Searches That End Without a Visit

The ordinary assumption behind a corporate website is that people will eventually reach it. Everything else — the coverage, the directory entries, the search listings, the summaries — was understood as a route, and the site was the destination where the organisation finally spoke for itself. That assumption has become measurably wrong for the majority of searches, and the consequences are structural rather than promotional.

The clearest recent measurement comes from the Pew Research Center, which in July 2025 published an analysis of 68,879 searches performed by 900 US adults in March 2025. It found that Google users clicked a traditional search result on 8% of visits to a results page containing an AI summary, compared with 15% of visits to pages without one. The same study found that users clicked a source link inside the summary itself on 1% of visits. The summary is not a signpost. It is where the enquiry stops.

The wider trend is the same and older. SparkToro’s clickstream analysis with Similarweb found that 68.01% of US Google searches ended without any click between January and April 2026, against 60.45% in 2024 and roughly 45% a decade earlier. Two thirds of the time, the search is the whole event.

What is actually changing

It is worth separating three things that are often reported as one.

The first is the volume of clicks, which is falling. The second is the share of results pages carrying a machine-written summary, which is rising: Pew observed a summary on 18% of the searches in its March 2025 sample, while Similarweb’s monitoring of US Google results across June 2025 to May 2026 recorded AI Overviews on more than 43% of searches. The third is the behaviour of the person doing the searching, which is where the interesting part sits. Pew found that 26% of sessions ended entirely after a results page carrying an AI summary, against 16% of sessions after a page without one. People are not being intercepted on the way somewhere. They are satisfied, and they leave.

Satisfaction is the awkward part. A summary that sends nobody onward is not necessarily failing its reader. It is failing only the party that expected the visit, and that party has no way to observe the encounter at all.

The record is now read at second hand

An organisation used to be able to treat its own site as the authoritative version of itself and everything else as an approximation of it. The approximation is now what most people read, and the site functions less as a destination than as source material for a description written elsewhere.

This inverts a familiar priority. Material written to persuade a visitor who has already arrived — the positioning statement, the narrative About page, the case study that assumes context — is written for a reader who increasingly does not appear. Material written to be extracted correctly by something that will never read the page in full is written for the reader who does: plain statements of what the organisation is, what it sells, to whom, where, and since when, in text that survives being lifted out of its surroundings.

Neither kind of writing is dishonest. But only one of them survives the trip.

The second-order effect on smaller sources

There is a distributional point here that is easy to miss when the headline figure is a single click-through rate. SparkToro and Datos, examining US and EU Google searches from September 2022 to May 2024, found that only 360 clicks per 1,000 US Google searches went to the open web, with Google-owned properties absorbing roughly 30% of clicks. The clicks that remain are not distributed evenly across everything a summary drew on. They concentrate.

For an organisation, this means that the sources feeding descriptions of it and the sources receiving visits from those descriptions are increasingly different sets. A trade directory, a registry entry, or a technical page may contribute materially to how an organisation is characterised while generating no measurable traffic whatsoever. Judged by referral analytics, such a page looks like nothing. Judged by its effect on the description in circulation, it may be one of the more consequential things published about the organisation.

Anyone allocating effort by measured traffic will therefore systematically underinvest in exactly the material that does the work. This is not a hypothesis about the future; it follows from the click data already collected.

What can and cannot be done about it

Three responses are available, and they are not equally serious.

The first is to attempt to recover the click — to write in ways calculated to leave the summary incomplete so that the reader must continue. This is a race against a system that is improving at completeness, and it degrades the material for the humans who do arrive. It is also, in aggregate, an argument for publishing worse pages.

The second is to treat the traffic as lost and disengage, on the reasoning that if nobody visits, nothing published matters. This misreads the data. The clicks fell; the reading did not. The material is being consumed more than before, by more parties, with less attribution and no session recorded.

The third is to accept that the public record is now the thing being consumed and to maintain it on those terms, which means keeping it factually right, keeping it in step with the present, removing the ambiguities that force a reader or a system to guess, and ensuring it is confirmed somewhere other than the organisation’s own domain. This yields no reportable traffic gain. It changes what is said about the organisation in the places where the enquiry now ends.

The measurement problem is real and worth stating plainly rather than solving falsely. There is no analytics view of a search that ended without a click. The organisation cannot see the impression it made, cannot see the description that produced it, and receives no signal when either changes. What it can do is check the inputs, which remain public and readable, and take the absence of traffic as an unreliable proxy for the absence of attention.