Do AI summaries on Google reduce clicks to actual websites?

Pew Research Center's analysis of actual browsing behavior — not self-reported survey answers — finds that when a Google search page shows an AI-generated summary, users click through to any link far less often than when a search produces no AI summary. Drawing on tracked browsing data from 900 U.S. adults in the KnowledgePanel Digital panel across March 2025 (68,879 unique Google searches, of which 12,593 produced an AI summary), Pew reports that "users who encountered an AI summary clicked on a traditional search result link in 8% of all visits," while those who did not see a summary "clicked on a search result nearly twice as often (15% of visits)." Clicking the AI summary's own cited links was rarer still, happening in "just 1% of all visits to pages with such a summary." Users were also more likely to end the browsing session outright after an AI-summary page (26%) than after a plain results page (16%).

Pew's own framing ties this directly to the publisher-traffic complaints noted at the top of the report: "online publishers recently have attributed declining web traffic to these summaries replacing traditional search results, claiming that many users are relying on the summaries instead of following links." The data on when summaries appear supports that mechanism — Pew finds AI summaries are triggered disproportionately by exactly the searches most likely to be informational rather than navigational: "Just 8% of one- or two-word searches resulted in an AI summary. But that share rose to 53% for searches with 10 words or more," and "60% of search queries that began with question words... resulted in an AI summary." The summary tends to appear precisely where a user is asking a question outright, and that same user's next action is overwhelmingly not to click anything — two-thirds of all searches in the study, summary or not, end with no result-link click at all.

This sits alongside the vault's existing measurements of the content-supply side of the same shift: How much of the internet is AI-generated now? and Has AI-generated content stopped growing on the web? both track how much of the web's new content is itself AI-produced. Pew's finding is the matching demand-side measurement: even where content keeps being produced, less of it is being visited, because an AI layer increasingly intercepts the query before a link is ever followed. Supply growing and demand for the underlying links shrinking are two halves of the same web-traffic story, measured independently here.

The design is observational, not experimental: Pew compares click rates on summary pages against non-summary pages but does not, and in this excerpt cannot, rule out that the queries which trigger a summary are intrinsically different from those that don't — longer, question-shaped, informational queries may have drawn fewer clicks even before AI Overviews existed, since someone asking "what is X" may never have clicked through at the same rate as someone searching a brand name. The excerpt also scopes to Google only ("due to technical limitations in our ability to identify AI-generated summaries on other search engines, this analysis includes only Google searches") and to one U.S. panel of 900 adults in one month, so it does not establish the scale of publisher revenue loss or whether the pattern holds on Bing, other AI search products, or outside the U.S. What it does establish, at the strength a large tracked-behavior sample allows, is that the publisher complaint — that AI summaries substitute for clicking rather than just sitting alongside it — survives a look at measured behavior rather than resting on anecdote alone.