The 68 percent problem

The oldest deal on the open web was struck in a search box. You made your pages legible to Google; Google sent you the click. Rank, click, convert. Three steps, and the middle one carried the entire economics of being findable. In the first four months of 2026, that middle step failed to happen in 68 out of every 100 American Google searches. Not because people stopped asking questions. Because the answer now arrives before the click is needed.

This is the "why now" file. Everything else we publish about AI visibility rests on the claim that the click-based funnel is dissolving, so this piece does one job: it lays out the three most defensible numbers in the field, each with its method and its window, and shows that they point the same way. A clickstream panel, an independent behavioral study, and publisher-side analytics. Three rulers that share no data source. One direction.

Section OneTwo thirds, and rising

Start with the widest ruler. Rand Fishkin of SparkToro analyzed Similarweb's US clickstream panel (desktop and mobile) for January through April 2026 and found that 68.01 percent of US Google searches ended without a click on any result. In 2024, the same panel put that figure at 60.45 percent. That is a rise of 7.56 percentage points in roughly two years, on a behavior that had been drifting for a decade.

SparkToro, R. Fishkin, "In 2026, Less than One Third of Google Searches Still Send a Click," 2026. Similarweb US clickstream panel (desktop + mobile), January–April 2026.

Read the definition before you repeat the number, because the definition is what makes it strong. A "click" here is generous: clicks on organic results count, clicks on ads count, clicks to Google's own properties count. Follow-up searches are excluded, so rephrasing a query does not inflate the figure. Even with that generous definition, more than two thirds of searches go nowhere. The searcher asks, reads the results page, and leaves. Or never leaves Google at all.

The same analysis carries a second, separate metric, and it deserves its own sentence because the two are routinely blended into nonsense. The share of searches that generate at least one click fell by 9.51 percentage points, a 22.9 percent relative decline. That is not the zero-click increase restated; it is a different ruler laid against the same behavior. The 7.56-point figure measures how much more often searches end clickless. The 9.51-point figure measures how much rarer the click-generating search has become.

Two searches out of three now end where they began: on Google.

Section TwoWhat the answer does to the click

The panel tells you that the click is disappearing. The Pew Research Center study tells you what happens in the moment it disappears, and it is the strongest behavioral evidence in the field. In July 2025, Pew published an analysis of 68,879 real Google searches made by 900 US adults on its KnowledgePanel, drawn from their actual March 2025 browsing. Not a survey of what people say they do. A record of what they did.

Pew Research Center, "Do people click on links in Google AI summaries?," July 22, 2025. 68,879 searches from 900 US adults, KnowledgePanel, March 2025 browsing data.

About 18 percent of those searches (12,593 of them) produced an AI Overview, Google's machine-written answer above the results. When that answer appeared, the click rate on traditional results dropped from 15 percent to 8 percent. The presence of the AI answer roughly halves the already-thin chance that any website receives the visit.

And the links inside the AI answer, the citations, the sources the machine names? Clicked in 1 percent of visits. One in a hundred. People read the answer, take it, and move on: 26 percent of visits with an AI summary ended the browsing session entirely, against 16 percent without one. The answer is not a doorway. It is a destination.

Eight percent click through the answer. One percent click into it. Everyone else is done.

Worth stating plainly: Pew is an independent research organization running its own panel. This is not Google's account of Google, and not a vendor with a product to sell against the trend. It is the closest thing the field has to a neutral instrument, and it agrees with the clickstream.

Section ThreeThe receiving end

If the click is dissolving, someone's analytics dashboard should be recording the loss. Someone's is. Chartbeat, the real-time analytics service running inside publisher websites, measured traffic across more than 2,500 publisher sites, and the Reuters Institute cited its data in the 2026 edition of its annual journalism trends report: Google search referral traffic to publishers fell by roughly a third globally, and by 38 percent in the United States, in the year to November 2025.

Chartbeat data (2,500+ publisher sites), cited in Reuters Institute, "Journalism and Technology Trends and Predictions 2026"; reported by Press Gazette.

This is the third method: not a panel watching searchers, but meters installed on the receiving end, counting arrivals. And it adds the detail the averages hide. The US decline, at 38 percent, ran more than twice as deep as Europe's 17 percent: AI answers rolled out earlier and wider in American search. And the loss is brutally regressive. The largest publishers lost 22 percent of their Google traffic over two years. Mid-size sites lost 47 percent. Small sites (under 10,000 daily pageviews, which is to say almost every small business with a content strategy) lost 60 percent.

Section FourAnswer → done

Put the three rulers side by side. The panel says two thirds of searches end without a click, and the share is rising. The behavioral study says the AI answer halves the remaining clicks and that its own citations are followed one time in a hundred. The publisher meters say the traffic loss is real money, already booked, and steepest for the smallest. The funnel that organized twenty years of marketing (rank, click, convert) is losing its middle step. What replaces it is shorter and stranger: answer → done.

The demand did not go anywhere. People ask more questions of machines than they ever typed into search boxes, and the machines answer by synthesizing what the web says about you. What vanished is the visit: the moment where your pixels, your funnel, your retargeting had their chance. The buyer still decides. The deciding now happens inside an answer you never see, assembled from sources you may or may not be in.

The new unit of value is presence in the answer, not passage through the page.

That is the strategic inversion, and it is worth saying without euphemism. For two decades, the ranking was the asset and the click was the payoff. Now being the cited source inside the answer is the asset. And it pays even when the 1 percent don't click, because the answer carries your name, your claim, your framing to the person deciding. Presence without traffic used to be a failure state. It is becoming the product. The businesses that adapt first will be the ones that stop measuring visibility in sessions and start measuring it in citations: how often the machine's answer contains them, describes them accurately, and hands them the recommendation.

What earns that citation is not a mystery, and it is not a trick. Machines assembling answers reach for sources that make extraction easy: claims stated plainly, numbers with their windows attached, an entity whose name, offer, and location read the same on every page that mentions it. The work looks less like chasing keywords and more like publishing the reference document on the things you actually know: the page a careful machine would quote because quoting it is safe. That is a writing and clarity problem before it is a technology problem, which is precisely why small operators are not locked out of it the way they are being locked out of the click.

In ClosingThe legibility test

None of this rewards louder marketing. A machine assembling an answer cannot cite what it cannot parse: unclear claims, inconsistent facts, unstructured pages, a brand story that changes from page to page. Machines can only act on what they can clearly read. That has been our thesis about brands from the first file, and the three numbers above are what it looks like when the reading, not the clicking, becomes the whole game. The 68 percent problem is not that the click died. It is that most businesses are still optimizing for the step that dissolved, and are illegible at the step that replaced it.

The question that matters is no longer "where do we rank." It is "when the machine answers a question we should win, are we in the answer?" You can measure that today: check how visible your brand already is to the machines at /signal-index/, or write to us at /contact/.