How do AI Overviews change clicks from search results?

What did Pew find when an AI summary appeared?

Pew measured fewer traditional-result clicks and more ended browsing sessions on pages with an AI summary. The results below use Pew's tracked-visit denominators.¹

Pew counted an overlapping URL as an AI-summary click when it appeared both in the traditional results and in the summary. That rule keeps the two click categories tied to the links available on the page rather than to an unobserved intention.¹

How common were AI summaries in Pew's Google sample?

AI summaries appeared in 18% of the 68,879 unique Google searches in Pew's click dataset. 12,593 searches produced an AI summary alongside traditional results.¹

The underlying Pew report also found that 58% of respondents conducted at least one search using a search engine like Google that produced an AI-generated summary alongside traditional results. That is a respondent-level exposure measure, not the share of all searches that showed a summary.³

How did Pew collect the click data?

Pew purchased March 2025 browsing data from 900 U.S. adults in KnowledgePanel Digital. Respondents agreed to install an app that tracked their online browsing behavior. Pew received URLs visited in a web browser on a tracked device from March 1 to 31, 2025, covering approximately 2.5 million visits to 1.1 million unique URLs.³

For every collected google.com search, Pew used a third-party web-scraping service to rerun the search and collect the first results page. The collection included traditional result links, any AI-summary text, and up to three source URLs in the summary. Pew collected those result pages from April 7 to 17, 2025.¹

The click result therefore joins two records: the tracked person's next visited URL and the links Pew collected from the corresponding results page. This is why the sample, device, date window and click definition belong beside every percentage.

What do these click statistics mean for a business website?

The study shows lower traditional-result click activity on Google visits where an AI summary appeared in this sample. It does not give a website its expected click-through rate. A business owner reading the result should keep the query type, date, device and click definition with any comparison.

The result is useful as context for the wider shift from classic results to generative search surfaces. For the broader collection of measured findings, see the research hub. For the Google Maps context around local discovery, see Google Maps.

What do these numbers not tell you?

• They do not establish causation. Pew observed browsing behavior and compared visits with and without an AI summary. The study did not randomly assign people to results pages.
• They do not describe local searches specifically. The 68,879-search dataset was not limited to local intent, a business category, a city or a Google Maps results page.¹
• They do not cover every device or app. Pew's tracked browsing data may omit work or other untracked devices, activity in apps outside web browsers, real-time algorithmically curated content and pages that require a login.³
• They do not show that a cited source caused a visit. The 1% figure counts a next URL that matched one of the first three URLs cited in a summary. It does not measure whether a reader trusted the summary or why the reader opened the page.
• They do not guarantee that an AI summary is correct. Google's Search help page warns that AI responses may include mistakes.²

FAQ

Can I use Pew's click rates as my site's expected click-through rate?

No. Pew's percentages describe next-page activity in a tracked browsing panel, not the future performance of one website, query set or local market.

Can an AI Overview contain mistakes?

Yes. Google warns that AI responses may include mistakes. For important information, Google says to check it in more than one place, including the supporting links and other Google Search results.²