Do AI answers in search mean fewer clicks to websites?

What Pew measured, and how

Pew analysed the March 2025 browsing of 900 U.S. adults in KnowledgePanel Digital, a panel whose members agreed to install an app that tracks their browsing. The tracking covered about 2.5 million visits to 1.1 million unique URLs, according to Pew's May 2025 report. That report counted 2,457,176 visits in all.

For each respondent, researchers examined every search engine visit during March 1 to 31, the search terms used, and the next URL visited afterward. Between April 7 and 17, 2025, they re-ran the searches on google.com through a third-party scraping service. The service captured the first page of results, any AI summary text, and up to three source URLs cited in the summary.

The dataset held 68,879 unique Google searches. Of those, 12,593 produced an AI summary. Pew studied Google only, because of technical limitations in identifying AI summaries on other search engines.

Clicks were matched by URL: the next page a person visited was compared with the links on the results page and with the first three sources cited in the summary. A link that appeared in both places was counted as a summary click.

The numbers: 8% against 15%

When a results page carried an AI summary, users clicked a traditional result link in 8% of visits. When it did not, they clicked in 15% of visits. Pew describes the second figure as nearly twice the first.

The arithmetic is simple. A drop from 15% to 8% is a reduction of seven percentage points, or about 47% in relative terms. For every 100 visits to a page without a summary, about 15 produced a click on a result. For every 100 with a summary, about 8 did.

The links inside the summary did not make up the difference. Users clicked one in about 1% of visits to pages that had a summary.

Pew sorted each visit into one of four outcomes: clicking a link, searching Google again, leaving for another site, or ending the session. A session counted as ended when the user left the browser for five seconds or more.

On that measure, 26% of visits to pages with a summary ended the browsing session, against 16% for pages without one. Regardless of whether a summary appeared, about two-thirds of searches ended with the user searching Google again or leaving for another site without clicking a result link.

Which searches get a summary

Pew found that 18% of the searches in its sample produced an AI summary, roughly one in five. Across people, 58% of the 900 respondents ran at least one Google search in March 2025 that returned one. The two figures use different bases: one counts searches, the other counts people.

The share varied sharply by query. Only 8% of searches of one or two words produced a summary, while 53% of searches of ten or more words did. Searches that began with a question word such as who, what, when or why produced one 60% of the time. Searches containing both a noun and a verb produced one 36% of the time.

In this data, longer searches, questions and full sentences tended to produce a summary more often than short keyword searches.

Who gets cited

Wikipedia, YouTube and Reddit were the three most-linked sites in both the summaries and the standard results. Together they made up 15% of summary sources and 17% of standard-result sources. Wikipedia appeared somewhat more often in summaries, and YouTube somewhat more often in standard results.

Government sites were cited more often in summaries: .gov domains made up 6% of summary sources against 2% of standard results. News sites accounted for 5% in each.

Most summaries named several sources. Pew found that 88% cited three or more and 1% cited one. The median summary ran to 67 words, in a range from 7 to 369.

A drop from 15% to 8% is a reduction of seven percentage points, or about 47% in relative terms.

Where Google disagrees

On August 6, 2025, Liz Reid, VP, Head of Google Search, published a post titled "AI in Search is driving more queries and higher quality clicks." It said total organic click volume from Google Search to websites has been relatively stable year-over-year. It also said average click quality had risen, and that Google was sending slightly more quality clicks than a year earlier.

Google defines a quality click as one where the user does not quickly return to the results page. The post attributes reports of steep traffic declines to flawed methods, isolated examples, or changes that predate AI features in Search.

The post also concedes that some queries may not produce clicks. Its example is a quick-answer search such as when the next full moon is, and it says the same holds for the Knowledge Graph and sports scores. For many other types of questions, it says people still click through to dig deeper, explore further or make a purchase, and that these clicks are more valuable.

Google adds that traffic is shifting between sites, with some losing and others gaining. It names forums, videos, podcasts, and first-hand or in-depth content as gaining. The post also says AI responses carry prominent links and visible citations, and that websites have control over how content is highlighted or included in Search, through open web protocols.

The two sets of claims are not directly comparable. Pew compared click rates on results pages with and without a summary, in one month, for a panel of U.S. adults. Google described total organic click volume from Google Search to websites. The Google post gave no percentages for clicks, click quality or query growth.

What the evidence does not show

Pew's data does not measure traffic to any one website, and it does not say what happened to revenue. It also reflects how pages looked in April 2025. Pew notes that AI summaries can change over time.

The browsing data has further limits. Respondents had to install a VPN to register their devices, so work and other untracked devices may be missed. Mobile apps, algorithmically curated feeds and pages that require a login were not captured.

Pew's May 2025 report covered visits and exposure to AI summaries but contained no click-through figures, so the click findings come from the later July 2025 analysis, which used the same data.

For an operator, the useful step is to check this against their own numbers. A site can compare click-through rates on queries that trigger a summary with those that do not. The Pew figures give a benchmark for that comparison, not a forecast for any particular site.