AI summaries cut clicks to the rest of the web on the searches where they appear, according to the independent studies with the strongest designs, and the cut is large. Pew Research Center’s analysis of real browsing data found that people clicked a traditional result on 8% of visits to Google results pages with an AI summary, against 15% on pages without one. A randomized field experiment by researchers at the University of Pennsylvania and Northeastern found that hiding AI Overviews raised click-through. The citations inside AI answers are unreliable as well, which matters to a brand as much as the traffic does.

I lead Innovation and AI at Tandem Interactive, an agency whose clients depend on search, and many of the widely shared numbers on this subject come from companies that sell SEO software. This essay sticks to sources without a product attached: research institutions and peer-reviewed studies, with university preprints marked as preprints and Google’s own statements labeled as Google’s claims.

Do AI Overviews reduce clicks?

The studies with the strongest designs agree that clicks fall, and they reach that conclusion from different directions. Pew tracked the browsing of 900 U.S. adults through March 2025 and found that on visits to Google results pages carrying an AI summary, people clicked a traditional result 8% of the time, against 15% on pages without one. They clicked a link inside the summary on 1% of those visits, and they ended their browsing session on 26% of pages with an AI summary compared with 16% of pages with only traditional results.

  • No AI summary clicked a result 15%
  • With an AI summary clicked a result 8%
  • Inside the summary clicked its link 1%
Share of visits to Google results pages where the user clicked a traditional search result, with and without an AI summary on the page, and the share of visits where they clicked a link inside the summary itself. Browsing data from 900 U.S. adults in March 2025, from Pew Research Center.

Pew’s data is observational, so it shows what happens on pages with summaries without proving that the summaries caused it. The searches that trigger summaries also differ from the ones that do not, and 53% of searches with ten or more words produced one. The causal test arrived in August 2026, when Wang and colleagues published a preregistered randomized experiment in which 1,100 Google users ran a browser extension that either hid AI Overviews or routed their searches into AI Mode.

Hiding AI Overviews raised click-through by 8.8 percentage points once the estimate was adjusted for how often the hiding worked. A change Google made to its page code partway through the study meant only about half of the Overviews were hidden. Assignment to AI Mode cut click-through by 18.8 points and reduced the share of users clicking through to news sites and to Reddit. The paper is still a preprint.

Chatbot users are still a minority of the people searching, though their numbers are growing quickly. The Reuters Institute’s 2026 Digital News Report, which covers 48 markets, found weekly use of AI chatbots for news rose from 7% to 10% in a year, and Pew’s 2026 survey found 42% of U.S. adults say they use AI chatbots to search for information.

Click-through from those tools is where the gap opens. Across the 27 markets the Reuters Institute surveyed, 4% of all respondents said they always or often click through to news sources from AI, compared with 19% from search and 17% from social media. In a separate Reuters Institute survey across six countries in 2025, about a third of the people who had seen AI-generated answers in search said they always or often click the links in them.

Are AI search citations accurate?

AI search citations go wrong often enough that every independent audit I found flagged them. Columbia’s Tow Center ran sixteen hundred queries through eight AI search tools, asking each to identify the original source of a news excerpt, and the tools answered more than 60% of queries incorrectly, from 37% for Perplexity to 94% for Grok 3.

The European Broadcasting Union and the BBC ran a larger audit with 22 public service media organizations across 18 countries, and journalists found at least one significant issue in 45% of AI assistant answers about the news, with sourcing the biggest single cause at 31% of responses.

  • Gemini 76%
  • Copilot 37%
  • ChatGPT 36%
  • Perplexity 30%
Share of AI assistant answers about the news that journalists rated as having at least one significant issue, by assistant, from the free versions tested in spring 2025. From the European Broadcasting Union and BBC News Integrity in AI Assistants study.

Peer-reviewed work found the pattern earlier. Liu, Zhang, and Liang at Stanford reported at EMNLP 2023 that only 51.5% of sentences from generative search engines were fully supported by their citations, and only 74.5% of citations supported the sentence they were attached to. They also found that citation precision fell as perceived usefulness rose, with a correlation of −0.96, so the answers that seemed most helpful were often the ones with inaccurate citations.

Google’s own AI Overviews were audited claim by claim this year in a preprint from Washington University in St. Louis, which checked 98,020 claims against the pages each Overview cited and found 11.0% unsupported, with 4.1% contradicted by the cited content and 7.0% not addressed in it. The same team found that 29.8% of the domains Overviews cited did not appear on the first page of results for the query at all.

What does Google say about AI Overviews and clicks?

Google’s public position is that traffic is holding up. In August 2025, Liz Reid, who leads Google Search, wrote that total organic click volume from Google Search to websites had been relatively stable year over year and that Google was sending slightly more quality clicks, meaning clicks where users do not quickly return to the results.

Google has not published the data behind either statement. The independent studies measured what happens on the searches where AI features appear, so both accounts can only hold if those losses are made up somewhere else. The same post points to one way that happens when it says user trends are shifting traffic to different sites, decreasing it for some and increasing it for others. That is little comfort to a site on the losing side of the shift.

What are regulators doing about AI Overviews?

The UK’s Competition and Markets Authority designated Google with strategic market status in general search in October 2025 and confirmed that AI Overviews and AI Mode are in scope of that designation. In June 2026 it announced that publishers will get effective tools to prevent their content being used to power AI features in search such as AI Overviews, a measure the regulator itself describes as a world first.

What should an SEO program change?

In rough order of how much they matter:

  1. Report traffic from queries that trigger AI features separately, since clicks on traditional results roughly halve on pages where a summary appears.
  2. Judge informational pages by what they contribute beyond the visit, because fewer of the people they help will ever click through.
  3. Track whether AI answers cite your pages and read what they claim when they do, using repeat sampling so noise does not pass for movement.
  4. Keep the pages engines cite precise and current, because an answer that misstates your page still carries your name.
  5. Treat any single click-through figure as a snapshot, including the ones in this essay, because the interfaces changed while the studies ran.
  6. Revisit opt-out settings as regulator-backed controls arrive, starting with the UK requirement.

What hasn’t been settled?

No peer-reviewed study I could locate measures the causal effect of AI Overviews across a broad set of sites, and the preprints disagree about size. The University of Washington study of English Wikipedia is the cautionary example. Its first version in February 2026 estimated that AI Overviews cut daily traffic to English articles by about 15%, and its sixth version in September put the decline in search traffic near 5% after the authors revised their method. Any number in this area needs its version date checked before it is repeated, including the numbers above.

FAQ

Do AI Overviews reduce organic traffic?

They reduce clicks on the searches where they appear, according to every independent study with a strong design that I found. Pew measured clicks on traditional results at 8% of visits to pages with an AI summary against 15% without one, and a 2026 randomized experiment found hiding AI Overviews raised click-through.

Rarely in browsing data, and more often in what people say about themselves. Pew’s browsing data recorded a click on a link inside the summary on 1% of visits to pages with one, while in a 2025 Reuters Institute survey about a third of people who had seen AI answers said they always or often click those links.

Are AI Overview citations accurate?

Most AI Overview claims match their cited pages, and roughly one in nine does not. A 2026 preprint from Washington University in St. Louis checked 98,020 claims against their cited pages and found 11.0% unsupported, including 4.1% contradicted by the cited page, and a broader audit of AI assistants found significant sourcing problems in 31% of answers.

Does Google say AI Overviews hurt traffic?

Google says the opposite, and its head of Search wrote in August 2025 that total organic click volume to websites was relatively stable year over year and that quality clicks had slightly increased. Google did not publish the data behind either claim, so neither one can be checked from outside the company.

Separate the queries that trigger AI features from the rest, and add presence and citation tracking beside clicks, because fewer of the people your pages help will arrive by clicking. My measurement essay covers how to track presence without mistaking noise for movement.

Sources