When Google shows an AI summary, 8% of people click a normal result, against 15% when it does not.
That pair of numbers comes from Pew Research Center's study of 68,879 real searches, and it is the cleanest evidence yet of what zero click search does to a website.
The awkward part for a small UK business is that the mention still happens. Your firm gets named in the summary, or quoted by ChatGPT, or shown in the map pack. The person reads it, decides, and never arrives on your site. Your analytics show a flat line while the market has, in fact, heard of you. That gap between being cited and being visited runs through the whole generative engine optimisation evidence base.
This article sets out the numbers, what Google says in reply, and five things to measure instead of sessions. It ends with a rule for deciding when a zero-click result is fine and when it is costing you money.
What zero click search means in 2026
A zero click search is a Google search that ends without a click on any result, paid or organic. The searcher reads the page, gets what they need or gives up, and either searches again or leaves. The term comes from Rand Fishkin at SparkToro, whose clickstream studies are the standard reference.
The 2024 SparkToro and Datos study put the rate at 58.5% in the US and 59.7% in the EU. For every 1,000 US searches, 360 clicks reached a site that Google neither owned nor was paid by.
The 2026 update, on Similarweb panel data from January to April, found 68.01% of US searches ended without a click, up from 60.45% on the same basis in 2024. Fishkin's advice from that post is blunt:
"Replace traffic as a KPI for your digital marketing efforts." Rand Fishkin, SparkToro, 2026 zero-click study
The UK is worse than the US. SparkToro's country breakdown, published in June 2026, found 69.5% of UK Google searches ended with no click, the highest of the four markets studied. Only 232 clicks per 1,000 UK searches reached the open web, and 43.1% of UK searches ended the session outright. Germany was 62.1%, France 65.3%.
The method matters. These panels cover desktop and mobile web browsing, not the Google app, and the 2024 and 2026 studies use different panels. Treat the trend as solid and the second decimal place as noise.
What the Pew study found about AI Overviews
Pew's study is the one to quote because it measured real behaviour, not survey answers. In March 2025, 900 US adults on an opt-in panel had their browsing tracked. Of 68,879 unique Google searches, 12,593 produced an AI summary, which is 18% of the total.
xychart-beta
title "Pew Research, March 2025: share of Google search visits that clicked a result"
x-axis ["Clicked a traditional result", "Clicked a link inside the AI summary", "Ended browsing session"]
y-axis "Per cent of visits" 0 --> 30
bar [15, 0, 16]
bar [8, 1, 26]
The first bar in each pair is pages without an AI summary, the second is pages with one. (There is no AI link to click when no summary appears, so that first value is zero by definition.)
The figures to take from it: 8% versus 15% for a click on a traditional result, 1% for a click on a source inside the summary, and 26% versus 16% for ending the browsing session. Longer, question-shaped searches triggered a summary far more often. Searches of ten or more words got one 53% of the time, and queries starting with who, what, when or why got one 60% of the time.
Two further points matter for a UK business. In Pew's data, Wikipedia, YouTube and Reddit were the three most-cited sites in both AI summaries and standard results, and .gov domains made up 6% of AI summary sources against 2% of normal results. Official and community sources get the links. Commercial sites get summarised.
What Google says, and why both sides can be right
Google's position is that total clicks are holding up. In August 2025, Liz Reid, head of Google Search, wrote that organic click volume was "relatively stable year-over-year" and that Google was sending slightly more quality clicks than a year earlier. She defined the term: "by quality clicks, we mean those where users don't quickly click back". Reports of steep traffic drops, the post said, rested on weak methods or isolated examples.
This is not a contradiction of Pew or SparkToro. Google is reporting a total across all sites and all searches. Search volume is growing, so a smaller share of a bigger number can stay flat. Pew measured the rate per search. Both can be true, and the second is the one that describes your site.
The practical reading is this. If your traffic held steady over the past two years, you have lost share of a growing market, not kept pace. If it fell, you are in the group Google describes as isolated examples, which is cold comfort when the invoices are the same.
Why your analytics hide the effect
The two reports you would reach for first cannot show you a zero-click outcome, because nothing happened on your site to record.
Search Console does count AI clicks. Google's performance report documentation says a click on an external link inside an AI Overview or AI Mode response counts as a click, standard impression rules apply, and every link in an AI Overview is given the same single position. The problem is that these rows are merged into the normal web search totals. You cannot filter them out, so you cannot tell an AI impression from a page-two impression.
The separate Search Generative AI performance report, launched on 3 June 2026 and rolled out to all sites by 31 August 2026, shows impressions, pages, countries, devices and dates for AI features. It is one of the few concrete measurement additions in Google's May 2026 guidance. It does not show clicks or queries. You can see that a page appeared inside an AI answer. You cannot see what was typed or whether anyone followed the link.
GA4 has the opposite gap. A person who read about you in an AI answer, closed the tab, and typed your name a week later arrives as "direct" or as a branded organic click. The AI exposure that caused the visit is invisible, and last-click reporting hands the credit to the final step.
Five things a UK SME should measure instead
The measures that still move when nobody visits are the ones tied to your name and your phone number. None of them is perfect. Together they show whether awareness is rising while sessions are flat.
Branded search impressions in Search Console. Filter the Queries report to searches containing your business name and compare the last three months with the same three months a year earlier. Watch impressions, because clicks are the thing that stopped happening. Someone who typed your name after an AI answer counts here even if they then rang you from the map pack.
Direct traffic trend. In GA4, take the Direct channel as a monthly series and plot it against the same month last year. Strip out any month where you ran print, radio or a mailshot. A steady rise in direct sessions with no offline campaign behind it usually means people are learning your name somewhere you cannot see.
Google Business Profile views, calls and direction requests. The Business Profile performance page reports unique profile views on Search and Maps, call button taps, direction requests, website clicks and message conversations, with one person counted once per day. For a local firm, calls and directions are the conversions that a zero-click search produces. Export them monthly, because the dashboard is poor at long comparisons.
"How did you hear about us?" at the point of enquiry. Add a free-text field, not a dropdown, to the contact form and to the phone script. Dropdowns miss "ChatGPT told me" because nobody thought to list it. Review the answers every month and count the ones that name an AI tool, Google without a click, or a recommendation that traces back to a search. We ask new clients to add this field before any campaign starts, so the baseline exists.
Attribution paths in GA4. The key event attribution paths report splits each converting path into early, mid and late touchpoints and shows which channels appear in each segment. If organic search keeps showing up as an early touchpoint while direct takes the last click, the search visit is still doing work even though it never closes the deal on its own.
A sixth measure, AI citation tracking, is worth adding once the five above are in place. Running the same 30 to 50 buyer questions through ChatGPT, Perplexity and Google AI Mode each month and recording whether your firm is named gives you the mention rate that no Google report shows. That is the core of the AI visibility work we do, and it only means something set against the branded search and enquiry figures.
When a zero-click result is fine, and when it is a loss
A zero-click result is fine when the search was never going to convert on your site, and it is a loss when the click was the only route to the sale.
Awareness searches are the first case. Someone asking "what does a conveyancing solicitor do" or "how long does a boiler service take" was reading, not buying, and for law firms that early mention still builds the name. If the AI summary names your firm or quotes your page, you got the mention for free and saved the server a visit. Measure it through branded search impressions, and do not grieve the session.
Local intent is mostly fine too. A person who finds a Kettering physio or dental clinic in the map pack, reads the reviews and presses the call button has done what you wanted. The click to your website was a detour. Business Profile calls and directions carry the result, and that is where local SEO effort pays back.
The loss comes in three situations. First, the summary answers the whole question with your content and the reader has no reason to remember you. A table of your prices quoted without your name helps a competitor who gets the next click. Second, the enquiry needs a form, a configurator or a booking page that lives on your site and nowhere else. Third, the AI answer names you alongside four rivals and the person picks on the summary alone, which turns your differentiator into a coin toss.
The rule we use: if branded search impressions and Business Profile actions are rising while organic sessions fall, the loss is cosmetic and the budget should go to being named more often. If all three fall together, the mention is not happening either, and the problem is visibility, not clicks. If sessions fall and enquiries fall but branded search holds, the site itself is losing the people who do arrive, and that is a conversion problem wearing a zero-click costume.
Run that check quarterly, on the same date ranges, before deciding whether the drop in traffic needs fixing at all. Half the time the honest answer is that nothing is broken except the metric you were watching.
If you want help setting up the measurement, or finding out how often UK AI answers already name your business, talk to us about AI visibility.