Click Behavior on Google Pages With AI Overviews Present
Informational queries lose the most clicks when AI Overviews appear.
Click behavior on Google pages with AI Overviews present fractures by query intent. Informational searches lose the most clicks because the answer can replace the website visit, while navigational and transactional searches retain a reason to leave Google. Intent is therefore the master variable for predicting traffic outcomes, with device, other results-page features and citation presence shaping each site's exposure.
How AI Overviews work on the results page and why they change the click calculus
Google places a generated summary above all organic results in most cases, commonly called position zero. Ahrefs' research found that the vast majority of AI Overview appearances occupy this top position. The summary draws on multiple sources to synthesize an answer before the user reaches an organic listing. Unlike a featured snippet, which pulls a passage from one page, an AI Overview does not give a single source ownership of the answer position. It can therefore resolve the query without requiring a visit to any of the pages that supplied the information.
That placement changes the decision from which result to click to whether a click is necessary. An organic listing may retain its ranking while losing the need that previously sent the user to it. Agarwal and Sen's randomized experiment, published on SSRN in April 2026, found that removing AI Overviews substantially increased outbound organic clicks while sponsored clicks stayed flat. The observed effect was not simply a redistribution of visits across organic and paid listings. The substitution was specific to organic website visits, with sponsored clicks unaffected in the experiment.
How widespread AI Overviews have become across the query landscape
AI Overviews appear in a large and growing share of Google searches, especially when the queries being studied lean toward informational research. Their presence is now a baseline condition for exposed content portfolios. Agarwal and Sen described the share of observed queries that triggered an AI Overview in their experiment as "not marginal". The practical significance is sustained exposure across a body of queries.
But how often they appear depends on the dataset, device and mix of queries being measured. An industry panel concentrated on informational and commercial research searches will not describe the same landscape as a broad sample of all queries. Most AI Overview appearances are on desktop, but appearances on mobile are growing faster. Those differences make a universal prevalence estimate less useful than a clear account of what a sample contains. A site's exposure therefore depends on how much of its own search demand falls within the exposed query sets.
Why query intent — not ranking position — is now the master variable for predicting click loss
Query intent determines how often AI Overviews trigger, and that trigger rate determines exposure to their click effects. Seer Interactive's April 2026 study found that informational queries triggered AI Overviews much more often than commercial or transactional queries. Ahrefs' keyword analysis likewise found that nearly all keywords triggering AI Overviews were informational in intent. These findings describe different measures, namely the trigger rate within an intent category and the intent composition of triggering keywords. Both place informational demand at the center of the exposure.
The next link in the chain is whether a summary can satisfy that demand. Informational searches often need an answer, whereas transactional and navigational searches need a destination where the user can act. Intent thus governs both the likelihood of encountering an AI Overview and the likelihood that it replaces a visit. Ranking position alone captures neither distinction. Two pages at position one can have radically different click-through rate outcomes because an AI Overview appears above one and not the other.
This is why intent, rather than rank, is the master variable for predicting click loss. It makes rankings alone insufficient as a traffic forecast. The content's intent profile establishes the exposure before ranking position can explain the remaining opportunity.
What happens to clicks on informational queries when an AI Overview appears
Informational queries bear the clearest click losses because a synthesized answer can complete the task on the results page. A definition, a date or a quick explanation may leave no residual need to open a website. Evergreen and how-to content is exposed to this substitution because the information itself is often the destination. The relevant loss is the disappearance of the next step that previously connected the search to a publisher's page.
Agarwal and Sen's experiment found a large, statistically significant increase in outbound clicks on informational queries when AI Overviews were removed. Ahrefs' updated keyword study also found a substantial decline in position-one click-through rate when an AI Overview was present. The experimental finding identifies the effect of the feature, while the keyword study shows the same direction in a different dataset. Even an informational page with a strong organic ranking can lose clicks when the results page gives an answer that meets the user's need.
That distinction matters when interpreting a traffic decline. A page can still rank well among organic results as the pool of organic clicks shrinks.
How navigational and transactional queries follow a different pattern
Navigational and transactional searches preserve a reason to click that an informational summary cannot usually remove. A user seeking a specific site still needs to reach it, while a user intending to buy or book needs a destination where that action can occur. Seer Interactive's April 2026 study found that transactional queries triggered AI Overviews at a low rate relative to informational queries. Agarwal and Sen's experiment found that click effects for navigational and transactional queries were not statistically distinguishable from zero. The experiment therefore did not detect the substitution observed for informational searches.
This contrast explains why a universal traffic-loss assumption is misleading. Transactional and navigational content remain relatively insulated because of lower trigger exposure and a continuing need for a destination. However, transactional trigger rates have risen, narrowing the gap, and an AI Overview may appear within a page containing other features that compete for attention. The insulation of a query category remains conditional as trigger rates and competing features change.
How device type and SERP feature combinations amplify or moderate the intent-based effects
Desktop audiences skew toward research-heavy informational queries, making the overlap between desktop use, informational intent and AI Overview presence especially exposed. The early-2026 pattern included a desktop decline and a mobile increase over the same stretch. A blended device total can therefore conceal a divergence that matters to a site's diagnosis.
An AI Overview may appear with a local pack, product grids, People Also Ask and video carousels rather than above an otherwise simple set of organic links. That feature stacking changes the context in which the user encounters the organic result. It also limits what an intent label alone can explain about an individual keyword's performance. Transactional intent may leave the user with a need to act even as the organic listing gets a smaller share of attention.
How much a page is exposed therefore depends on query intent, which devices its audience uses and how many results-page features appear alongside it. A useful diagnosis distinguishes a change in the query mix from a change in the results page presented to the same audience. Without that split, an overall click-through rate shows what happened but leaves the cause unclear.
Why being cited inside an AI Overview creates a separate traffic and visibility layer from ranking
Citation creates a second route to visibility within the feature that removes the need for many organic visits. A page can supply the synthesized answer without occupying a leading organic position, just as a leading organic page can be absent from the answer's citations. BrightEdge's research found that only a minority of cited sources ranked in the organic top results. The weakening overlap between citation and ranking means that one cannot reliably stand in for the other.
Citation presence correlates with click-through rates that hold or grow, while absence correlates with decline. That relationship creates a separate opportunity for informational pages exposed to organic click loss. A cited page may win back traffic through the citation even as its ordinary listing gets fewer clicks. An uncited page faces both the substitution of its potential visit and exclusion from the answer's source visibility.
This structural distinction leaves open whether citation offsets every loss. It also does not establish that ranking has ceased to matter in every search context. A rank-only report cannot distinguish a page represented inside the answer from one that appears solely beneath it.
What the early-2026 partial rebound and Google's counter-claims actually show
The early-2026 partial rebound in click-through rates qualifies the decline narrative without reversing it. The rebound was real but modest, lifting the floor rather than restoring performance to pre-AI Overview levels. It does not remove the intent-based split or show that informational pages have regained their former click opportunity. A recovery from a depressed level is different from recovery to the earlier baseline.
Google's traffic-growth claim should also be kept separate from the question of how click loss differs by intent. The claim is not independently verifiable from the evidence provided here, so it cannot settle that question. An aggregate growth claim does not by itself establish what happens to clicks on informational queries when an AI Overview appears. Google's further argument that remaining clicks are higher quality is more directly testable. Agarwal and Sen found no meaningful difference in downstream engagement between users who saw AI Overviews and users who did not, including bounce, back-button rate and time on page.
Those findings do not back the claim that AI Overviews filter out only low-quality visits. The randomized design gives the experiment stronger causal weight than a before-and-after traffic comparison. Its results also align directionally with the observational evidence of organic click pressure. The calibrated conclusion is neither uninterrupted deterioration nor a return to the previous search environment. There has been a partial rebound, but the strongest causal evidence still shows an intent-specific loss of organic visits.
How to map your own query portfolio against the intent-based risk and citation-opportunity framework
Begin by classifying owned pages according to the dominant query intent they serve, including informational, navigational, transactional and local intent. This establishes the expected exposure before individual ranking movements distract from the larger pattern. Next, find which target queries now trigger AI Overviews and which owned pages appear beneath them. Tools that track SERP features by keyword provide the required view of the results page. Longer, multi-word queries offer a useful triage signal, while Pew Research Center's noun-verb combination finding provides another heuristic for identifying potential triggers.
Then split the exposed group into citation-present pages and citation-absent pages. The audience's device profile and the other features present on the target results pages provide additional signals. Those signals help separate the exposure that comes with informational intent from the added pressure of device mix and feature stacking. The aim is to locate the intersection that explains performance, rather than assign every decline to AI Overview presence alone.
Measurement should include AI citation share, branded query volume and conversion rate by traffic source, as well as pageview totals. Session volume alone can misread an outcome when a page loses raw clicks while gaining qualified visitors through citations. That possibility should be measured rather than assumed, especially given the experimental evidence against a general claim of better remaining clicks. Platforms that track SERP features, intent classification and citation presence at scale supply the capabilities this workflow requires. The practical shift is from monitoring rank alone to diagnosing which searches still require a visit and whether the site is visible on the path the user takes.
