Build1 publisher3 min readPublished
Pew's seven-point click gap shrinks to 1.3 points across the whole sample
Pew's click-rate gap between visits with and without an AI summary is measured across two different groups of searches in one month, so how much of it reaches a page depends on that page's query mix.
The Engineer · Build desk
What happened
- Pew Research Center's March 2025 panel logged 68,879 Google searches made by 900 U.S. adults.
- Users clicked a traditional search result in 8% of visits where an AI summary appeared, and in 15% of visits where none did.
- Links inside the summaries were clicked in 1% of all visits, and 88% of the summaries cited three or more sources.
Compiled by The EngineerSomething wrong?How this is made
Why it matters
- constraint The study observed what people did, not how Google chose its citations, so it gives a team no route to getting cited in a summary and no way to test one.
- decision Whether sessions are still a fair proxy for a page's search performance now has to be decided page by page, because the summary trigger rate depends on the length of the queries that page wins.
- exposure Government sites took 6% of AI-summary citations against 2% of standard search links, so a site's share of citations can rise while its click count does not.
The 8% and the 15% are not two readings of the same searches. They are two groups of visits from the same month, sorted by whether Google put a summary on the results page [2]. Weight the two rates by how often that happened and the sample-wide traditional click rate comes to 0.18 x 8 + 0.82 x 15 = 13.7%, about 1.3 points under the no-summary rate [3][2]. That treats 15% as what the summary visits would have returned with no summary present, and the study's one-month design cannot test the assumption [1].
Summary presence tracked query length. Summaries appeared in 8% of one- and two-word searches and in 53% of searches running ten words or more [4]. The with-summary group is therefore weighted toward long specific queries and the without-summary group toward short ones. The dev.to write-up of the study states the limit directly: the zero-click difference "does not establish that an AI Overview was the sole reason a user did not click" [11].
A single page's exposure depends on its query mix. A page winning one- and two-word queries sits at 0.08 x 8 + 0.92 x 15 = 14.4%. A page winning ten-word procedural queries sits at 0.53 x 8 + 0.47 x 15 = 11.3%, a quarter below the 15% figure [3]. The same two click rates land 3.1 points apart depending on which queries the page ranks for [3].
Pew's in-summary click figure does not sit with the rest. Links inside summaries were clicked in 1% of all visits [6]. If visits carrying a summary are the same 18% share, that is roughly 5.6% within summary visits [4], which on top of the 8% traditional click puts about 13.6% of summary visits ending in an outbound click against 15% elsewhere [5]. The zero-click numbers point the other way: 26% of cases with a summary ended without a website visit, against 16% without one [7], a 10-point gap [6]. Three denominators are in play here, all visits for the 1%, visits with a summary for the 8%, and sessions for the 26% [2][6][7], so this data does not fix the size of the outbound loss.
One reporting consequence survives all of it. The write-up says a page "can remain visible in results while receiving fewer visits because a searcher's immediate question has already been answered on Google" [12], and that a declining visit count "may reflect a changed results page, rather than a loss of rankings or a decline in demand" [13]. Position and impressions have to be read separately from sessions, cohorted by query length, before a traffic drop gets logged as a ranking problem.
What to watch
- A split of clicks by query length would show whether the summary or the query drives the seven-point gap; the write-up does not report one.
- A repeat wave of the panel over the same queries would supply the baseline a single-month design cannot give.
- If Google's trigger rate for ten-word-plus queries moves off 53%, the exposure for long-tail informational pages moves with it.