Everyone agrees AI search is killing the click, and almost nobody agrees on by how much. You will read that 68% of searches are zero-click, that AI Overviews cut clicks by nearly 60%, that AI Mode is 93% zero-click, and that AI assistants send a rounding-error of referral traffic — and these figures are quoted side by side as if they were the same measurement disagreeing, when in fact they are four different measurements that happen to point the same direction. This guide is the field guide to those numbers: what zero-click traffic means in the AI-search era precisely, why the stats scatter so wildly (they count three distinct things — the share of searches that end with no click, the drop in click-through when an AI answer is present, and the trickle of referral traffic AI assistants actually send), and a surface-by-surface map of what each AI surface keeps versus forwards, from the classic SERP through AI Overviews, AI Mode, standalone assistants like ChatGPT and Perplexity, and the agentic frontier where software, not a person, is the visitor. It is honest about the counter-narrative — Google insists total click volume is roughly stable, and the arriving AI visitor appears to convert better than the one you lost — and it ends on the conclusion the numbers force once you stop arguing about the exact percentage: the click is no longer the unit of distribution, the branded impression and the citation are, and that reprices content from cost-per-click to cost-per-impression — which is a production problem a content engine like Kompozy is built to solve.
Everyone agrees AI search is taking the click. Almost nobody agrees on how much — and the disagreement is not really a disagreement. You will see it quoted that 68% of searches are zero-click, that AI Overviews cut clicks by nearly 60%, that AI Mode is 93% zero-click, and that AI assistants send a rounding-error of referral traffic. Those four numbers get stacked next to each other as if they were rival estimates of one thing, which is why the topic feels like statistical chaos. They are not rivals. They are four different measurements, each answering a different question, and they happen to all point the same way: away from the click.
So this guide does two things most zero-click coverage skips. It untangles the numbers — what each one actually counts, so you stop comparing a rate to a decline to a referral share — and it maps, surface by surface, what each AI search environment keeps versus forwards, because "AI search" is not one thing and the click economics of an AI Overview, an AI Mode thread, a ChatGPT answer, and an autonomous agent are wildly different. Then it draws the one conclusion that survives no matter which percentage you trust, and is honest about the counter-case while doing it.
Zero-click search is not new. For years a chunk of Google searches ended with no outbound click because a featured snippet, a knowledge panel, a weather box, or a maps result answered the query on the results page. The classic zero-click search phenomenon was about Google surfacing a direct answer and keeping the user. What changed with AI is the scope of what can be answered in place and the arrival of a second mechanism entirely.
The scope change is that a generative answer can resolve a far wider class of queries than a snippet ever could — not just "what time is the Super Bowl" but "compare the three best options for my situation and tell me why," synthesized from many sources into one response. The genuinely new mechanism is agentic: an AI agent that browses, compares, and completes a task for the user, so that the thing visiting pages is software, not a person, and no human click occurs at any point. Old zero-click withheld the click on simple queries; AI zero-click resolves complex ones and, at the frontier, removes the human from the loop. That agentic layer is its own discipline, covered in AI search and agentic web optimization.
Here is the untangling, because once you see that the famous figures answer different questions, the apparent chaos resolves into a consistent picture. There are three distinct measurements in circulation, and almost every confused argument about zero-click is really two people quoting two of them at each other.
This is the share of all searches that end without any outbound click. SparkToro's analysis of Similarweb clickstream data put it at about 68% of US Google searches in the first four months of 2026, up from roughly 60% in 2024. Read it carefully: this is every search ending without a click to the open web, which includes searches resolved by snippets, maps, knowledge panels, and clicks that stay inside Google's own properties — not only searches resolved by AI. SparkToro is explicit that it does not isolate how much of the rise is specifically AI Overviews. So 68% is the honest headline for "how often does a Google search fail to send anyone to a website," but it is not "how much AI is costing you" — it is broader than that.
This is a different question: when an AI answer does appear, what happens to the clicks a result would otherwise have gotten? Here the measurements cluster but use different methods. Ahrefs, studying position-one rankings, found click-through fell by around a third shortly after AI Overviews rolled out and by nearly 60% in later 2025 data when an Overview was present. Pew Research, tracking real users in 2025, found that on searches with an AI summary only about 8% of visits ended in a click to a traditional result, versus 15% without a summary — and that the links inside the summary itself were clicked on only about 1% of visits. This is the number that actually describes the AI penalty on a given query, and it is conditional: it applies to the roughly 20%-plus of searches where an Overview shows, not to all searches.
The third number is the one people reach for when they ask "how much traffic is ChatGPT sending me," and today the honest answer is: not much, but rising fast and punching above its weight. Analyses through 2025 and into 2026 put referrals from standalone AI assistants — ChatGPT, Perplexity, Gemini, Claude — at a small fraction of a percent of total web visits for most publishers, even as the absolute numbers grew several-fold year over year. ChatGPT dominates that referral mix; Gemini has been climbing. Crucially, several 2026 analyses found these AI-referred visitors convert meaningfully better than general search traffic, plausibly because the assistant pre-qualified them before sending them on. So the referral share is tiny in volume and disproportionate in quality — the opposite profile of the clicks it is replacing.
Stack the three and the confusion dissolves: 68% is how often a search sends no one anywhere, ~60% is how hard an AI answer cuts the click on the queries where it appears, and the referral trickle is what the chat assistants hand back. Different denominators, same conclusion. Anyone arguing that "zero-click is overblown because ChatGPT referrals are tiny" is answering question three to rebut question one, which is a category error.
"AI search" is a bucket term hiding surfaces with completely different click economics. Sorting them from most-forwarding to most-extractive is the single most useful mental model here, because your exposure depends entirely on which surfaces your audience uses.
A traditional ten-blue-links SERP still sends an outbound click on roughly 15% of visits — Pew's baseline for a search with no AI summary present. It is the baseline the AI surfaces are measured against, and it still exists for a large share of queries. This is the surface where the old click economics mostly hold.
An AI Overview sits on top of an otherwise normal results page, so the links are still there, one scroll down. The click survives but is cut hard — Pew's ~8%-versus-15% gap and Ahrefs' near-60% CTR drop both describe this surface. Overviews now appear on more than a fifth of searches and reach billions of users, so this is the surface doing most of the measurable damage to informational traffic today, simply because of its reach. The full stable-traffic-versus-publisher-data debate specific to Overviews is in Google AI Overviews and web traffic.
AI Mode is a fully conversational thread with no results page underneath it in the old sense. A 2025 analysis put it near 93% zero-click — only about 7% of AI Mode searches click out — because its design decomposes your question into sub-queries, answers them in parallel, and lets you go deeper in follow-up turns, resolving a whole research session inside the thread. It is still a small share of total search (a fraction of a percent of searches transitioned into it in early 2026) but passed a billion monthly users within about a year, so it is a trajectory to plan for, not a current emergency. The dedicated breakdown is Google AI Mode and zero-click traffic.
ChatGPT, Perplexity, and the rest are off-Google surfaces entirely: they answer in their own interface and cite sources inconsistently, sending the small-but-high-quality referral trickle described above. The frontier beyond them is agentic — an assistant that doesn't just answer but acts, browsing and transacting for the user. In that case the "visitor" is the agent, and a human click may never happen at all; your page is read by software and the outcome is a completed task, not a session in your analytics. This is the logical endpoint of zero-click, and it changes what "being visible" even means.
A first-mover page an engine might itself cite has to state the other side fairly, because overstating zero-click is its own failure. Two caveats are real. First, Google maintains that total organic click volume to the web is roughly stable year over year, and the logic is not absurd: search volume keeps growing, so a lower click-through rate spread over more searches can net out close to flat in aggregate. The catch is that a stable total masks a savage redistribution. The surviving clicks concentrate on transactional, branded, and navigational queries — someone ready to buy, or searching your name — while informational and how-to content, exactly what an AI answer resolves in place, bleeds its clicks. So "the web's total is stable" and "my explanatory blog lost half its traffic" can both be true, and usually are. This split is the core of SEO in the age of AI Overviews and worth internalizing before you panic or dismiss.
Second, the clicks you do get from AI surfaces appear to be better. The conversion-lift findings on AI-referred visitors suggest the few who arrive are more qualified, so a smaller number of higher-intent sessions can be worth more than the volume it replaced. None of this makes zero-click a non-issue — it reframes it. The problem is not that all traffic vanished; it is that the free, high-volume, informational-query traffic that funded a lot of content businesses is the specific thing being withdrawn, and that it is being withdrawn unevenly. Measuring your own exposure rather than arguing about the web-wide average is the only move that matters, and the per-query method is in how to measure traffic lost to AI Overviews.
Pick whichever percentage you trust — 68, 60, 93, or the referral trickle — and they all deliver the same verdict: the click has stopped being a reliable unit of distribution. For twenty years you could count visits and reason about reach because a ranking turned into a click turned into a session. On a search layer that answers in place or acts through an agent, that chain is broken, and no amount of ranking harder re-forges it, because you can hold position one and still lose the click to the answer sitting above you. What replaces the click as the thing worth counting is two-fold: the citation — being the named source the answer is built on — and the branded impression that citation creates, which returns later as direct visits and branded search even though it sent no click today. Being paid in impressions and citations instead of sessions is the actual economics of zero-click, and the conversion side of it is worked through in AI referrals and the zero-click content strategy.
That verdict has a measurement consequence most teams discover the hard way: the dashboards built around the click will tell you the strategy is failing precisely when it is working, because the metric you are watching is the one you deliberately stopped depending on. Reorienting what you track — toward citations, share of the answer, impressions, and the branded demand that follows — is its own job, and the reason a sensible content program can look like it is dying on a sessions graph while it is actually fine; that measurement break is the whole subject of AI content didn't stop working — your metrics did. And once the click is no longer the unit, the strategic move is to stop fighting for a click the surface has decided to keep and start building distribution across the channels the search layer cannot gate — the portfolio argument in content strategy beyond Google traffic. For the reframe of visibility itself, AI search visibility and zero-click discovery covers why being indexed no longer means being visited.
Follow the conclusion to its blunt operational end. If you are paid in branded impressions and citations rather than clicks, then the number that governs your content economics is no longer cost-per-click — it is cost-per-on-brand-impression: how cheaply you can put a recognizable, on-message piece of your brand in front of a human or into the source pool an answer engine samples. Zero-click doesn't make content worthless; it reprices it. And at a lower price per impression, the winning posture is obvious — mint far more on-brand impressions, across far more surfaces, for the same effort. That is not a strategy problem. It is a throughput problem, and it is the specific thing Kompozy is built to solve.
The leverage is the surface where zero-click quietly works in your favor: native in-feed content. A reel, a carousel, a text post, or a short video earns a human impression right there in the feed, with no outbound click required and no answer engine standing between you and the viewer — the one place the zero-click dynamic is a feature, not a tax. Kompozy is a full AI content generation and multi-platform publishing engine that turns one source idea into 18 native output formats at once — avatar Persona Shorts and clipped video for the feeds, carousels and image and text posts for in-feed reach, plus a blog article and an email newsletter that are text-native and therefore directly readable and citable by a model. One input becomes a week of impressions spread across the eight social platforms plus blog and email, which is how you drive the cost-per-impression down instead of grinding out one piece at a time.
The reason volume doesn't become noise is consistency, which is also what the citation half of the equation rewards: a written Persona Brief governs voice so every piece reads as the same recognizable source, and HyperFrames renders every visual pixel-consistent, so fifty impressions register as fifty touches from one brand rather than fifty strangers — the repetition that converts a zero-click impression into branded recall and later direct demand. Autopilot schedules and publishes the whole set from one queue behind a per-post review gate, so the cadence holds instead of trailing off the first busy week. The boundary, stated plainly because this space overclaims: Kompozy does not measure your citations, read your Search Console impressions, or make Google send back a click it has decided to keep — you still own the measurement and the editorial judgment about what deserves a first-hand take. What it removes is the production ceiling that otherwise makes "just make more on-brand impressions everywhere" a slogan nobody has the hours to execute. In a zero-click world, that ceiling is the whole game.
It is search demand that resolves without any click through to a website. In the AI-search era this happens two ways. First, the answer appears in the search surface itself — a Google AI Overview, an AI Mode thread, a ChatGPT or Perplexity response — and satisfies the query in place, so the user never needs a result. Second, and newer, an agent carries out the task on the user's behalf, so software rather than a person is the only thing that ever touches a page. Either way the query is served but no human visit is generated, which is the distribution problem the term names.
Because the headline figures measure three different things that get quoted as if they were one. The zero-click rate — about 68% of US Google searches in early 2026 — is the share of all searches that end with no outbound click, and it includes searches resolved by featured snippets, maps, and knowledge panels, not only AI. The click-through decline — AI Overviews cutting the top result's clicks by nearly 60% — measures what happens to a result's CTR specifically when an AI answer is present. And AI-referral share — the small-but-growing trickle from ChatGPT, Perplexity, and Gemini — measures what the standalone assistants send you. Three denominators, three questions, one direction.
A classic results page is the most forwarding — Pew found a traditional result gets clicked in about 15% of visits when no AI summary appears. AI Overviews sit above that same results page, so clicks still happen but at a reduced rate — the same Pew study found only about 8% of visits with a summary ended in an organic click. AI Mode is the most extractive mainstream surface: a 2025 analysis put it near 93% zero-click, meaning only about 7% of AI Mode searches clicked out. Standalone assistants send little referral traffic today but it is growing. Agentic browsing is the terminal case — the software is the visitor, so a human click may never occur at all.
Partly, and the nuance matters. Google maintains that total organic click volume to the web is roughly stable year over year, and there is a real argument underneath it: search volume keeps growing, so a lower click-through rate on more searches can net out flat. But a stable total hides a brutal redistribution — the clicks that survive concentrate on transactional, branded, and navigational queries, while informational and how-to content, the kind an AI answer resolves in place, loses its click. So a site heavy in explanatory content can see its own traffic fall sharply even if the web-wide total holds. Average and individual are not the same story.
Two ways that survive the zero-click shift. First, the branded impression: being the named source inside an answer builds recognition that shows up later as direct visits and branded search, even when the answer sent no click. Second, quality over quantity on the clicks that do come — multiple 2026 analyses found visitors arriving from AI assistants convert meaningfully better than general search traffic, plausibly because the assistant pre-qualified them. So the value moved from volume of sessions to the recognition you earn by being cited and the higher intent of the few who still click.
AI search zero-click traffic is search demand that ends with no website click — the AI answer resolves the query in place, or an agent acts for the user. Around 68% of US Google searches end without a click, and AI Overviews cut the top result's click-through nearly 60% when present. The published figures disagree because they measure three different things, so the honest conclusion is that the click is no longer the unit — the branded impression and the citation are.
Get started → · ← All guides · Compare Kompozy vs other tools