Most claims about what AI search is doing to website traffic come from vendors with a product to sell. The Pew Research Center study published in July 2025 is unusual because it measures behaviour rather than opinion and publishes its method.
What it measured
Pew used the browsing data of 900 United States adults drawn from its KnowledgePanel, covering every Google search they made between 1 and 31 March 2025. Eighteen percent of those searches produced an AI summary.
On search pages that carried an AI summary, users clicked a traditional search result on 8 percent of visits. On pages without one, they clicked on 15 percent of visits. They clicked a source cited inside the summary itself on 1 percent of visits. They ended the browsing session entirely on 26 percent of pages with a summary, against 16 percent without.
Why the citation figure is the important one
A great deal of advice about answer engines is organised around being cited. The Pew figure puts a ceiling on what a citation is worth as traffic. One percent of visits produced a click on a cited source, against 8 percent on an ordinary result and 15 percent where no summary appeared.
That does not make citation worthless. It makes it worth something other than traffic. A citation places your name inside the answer a buyer reads, which is a presence effect rather than a visit, and it should be valued and measured as one.
What we take from it
The first conclusion is arithmetic. If a meaningful share of your search impressions now resolve into a summary, the traffic decline that follows is structural and no amount of ranking work reverses it. Treating it as a ranking problem produces a year of expensive effort against a cause that is not there.
The second is about what a page is now for. A page that exists to capture a click and then convert it is competing for a shrinking event. A page that exists to be the complete, quotable answer is competing for the thing that is growing, and it does so whether or not the reader ever arrives.
The honest limitation is that this study covers general consumer search behaviour in one country over one month, not commercial buying research specifically. We treat it as strong evidence about the direction and weak evidence about the magnitude inside any particular market, including yours. The way to get the second is to measure your own.