Blog · Data and trends

Zero click, and what replaces the click, checked in your logs

Published 29 May 2026 · Updated 8 October 2026 · 8 min read

Zero click, and what replaces the click, checked in your logs
Photograph: JL Cabrera / unsplash

The short answer: Zero click means the answer is delivered on the results surface, so the visit never happens. What replaces the click is the mention: your name, inside somebody else's sentence, read by a buyer you will never see in your analytics. The work shifts from winning a position to being the source a model quotes, and the measurement shifts from sessions to named appearances with correct claims attached.

The phrase has been around since featured snippets, and for a decade it mostly described a nuisance: weather, conversions, opening hours, questions nobody was going to buy anything after asking. Generative answers moved it up the funnel. The question that used to produce ten links and a comparison afternoon now produces a paragraph with three brands in it.

The number that is worth quoting, and what it actually measures

Pew Research Center tracked the real browsing of 900 United States adults through March 2025, covering 68,879 Google searches. On result pages carrying an AI summary, users clicked a traditional result in 8 percent of visits. On pages without one, 15 percent. Clicks on the sources cited inside the summary happened in about 1 percent of visits. Of the 68,879 searches, 12,593 produced an AI summary, so roughly one search in five, on that panel, in that month.

Read that carefully, because it is routinely misread. It is not a claim that your traffic halves. It is a per-visit click rate on a mixed basket of queries, measured on one engine, in one country, in one month, before AI Mode was broadly available. The honest reading is narrow and still damning: when a summary appears, the page below it gets engaged with about half as often, and the citation links are decorative rather than a traffic channel. If you are assembling a slide, the discipline of AI search statistics worth quoting is to carry the sample, the surface and the month alongside the figure.

What replaces the click is a mention inside a generated sentence

Three things are produced by an answer that produces no visit.

A named appearance. Your brand enters the consideration set without a session. The buyer now holds an opinion formed from a sentence you did not write, assembled from sources you may not have chosen. If that sentence contains a stale price or the wrong category, the damage is real and silent.

A deferred, higher-intent visit. People who do arrive have already been pre-qualified by the answer. Google's own documentation on AI features says that when people click from result pages carrying an AI Overview, those clicks are higher quality, meaning users are more likely to spend more time on the site. That is Google's characterisation of Google's surface, with no underlying numbers published and no equivalent statement from the other assistants, so it is a claim to test against your own analytics rather than one to repeat. The practical consequence is that a traffic number falling while pipeline holds is a normal shape, not a contradiction, and do AI referrals convert better is the right place to argue about by how much.

A retrieval event you can see in logs. The model fetched something. That fetch is recorded on your server even when no human follows. It is the only part of the sequence you own outright.

The mechanism that makes impressions unreadable

One typed question is not one retrieval. Google's Search Console documentation describes AI Mode as grouping the user's question into subtopics and searching for each one simultaneously. The other assistants do not document their fan-out this way, and from outside we can only infer it from the shape of the requests that reach a server, so treat AI Mode as the documented case and the rest as a reasonable guess. Where it holds, a buyer asking which tool fits a forty-person finance team can trigger a search for pricing, a search for integrations, a search for a compliance term, and a search for a competitor's name.

Two consequences follow, and both wreck naive counting.

Your page can be fetched four times for one human question, which inflates any crawl-derived estimate of demand. Your page can also satisfy the sub-query about pricing while a rival satisfies the one about integrations, so both of you appear in a single paragraph and neither of you receives a visit. There is no position, no impression share and no CTR underneath this. The arithmetic that held search reporting together does not survive contact with it, which is the subject of attribution when there is no click.

A check you can run this week on your own logs

Pick the last thirty days of raw access logs. You want two counts, from the same window.

First, requests from assistant user agents. Count hits where the user agent matches the documented crawler strings for the major engines, separating the training-oriented fetchers from the live retrieval ones where the documentation distinguishes them. Group by URL path.

Second, sessions with an AI assistant referrer. Count visits whose referrer host belongs to an assistant or AI search surface. Group by landing path.

Divide the first by the second, per path. The ratio is the number of machine fetches your content absorbs per human visit it earns back. We run this check inside audits, on whatever log window the client can hand over, and the pattern we keep seeing is that the ratio is both high and very uneven between page types on the same site. We have not published a distribution for it, because we have no standing sample: the sites are self selected, the log windows differ, and we have never measured the same set twice. Read what follows as a way to interpret your own ratio, not as a benchmark to compare against. High ratio with no mentions in your prompt runs means you are feeding the answer and not being named in it, which is a corroboration and identity problem rather than an access one. Low ratio on a page you expect to be cited usually means the page is not being retrieved at all, which is an access problem and is the cheaper of the two to fix.

The referrer half of this check undercounts. The documented browser default is strict-origin-when-cross-origin, which sends only the origin on a cross origin navigation, so an assistant referral arrives without its path. Whether any given assistant surface strips more than that is not something we can read from outside, and we have not tested each one, so we assume some referrals land in the direct bucket rather than asserting how many. The ratio is a direction, not an accounting figure, and you should say so out loud when you present it. The sample audit report shows how we attach the evidence to a finding like this so the number can be argued with rather than believed.

What to do differently, in order of payback

Make the answer liftable. If the buyer's question is answered in a forty word passage near the top of the page, with the entity named inside the passage rather than three paragraphs up, you are a cheaper source for the model to use. This is the whole of what is answer engine optimisation compressed into one habit.

Fix the facts that travel. A price, a headcount, a category label, a founding year. These are the details we most often find repeated back to us wrongly in audit runs, and a wrong one tends to survive several re-runs rather than correct itself. Where your own pages contradict each other, we cannot predict which version an engine takes, and in our runs it has not reliably been the newest one.

Change the report before the quarter ends. Replace sessions as the headline with named appearances across a fixed prompt set, the correctness of the sentence you appear in, and assisted pipeline from an on-site question asking how the visitor heard about you. The last one is self-reported and noisy. It is also, for now, the only instrument that sees the conversation that happened without you.

Honest limits

The Pew figures are one engine, one country, a 900 person panel and one month, March 2025, before AI Mode was broadly available, and we have found no per-intent breakdown for commercial research queries, which is the cohort most readers of this post actually care about. The fan-out described here is documented for Google's AI Mode only; we are inferring it elsewhere from server logs, which show us requests and not the reasoning behind them. The log ratio check has no benchmark attached because we have no standing sample to produce one, and the referrer side of it undercounts by an amount we cannot measure. Google's claim that AI Overview clicks are higher quality is Google's own, unaccompanied by data, and we are not in a position to audit it.

FAQ

Does zero click mean search traffic is finished? No. It means a slice of informational demand is now answered on the surface. Transactional and navigational intent still clicks, and queries with no AI summary still behave roughly as they did. The loss is concentrated, not uniform, so measure it per page rather than per site.

Can I count impressions in an AI answer? On Google, yes. Its Generative AI performance report, available to all sites since 31 August 2026, reports impressions for AI Overviews and AI Mode on their own, drawn from the same Web search type that carries them in the overall report. It gives impressions and not clicks, and we have found no equivalent from the other assistants. What you can count everywhere is your appearance rate across a prompt set you control, re-run on a schedule.

Should I block AI crawlers to protect my content? Where an engine documents that a named fetcher is what supplies its answers, blocking that fetcher takes you out of those answers. For a paywalled archive that may be correct. For a product site it usually trades a visible problem for an invisible one.

Is a mention worth anything if nobody clicks? It is worth what a mention in a trusted recommendation has always been worth, which is considerable and hard to invoice for. The measurable proxy is whether buyers arrive already using your vocabulary, and whether your name appears in answers to the questions that precede a purchase.

Sources

Related reading

AI search statistics worth quotingDo AI referrals convert betterAttribution when there is no clickWhat is answer engine optimisationHow buyers are actually using AI searchCitation churn, and how fast the cited set changes