Do AI Overviews Actually Reduce Clicks to Websites?
Pew Research tracked real browsing data from 900 U.S. adults. The click-through numbers it found are the most concrete evidence yet of what an AI summary does to a search.
Pew Research Center tracked real browsing data from 900 U.S. adults across 68,879 Google searches in March 2025, and found that people clicked a traditional search result in just 8% of searches that displayed an AI-generated summary, compared to 15% of searches that didn't. Only 1% of all visits involved clicking a link inside the summary itself. This is actual measured behavior, not a survey asking people what they think they'd do, which makes it some of the most concrete evidence available on what an AI Overview does to click-through behavior.
What Pew Research Actually Measured
The study drew on Pew's KnowledgePanel Digital panel: 900 U.S. adults who agreed to share real browsing data, tracked from March 1 to March 31, 2025. Researchers then reconstructed which of those 68,879 unique Google searches had produced an AI-generated summary by re-scraping the same queries between April 7 and April 17, 2025. Of those searches, 12,593, about 18% of the total, returned an AI summary.
This design matters because it's observational, not experimental: Pew didn't force some users to see summaries and others not to. It measured what actually happened across a real, if limited, sample of real search sessions, which is a meaningfully different kind of evidence than a vendor's self-reported numbers or an SEO tool's estimate.
The Click-Through Numbers
The core comparison, straight from Pew's own published figures:
| Metric | Searches with an AI summary | Searches without one |
|---|---|---|
| Clicked a traditional search result | 8% of visits | 15% of visits |
| Clicked a link inside the AI summary | 1% of visits | Not applicable |
| Ended the browsing session on that page | 26% of visits | 16% of visits |
Two things stand out. First, the traditional-result click rate roughly halves when an AI summary appears (8% versus 15%), and clicking a link placed inside the summary itself is rarer still. Second, people were also more likely to simply stop browsing altogether after landing on a page with an AI summary (26%) than after one without (16%). Pew also found the median summary ran 67 words, and 88% of the summaries it examined cited three or more sources, meaning most summaries genuinely draw on multiple pages even though few of those pages get a click as a result.
Not Every Search Triggers a Summary, and the Pattern Isn't Random
Only 18% of searches in Pew's sample produced an AI summary at all, and which searches did wasn't evenly distributed. Question-based searches produced a summary 60% of the time, searches of 10 or more words did so 53% of the time, and short one-to-two word searches did so only 8% of the time. AI summaries concentrate on longer, more specific, more question-shaped queries, exactly the kind of query a definitional or explanatory page is written to answer.
Which Sources Actually Get Cited in the Summaries
Pew's scrape also captured what the AI summaries actually linked to, not just whether one appeared. The typical summary ran a median of 67 words, though the full range stretched from as short as 7 words to as long as 369. Wikipedia, YouTube, and Reddit together accounted for 15% of the links cited inside AI summaries, more than any other single cluster of sources in Pew's dataset. Government sites, domains ending in .gov, showed up inside AI summaries at roughly three times the rate they showed up among the ten traditional blue-link results on the same page, 6% versus 2%. And search queries phrased as complete sentences, not just direct questions, triggered a summary 36% of the time, landing between the 60% rate for question-format searches and the 8% rate for short keyword searches.
Read together, this points the same direction as the click-through numbers above: an AI summary leans on a narrower, more established set of sources than the average search result page does, which raises the bar for what a page needs to be, in size, authority, and clarity, to be one of the sources it draws from.
Is This the Same Thing as Gartner's "25% Search Volume Drop" Prediction?
No, and conflating the two overstates what either one actually shows. Gartner published a forecast in February 2024 predicting that traditional search engine volume would fall 25% by 2026 as generative AI tools substitute for search queries entirely, a prediction from Gartner's own modeling about the total volume of search activity shifting to chatbots and other AI agents. Pew's study measures something narrower and already observed: click behavior on searches that already show an AI summary, from actual 2025 browsing data. One is a forward-looking industry forecast; the other is a completed measurement of a specific behavior. As of this writing, no public dataset has independently confirmed or disproven Gartner's specific 25% figure with the same kind of transparent, sourced methodology Pew used, so that particular number is worth treating as an industry forecast, not a settled fact, however often it gets repeated as one.
What This Actually Means for a Site
The realistic takeaway isn't that ranking or being crawlable stopped mattering. It's that for the specific slice of queries most likely to trigger a summary (question-based, longer, definitional), the value of ranking well has partly shifted from "the click" to "being one of the sources the summary draws from and names." Since 88% of the summaries in Pew's data cited three or more sources, being cited, not just ranked, is the more attainable and more meaningful goal for that content type now. That's the same distinction covered in how to show up in AI search results: a page can rank normally in Google and still never get pulled into an AI-generated answer, and conversely, being one of several cited sources doesn't guarantee a click even when it happens. Neither ranking nor citation is a reliable substitute for the other; they're separate outcomes worth tracking separately. For the underlying access requirements that determine whether a page can even be considered for that source list in the first place, see AEO, GEO, and SEO: what each term actually means.
What a Site Owner Can Actually Measure Now
Until mid-2026, there was no first-party way to see how a specific page performed inside Google's AI Overviews at all. A site's AI-driven visibility was effectively invisible in Search Console, unless a site owner inferred it indirectly from a traffic dip. Google closed part of that gap on June 3, 2026, introducing dedicated Generative AI performance reports inside Search Console. This is Google's own account of what the report shows, not third-party inference, so it's worth being precise about what it actually adds and where it still stops short.
Two separate reports sit under Performance in the left-hand navigation: one covering AI Overviews and AI Mode inside Search itself, and a second, independent report for generative AI features inside Google Discover. Both report impressions, meaning how many times a link to your site was shown inside a generative AI feature, broken down by individual page, by country, by device (desktop, tablet, or mobile), and by date on a daily, weekly, or monthly basis in Pacific time. Data only goes back to May 18, 2026, and Search Labs experiments are excluded from both reports. Like Search Console's standard Performance report, each is capped at 1,000 rows, and both include an export button for the chart and the underlying table.
What the report deliberately doesn't show is the part most site owners actually want: no clicks, no click-through rate, no average position, and no query-level data showing which search triggered a given impression. That gap lines up with what Pew's browsing data already suggests. An impression inside an AI summary and a click on your site are two loosely connected events, and right now Search Console can only confirm the first one happened. A page can show a clear rise in the Generative AI report and still see no meaningful change in actual sessions, consistent with Pew's finding that only 8% of AI-summary searches produced any traditional click at all. Until Google adds click data to this report, the more reliable way to gauge whether AI visibility is translating into visits is comparing overall Search performance and content-level analytics before and after a page starts appearing in it, rather than treating the impression count itself as a proxy for traffic. For the diagnostic tools available once a page is confirmed indexed, see indexed but not ranking in Search Console.
In practice, three checks make the report worth opening on a regular basis: sort by page to see which URLs are actually surfacing inside AI Overviews at all, since many pages never will; watch whether impressions on a given page climb after a content update, since that's the closest signal available right now to "did this edit help get cited"; and cross-reference the pages showing up here against server logs or an analytics tool's referrer data for any AI-attributed sessions your stack can already detect, rather than waiting on Search Console to close the click-data gap on its own timeline.
Related: how to show up in AI search results, is AI search optimization actually different from SEO, AEO, GEO, and SEO: what each term actually means, indexed but not ranking in Search Console.
Frequently asked questions
What exactly did Pew Research measure?
Real browsing behavior, not a survey. Pew tracked 900 U.S. adults in its KnowledgePanel Digital panel through actual Google searches conducted in March 2025, then checked which of those search result pages included an AI-generated summary. It measured what people actually clicked, not what they said they would click.
Does this study prove AI Overviews are killing website traffic industry-wide?
It proves a specific, narrower thing well: within this sample, a click on a traditional result was less likely on a search that showed an AI summary than one that didn't. It doesn't measure total industry-wide traffic change, doesn't cover other AI products like ChatGPT or Perplexity, and one panel from one month isn't the same claim as a permanent, universal trend. Treat the specific numbers as evidence about this dataset, not a settled law of search.
Which kinds of searches are most likely to trigger an AI summary?
Per Pew's own breakdown, question-based searches (60%) and longer searches of 10 or more words (53%) were far more likely to produce an AI summary than short one-to-two word searches (8%). Definitional and explanatory content, the kind that answers a specific question in detail, is exactly the content most exposed to this effect.
Can Search Console tell me if AI Overviews are sending me clicks?
Not yet. Search Console's Generative AI performance report, live since June 2026, shows impressions inside AI Overviews and AI Mode broken down by page, country, device, and date, but it has no click, click-through-rate, or query data. It can confirm your page was shown inside an AI summary; it can't confirm anyone clicked through from it.
Part of the AI Search cluster.
Want this applied to your own site, not just read about it?
This is the free version, evidence-labeled and yours to read at no cost. Applying it to your own site (technical SEO, AI search visibility, and GEO in one pass) is separate, paid work at kuraib.site.