introduction while still wanting access to detailed and independently attributable information.
This makes verification a distinct but difficult-to-measure form of continued search. A traditional page may receive fewer casual visits yet matter more to people who have already formed a provisional view and want to test it. A summary that sends a person to one source can also narrow the visible set of alternatives, even when the person’s need is not merely to find an answer but to understand why that answer should be accepted. The eye-tracking study cannot estimate how often this happens, but it shows why the same downstream visit should not automatically be interpreted as the same kind of discovery event it was before generated summaries appeared.
For marketers, this changes the question from “Did the AI Overview steal the click?” to “What job was the ranked result doing after the overview?” That job may still be commercially important. A product or brand page that supplies specifications, availability, evidence, or a direct offer can matter later in the path even if it no longer wins the first moment of attention. Conversely, a high ranking cannot guarantee a visit when the overview has already resolved the user’s immediate need.
The evidence does not support a single strategy for every query. For a straightforward informational need, a summary may satisfy the searcher quickly. For a question where trust, detail, or disagreement matters, the ranked list may remain essential. The study does not establish how these patterns vary across all task types or user populations, so any operational response should treat them as conditions to investigate rather than universal rules.
What the mixed result means
The strongest conclusion is relational. AI Overviews change the first layer of the page, but traditional search remains a structured second layer. The overview can attract more gaze, concentrate early interpretation, and alter why a person reaches for an external result. Rank continues to shape which traditional result is seen and selected after that decision. Trust and usefulness then determine whether the searcher stops, verifies, or keeps browsing.
Several boundaries keep that conclusion from becoming too broad. The study is exploratory and lab-based, uses a simulated interface and pre-typed queries, and does not allow participants to refine a query or begin a later search session. The task-complexity groups and overview sizes were unequal, and the trust and usefulness findings come from a small qualitative response set. The findings therefore describe a credible pattern in the tested setting, not a population estimate of trust, a guarantee of search behavior, or evidence that a generated answer is more accurate than a traditional page.
The practical lesson is to preserve the ranked-results layer in both analysis and judgment. Lower initial attention does not equal total substitution. A page may still influence the decision as a source of detail or verification, but its role may be harder to see if reports track only first clicks. The next measurement challenge is larger than one search page: if an assistant forms a shortlist and moves comparison into a conversation, how will the resulting buying journey look when only its downstream pages remain visible?