Between January and April 2026, 68.01% of US Google searches ended without anyone clicking anything. No link, no ad, no result of any kind. SparkToro published that figure in June 2026, working from Similarweb clickstream data covering US desktop and mobile search across those four months.
If your website traffic has been sliding for a year and nobody has given you a good reason why, that is most of the reason.
The number, and where it comes from
A decade ago the zero-click share sat around 45%. In 2024 it was 60.45%. SparkToro’s analysis puts it at 68.01%, and publishes its methodology alongside the post.
One detail cuts against the usual story. Google’s AI Mode accounted for only 0.34% of searches in that window. Whatever is taking your clicks, it is not customers abandoning Google for a chatbot tab. The slide started long before AI summaries existed. Google has spent ten years putting more answers on its own page, from store hours to phone numbers, and a language model summary is the newest layer on a very old habit.
What an AI summary does to a click
Pew Research Center measured this directly rather than inferring it. In a July 2025 study, Pew used opt-in browsing data from 900 US adults in the Ipsos KnowledgePanel, covering every URL they visited in March 2025: 68,879 Google searches, 12,593 of them returning an AI summary.
On pages carrying an AI summary, people clicked a traditional result on 8% of visits. On pages without one, 15%. The link inside the summary did not absorb the difference, because people clicked those on 1% of visits.
More often the session simply stopped. 26% ended their browsing entirely after a page with an AI summary, against 16% without one.
Note the date. That is March 2025 browsing data, and AI summaries have spread since.
The damage by ranking position
Ahrefs published updated numbers in February 2026 comparing 300,000 keywords, half with an AI Overview present and half informational keywords without one, against aggregated Google Search Console desktop click-through data from December 2023 and December 2025. The presence of an AI Overview correlates with a 58% lower average click-through rate for the number one result. Ahrefs’ earlier study, published in April 2025, measured a 34.5% reduction using a different comparison window.
Further down the page the effect shrinks. Position two loses 50.8%, position three 46.4%, position ten 19.4%.
Which does not make position ten safer ground. It never earned many clicks, so there was less to take. An AI summary flattens the premium on ranking first, and that premium has been worth several times a mid-page ranking. If your whole search strategy is one page ranking first for one phrase, that is the asset losing the most value right now. Ahrefs calls this a correlation and so do we, since no experiment was run.
The part nobody reports, AI traffic converts better
Adobe Analytics reported in January 2026 that traffic to US retail sites from generative AI tools rose 693.4% year over year across the November to December 2025 holiday season. What those visitors did after arriving matters more than the growth rate: AI referrals converted 31% better than other traffic sources, were 33% less likely to bounce, spent 45% more time on site, and drove revenue per visit up 254%. Adobe pulled that from transaction data covering more than a trillion visits to US retail sites.
The caveat belongs up front. Adobe measured retail and other large sectors, not a two-truck HVAC company. Nobody has published a comparison like this for local service businesses at a scale we would trust, so carrying it over to a roofing company is analogy rather than evidence.
The mechanism is not mysterious. Someone arriving from an AI answer has read a summary, narrowed the field, and clicked anyway. Fewer visitors, further along, is a different kind of traffic and not obviously a worse one.
What people think of AI summaries when asked
Pew surveyed 5,153 US adults in August 2025 through its American Trends Panel. 65% at least sometimes run into AI summaries. Among those who have seen them, 20% call them extremely or very useful, 52% somewhat useful, and 28% not too useful or not useful at all.
Trust runs thinner than usefulness. 53% have at least some trust in what the summaries tell them, but only 6% trust them a lot.
A later Pew report from June 2026, based on 5,119 US adults surveyed that February, found 49% of US adults have used an AI chatbot, up from 33% in 2024, with looking up information the most common use at 42%. 60% say they read the AI summaries at the top of search results.
That is your customer this year: she reads the summary, finds it moderately useful, does not entirely believe it, and goes looking for something firmer before spending money.
What still earns a click
These statistics measure aggregate clicks and citations. They cannot tell you which of your pages will keep pulling traffic. What follows sticks to the data and flags where we are inferring.
Query types
Searches that do not trigger an AI Overview still behave the way they always did. The Ahrefs comparison set was informational keywords, and the 58% figure applies where an Overview appears. Overviews cluster on questions with a settled factual answer, which is not what someone types right before hiring a contractor.
Local recommendation searches end in a shortlist. BrightLocal found in March 2026, surveying 1,002 US adults, that 45% of consumers have used AI tools to find local businesses, up from 6% a year earlier, putting AI third behind Google and Facebook. Being named on that shortlist puts you in the middle of the errand, not the end of it: the next thing she does is look you up.
Transactional searches survive for a structural reason: price, booking, availability, hours, financing. No summary can hold a Thursday morning open for you. Nor can it answer what only you know, like whether you take a particular insurance, which counties you cover, or what you charge to come look at a job. If none of that is published on your site, no system can repeat it and no customer can find it.
Page types
Pew’s July 2025 analysis found 88% of AI summaries cited three or more sources, and only 1% cited a single source. Being one of several cited sources is reachable; being the only one is close to impossible. The pages that get cited have something specific enough to quote.
The same analysis found Wikipedia, YouTube and Reddit together accounted for 15% of all sources listed. General explanatory content is well covered, and your version of “what is a heat pump” is competing with an encyclopedia. Pages that finish something instead, a quote request, a booking, an application, do work no summary can do.
The last piece is our inference, so treat it as one. The median AI summary in Pew’s data ran 67 words. We read that to mean a summary can carry a fact but not a decision involving a contract, a roof, or someone’s teeth. That gap is where clicks still live. It is a hypothesis to check against your own numbers, not a finding.
Why your own website matters more now, not less
When traffic drops, the reflex is to decide the website stopped working. Sometimes that is true. More often the yardstick broke rather than the site. Sessions up or down made sense in 2015. In 2026 it measures how Google distributes attention, which says nothing about whether your business is any good.
Better questions: how many calls, form submissions and booked appointments came in last month, and did that number move with sessions? If sessions fell by a third and calls held flat, nothing broke. Your site did the same work with fewer, better-qualified visits, roughly what the Adobe pattern would predict if it holds outside retail.
Your own pages also gain structural weight. Every AI summary needs sources, and directories, review platforms and social profiles are all rented ground. If your service area, pricing approach, warranty terms, and the answers to the ten questions you field on the phone every week live only in your head, nothing can cite them.
None of the work this implies is glamorous. A page for each real question, in the words customers use. Prices or ranges where you can give them. A service area stated by county rather than “and surrounding areas.” Booking and quote pages a person can finish on a phone in under a minute. Most of the website design and development work we take on now is that shift, from a site that describes a business to a site that answers for it. So when an owner tells us traffic is down, the first thing we check is whether anything broke at all. Often nothing did, and there is no rebuild to sell.
One step this week. Open Google Search Console and pull the last 18 months of impressions and clicks side by side, then pull your call and form volumes for the same period. Read the three lines together. Impressions steady, clicks down, calls flat is the shift described here, not a broken website, and the work belongs on your pages rather than your rankings. Impressions down as well is a different problem with a different fix. Either way you now know which one you have before anyone quotes you for it.