Pew Puts A Number On AI Summaries: 8% Outbound Clicks With, 15% Without

Pew Research Center measured what happens to outbound clicks when Google puts an AI summary on the results page: users clicked a traditional result link on 8% of visits where a summary appeared, against 15% of visits where none did ^[1][2]. The measurement came from real browsing activity of 900 US adults across 68,879 Google searches in March 2025 ^[4][5], which makes it the first number solid enough to put into a referral-traffic forecast rather than into a complaint.

Run the arithmetic before running the strategy. The gap is 7 percentage points, a 47% relative decline in outbound clicking on summary pages ^[18]. Pew also found users clicked links inside the summary itself on roughly 1% of visits ^[3], so the generous reading, counting in-summary links as arrivals, is 9% versus 15%, still 40% lower ^[19]. Neither figure is the number you should plug in wholesale, because the input that matters is coverage: the share of your query mix that draws a summary at all. If half of it does, the blended outbound click rate is 12%, or 20% below the no-summary baseline ^[22]. That is a modelling exercise, and it is a different exercise from defending rankings.

The session-ending figure is the one that should worry anyone selling second and third pageviews. Browsing stopped after 26% of visits to a results page with a summary, versus 16% without ^[10], a 10-point gap and roughly 1.6 times the rate ^[20]. Pew reports the typical summary ran about 67 words ^[9], which is enough to close a simple informational query without a follow-up. About 58% of respondents ran at least one search that produced a summary during the month ^[8]; participants averaged roughly 77 searches each over the period ^[21].

Pew is explicit that the association is not causation, and that the data do not establish that summaries alone produced the difference ^[7]. The study is a snapshot of Google in the US in March 2025, not a measure of every search engine, country, audience, or query type ^[11]. It also does not show which categories of site lose the most, whether users judged the summaries accurate or useful, or how behaviour varies by query type ^[12]. Anyone quoting 8% versus 15% as a universal haircut is overreading it.

The operational consequence Pew draws is that visibility and referral traffic are now separate outcomes: a page can rank, be cited in a summary, and still be visited less ^[13]. That argues for decoupling ranking reports from traffic reports ^[14], and for measuring qualified leads, subscriptions and product actions alongside organic sessions ^[15]. Pew does not conclude that search optimisation is irrelevant ^[23]. It also flags a quieter problem for internal systems: when answers are synthesised at the interface layer, documentation and knowledge bases influence discovery without producing the page view that used to signal they were working, so provenance and update discipline carry more weight ^[17]. Content governance, meaning named ownership and review cadence for material that may be summarised before it is read, follows from the same logic ^[16].

What to watch is coverage drift. The 58% monthly reach and the 67-word summary length ^[8][9] are March 2025 readings from one panel; the multiplier moves with them, and Pew's data does not break coverage out by query category ^[12], so operators who want that number will have to instrument it themselves.