What did Pew actually measure?
On 22 July 2025 the Pew Research Center published a short read by Athena Chapekis and Anna Lieb under the title “Google users are less likely to click on links when an AI summary appears in the results”. It is not a survey. Nobody was asked what they thought they did. It is a behavioural study built from the browsing records of 900 US adults who are members of KnowledgePanel Digital, an online panel whose respondents agree to install an app that tracks the URLs they visit.
Pew received every URL each respondent visited on a tracked device between 1 and 31 March 2025, which gave it the searches those people actually ran and what they did next. The dataset contains 68,879 unique Google searches; 12,593 of them — 18% — produced an AI summary. Fifty-eight per cent of respondents met at least one.
To know which searches carried a summary, and what it linked to, Pew re-ran each search through a third-party scraping service and captured the first page of results, between 7 and 17 April 2025. That gap is the study's most important limitation, and Pew states it rather than hiding it: “The AI summary for any given search may change over time, and the results of this analysis reflect how search results appeared at the time they were collected.” A search run on 3 March was classified using a page scraped five weeks later. Some searches counted as carrying a summary may not have carried one when the person saw them, and some counted as clean may have.
Two further boundaries. Only Google is included, because Pew says it could not reliably identify AI summaries on other engines. And the unit is a browser visit to google.com by a US adult on a tracked device — not a query inside Google's app, and not a commercial query isolated from the rest. One thing still separates this study from almost every other number in the argument: Pew sells nothing.
How far do clicks fall?
The headline figure is a halving. On visits to a Google results page that carried an AI summary, users clicked a standard search result 8% of the time. On visits to a page without one, they clicked 15% of the time — nearly twice as often.
The figure that matters more for anyone hoping to be cited is the second one. Users clicked a link inside the AI summary on 1% of visits to pages carrying a summary. Not 1% of clicks: 1% of visits. The links a summary displays are, on this evidence, close to decorative as a traffic mechanism.
Behaviour after the search shifted too. Sessions ended on the results page on 26% of pages with an AI summary against 16% of pages without. Around two-thirds of all searches ended with the user either browsing elsewhere on Google or leaving without clicking anything.
% of visits to Google results pages, US adults, March 2025
Source: Pew Research Center, 22 July 2025, from the browsing records of 900 US adults on KnowledgePanel Digital, 68,879 Google searches collected 1–31 March 2025.
Bars are scaled to the largest value shown. Search results were re-scraped 7–17 April 2025 to identify which searches carried a summary, so classification lags the behaviour by up to six weeks.
Two details in the same report cut against the loudest version of the publisher complaint. The three most-linked sources in AI summaries are Wikipedia, YouTube and Reddit, and together they accounted for 15% of the sources listed in the summaries Pew examined — against 17% of the sources in standard results. And news websites accounted for 5% of the links in AI summaries and 5% of the links in standard results. On this dataset, the summary is not shifting the mix away from news. It is reducing the number of clicks that any mix produces.
Where the mix does move is government: 6% of links in AI summaries pointed at .gov sites against 2% of standard results. That is consistent with what independent evaluations find about which sources get reached for — reference and official material outranks reporting, a pattern we traced in which sources AI assistants actually cite.
Whether you meet a summary at all depends heavily on how people phrase the search. Just 8% of one- or two-word searches produced one, rising to 53% for searches of ten words or more. Sixty per cent of searches beginning with a question word produced a summary, against 36% of searches containing both a noun and a verb. The typical summary ran to 67 words; the shortest was seven and the longest 369. Eighty-eight per cent cited three or more sources and only 1% cited a single source.
What do the SEO datasets find?
The commercial datasets are larger, more current and less disinterested. They are also the ones most often quoted without their qualifications.
Ahrefs: 34.5%, and why it is not 64%
On 17 April 2025 Ryan Law at Ahrefs published an analysis of 300,000 keywords — 150,000 that triggered an AI Overview by March 2025 and 150,000 informational keywords that did not — comparing position-one clickthrough rate in Search Console data for March 2024 against March 2025.
The raw numbers look catastrophic: position-one CTR on the AI Overview keywords fell from 0.073 to 0.026, about 64%. Ahrefs does not claim 64%, and the reason is the interesting part. Clickthrough rates fell on the control keywords too. Against a forecast CTR of 0.040 for what those keywords would have done without an AI Overview, the AI-attributable difference is 34.5% — “a 34.5% lower average clickthrough rate (CTR) for the top-ranking page”. It is the only study here that builds an explicit counterfactual, and most citations of it drop the counterfactual while keeping the number.
Similarweb and Semrush: news queries, not all queries
Digiday's Sara Guaglione assembled several vendor datasets on 10 July 2025. From Similarweb: searches for news that end without a click to a news site rose from 56% in May 2024, when AI Overviews launched in the US, to nearly 69% by May 2025. General search referral traffic to a set of 1,000 web domains fell from 12 billion global visits in June 2024 to 11.2 billion in June 2025, about 6.7% year on year. US organic search referrals fell from over 2.3 billion visits in July 2024 to 1.8 billion in June 2025.
The 56-to-69 figure is the one most often recycled as a statement about search in general. It is not. It is news queries, measured against news sites. Similarweb's own senior insights manager, David Carr, told Digiday that “a one-month change may not necessarily signify that much”.
From Semrush, shared with Digiday: the share of US desktop queries showing an AI Overview rose from 6.49% in January 2025 to 20.22% in April, then edged down to 19.88% in May. SEOClarity, tracking over 100 million US keywords, put it at about 19% in June 2025. These are prevalence figures, not click figures, and they are the ones most likely to be stale by the time you read them.
SparkToro: the longest series
On 9 June 2026 Rand Fishkin published an analysis using Similarweb's desktop and mobile clickstream panel covering January to April 2026 in the United States. It found that 68.01% of Google searches ended without a click, against 60.45% in 2024, 49% in 2019 and roughly 45% in 2016 — a 12.5-point rise in two years, which Fishkin describes as the fastest move in a decade. The share of searches producing at least one click to the open web fell 9.51 points, a 22.9% decline. Only 0.34% of searches reached Google's AI Mode in that period.
Fishkin is explicit that the long series stitches together three providers — Jumpshot, then Datos, then Similarweb — and that “these aren't the same users or devices”. The clickstream also excludes Google's native app, where zero-click behaviour is more aggressive, which would push the real figure higher.
Where do the studies disagree?
They do not disagree about direction. Every dataset here points the same way: a search showing an AI summary produces fewer clicks than one that does not. The disagreements are about four other things, each of which matters more than the headline.
The unit. Pew counts visits by people, Ahrefs counts clickthrough rate per keyword at position one, Similarweb counts visits arriving at domains. A fall in clicks per search is not a fall in total clicks, because the number of searches is not fixed. This is the gap Google drives its rebuttal through, and it is a real one.
The population. Pew covers everything 900 US adults searched for. The 69% figure covers news queries hitting news sites. Ahrefs covers informational keywords at position one only. None of them is measuring the query mix of a B2B software company, and none is measuring yours.
The counterfactual. Only Ahrefs builds one. Pew compares searches with summaries against searches without, which compares different kinds of query rather than the same query under two regimes: long, question-shaped searches are both more likely to trigger a summary and plausibly less likely to end in a click anyway. Pew does not claim causation, and neither should anyone quoting it.
The attribution. Google changed a great deal between March 2024 and March 2026 besides adding AI Overviews: core updates, layout changes, more paid units, AI Mode. No study here isolates the summary from the rest. When one says clicks fell, it is reporting a correlation with a period, not the effect of a feature.
The honest summary is narrow. On the queries that trigger them, AI summaries are associated with roughly half the click rate; the link inside the summary is clicked about 1% of the time; and the aggregate effect on any one site is established by none of this. For the mechanism rather than the effect, see how ranking in Google differs from being cited by an AI system.
What does Google say about all this?
Google disputes the research and publishes nothing checkable in its place.
On the Pew study, a Google spokesperson told The Register, in a statement published by Thomas Claburn on 22 July 2025: “People are gravitating to AI-powered experiences, and AI features in Search enable people to ask even more questions, creating new opportunities for people to connect with websites. This study uses a flawed methodology and skewed queryset that is not representative of Search traffic.”
On 6 August 2025, Liz Reid, the head of Google Search, published a post making two claims: “Overall, total organic click volume from Google Search to websites has been relatively stable year-over-year”, and that “average click quality has increased and we're actually sending slightly more quality clicks to websites than a year ago”. A quality click is defined as one where users do not quickly click back. The same post dismisses “third-party reports that inaccurately suggest dramatic declines in aggregate traffic — often based on flawed methodologies, isolated examples, or traffic changes that occurred prior to the roll out of AI features in Search”.
Three observations. Google publishes no dataset, chart or methodology for either claim, so neither is falsifiable from outside. Both claims are about aggregate volume across all of Search, which is not what any of the studies measured: total clicks can hold steady while the click rate on summary-bearing queries halves, if query volume rises enough — both sides can be telling the truth about different quantities. And “quality clicks” is a metric Google defined, measures privately and reports to nobody.
The tool Google gives site owners does not settle it either. Search Console's generative AI reporting shows how often your pages appeared in AI Overviews and AI Mode, but reports no clicks, no clickthrough rate and no queries, and the impressions were already inside your existing totals. We covered what that report does and does not contain in Google now reports AI impressions — it does not report the clicks. Until that changes, no site owner can independently check either Google's claim or Pew's.
What should a founder take from this?
Four things follow that are worth more than the headline percentage.
Stop looking for your number in a study. None of these datasets is measuring your query mix, your industry or your brand. The useful comparison is your own Search Console impressions and clicks for the queries you care about, tracked across the period a summary started appearing for them. That is a worse dataset than Pew's in every respect except the one that matters, which is that it is about you.
Treat an AI citation as a branding event, not a traffic event. One per cent of visits produce a click on a link inside the summary. If a visibility service is priced on the traffic a citation will send, the Pew figure is the number to hold it against. If it is priced on being the source a summary paraphrases, that is a different and more defensible proposition — and a much harder one to measure, which is the subject of how a company should measure whether placement worked.
Look at your query shape before your content. Navigational and brand-name queries are the least exposed part of a search footprint; long informational queries are the most exposed. That split is more actionable than any site-wide average, and it points at where the risk sits in the shift from search optimisation to answer optimisation covered in SEO, AEO and GEO.
Expect the ground to move. AI Overview prevalence went from 6.49% of US desktop queries in January 2025 to about 20% by April, and the zero-click share moved 7.5 points between 2024 and 2026. A vendor quoting a figure in this field without its date is quoting a figure without its meaning.
Media Morpheus sells media placement, partly on the argument that coverage feeds the sources AI systems summarise. The Pew data constrains that argument as much as it supports it: coverage reaching an AI summary is reaching an audience that clicks through about 1% of the time. The visit is not the product any more. The mention is. Anyone selling you the visit should be asked which study they are using.
Methodology and limitations
Primary figures are from “Google users are less likely to click on links when an AI summary appears in the results” by Athena Chapekis and Anna Lieb, Pew Research Center, 22 July 2025, drawing on browsing data from 900 US adults on KnowledgePanel Digital for 1–31 March 2025: 68,879 unique Google searches, 12,593 of which produced an AI summary. Result pages were re-scraped 7–17 April 2025. Four limits apply. First, the classification of which searches carried a summary was made up to six weeks after the search, and Pew notes summaries change. Second, it covers Google only, on tracked browsers, so app searches and non-US behaviour are out of scope. Third, searches carrying a summary differ systematically from those that do not — they are longer and more often questions — so the 8%-versus-15% comparison is not a controlled experiment and Pew does not present it as one. Fourth, 900 people is a small panel for cutting by query type. The Ahrefs figure is from Ryan Law, 17 April 2025, comparing position-one clickthrough rate across 300,000 keywords for March 2024 against March 2025; its 34.5% is measured against a forecast counterfactual of 0.040, not the raw 64% fall in the same data, and Ahrefs sells SEO software. Similarweb, Semrush and SEOClarity figures are as reported by Digiday on 10 July 2025 and are vendor panel data covering specific query populations, principally news. The SparkToro figures are from Rand Fishkin, 9 June 2026, using Similarweb's clickstream panel for the United States, January to April 2026; Fishkin states the long series combines three providers and excludes Google's native app. Google's claims about click volume and click quality come with no published data and are reported here as claims.
Conclusion
Yes, on the queries where they appear. The only non-commercial measurement available found the click rate on a Google results page roughly halved when an AI summary was present, 8% against 15%, and found the links inside the summary clicked on 1% of visits. Vendor datasets agree on direction and disagree on magnitude by a factor of several, because they count different things over different populations. Google denies the aggregate effect and publishes nothing that would let anyone check. The safe conclusion is narrow and still useful: the click is becoming the scarce part of the transaction and the citation the common part, and the two should not be priced as if they were the same thing.
Frequently Asked Questions
What did Pew actually measure?
The Pew Research Center analysed the browsing records of 900 US adults on the KnowledgePanel Digital panel, covering every URL they visited on a tracked device during March 2025. The dataset holds 68,879 unique Google searches, of which 12,593 — 18% — produced an AI summary.
How far do clicks fall?
Users clicked a standard search result on 8% of visits to pages carrying an AI summary, against 15% of visits to pages without one. They clicked a link inside the summary on 1% of those visits. Sessions ended on the search page 26% of the time with a summary present and 16% without.
What do the SEO datasets find?
Ahrefs put the AI Overview effect at a 34.5% lower clickthrough rate for the top-ranking page across 300,000 keywords. Similarweb data reported by Digiday showed news searches ending without a click rising from 56% in May 2024 to nearly 69% in May 2025. SparkToro, using Similarweb's panel, put US zero-click searches at 68.01% for January to April 2026.
Where do the studies disagree?
On the unit they measure. Pew counts visits by a panel, Ahrefs counts keyword-level clickthrough rate from Search Console, Similarweb counts visits to domains. They also differ on population: Pew covers all queries, the 69% figure covers news queries only, and Ahrefs covers informational keywords at position one.
What does Google say about all this?
Google told reporters the Pew study “uses a flawed methodology and skewed queryset that is not representative of Search traffic”, and Liz Reid wrote in August 2025 that total organic click volume has been “relatively stable year-over-year”. Google published no data behind either claim.
What should a founder take from this?
That the direction is not in dispute and the magnitude is unknowable from outside. Measure your own traffic rather than a published average, treat a citation inside an AI Overview as a branding event rather than a traffic event, and look at how many of your queries are long or question-shaped, because those are the ones that trigger summaries.
Sources
- research Chapekis, A. and Lieb, A., 2025. Google users are less likely to click on links when an AI summary appears in the results. Pew Research Center, 22 July 2025.
- vendor Law, R., 2025. AI Overviews Reduce Clicks by 34.5%. Ahrefs, 17 April 2025. 300,000 keywords; Ahrefs sells SEO software.
- trade press Guaglione, S., 2025. AI platforms are driving more traffic — but not enough to offset ‘zero-click’ search. Digiday, 10 July 2025, reporting Similarweb, Semrush and SEOClarity data.
- vendor Fishkin, R., 2026. In 2026, Less than One Third of Google Searches Still Send a Click. SparkToro, 9 June 2026, using Similarweb clickstream data.
- operator Reid, L., 2025. AI in Search: driving more queries and higher quality clicks. Google, 6 August 2025. No underlying data published.
- trade press Claburn, T., 2025. Google AI Overviews are killing the web, Pew study shows. The Register, 22 July 2025, carrying Google’s statement on the Pew study.