Zero-Click AI Answers: Measuring Visibility Without Traffic

In January–April 2026, 68.01% of US Google searches ended without a click — up from 60.45% in 2024 and roughly 49% in 2019, per SparkToro's clickstream study using Similarweb panel data. Those clicks did not move to AI chatbots: across 74,752 sites, all AI chatbots combined sent 0.28% of total traffic in March 2026, while Google Search still sent 28.12%. Do that arithmetic and the strategic problem is obvious — the demand did not relocate to a channel you can report on. So the fix is not a new traffic source. It is a measurement change: chart impressions and clicks as separate series, measure whether you appear inside the answer, and stop leading with click-through rate.

Here is why that last point matters more than it sounds. In Q4 2025, one dataset showed informational CTR on AI Overview queries falling 61%. Clicks went from 398,000 to 400,000. Nothing had gone wrong with the business; everything had gone wrong with the metric. If you own the organic number in a monthly review, that failure mode is the one most likely to cost you a budget conversation this quarter.

Start by fixing the two numbers in your deck

Two figures dominate every zero-click slide, and one of them does not survive contact with a primary source.

The first is "65% of searches end without a click" — accurate for its moment, now behind the data. SparkToro's 2026 read is 68.01%, US Google, browser-based desktop and mobile. Know what that covers before you present it: it assumes a two-thirds mobile split, treats ten seconds of inactivity as a session end, and excludes searches made inside the Google mobile app, where zero-click features are more aggressive still. So 68% is a floor, and a US floor at that.

The second is "AI Overviews push that to 83%." It circulates widely, usually attributed to Similarweb — but Similarweb's own zero-click explainer does not state it. That page offers keyword-level texture instead: "what's the capital of Sweden" at 85% zero-click, "sunglasses" at 77%. Very different from a platform-wide claim. If a number's provenance dead-ends in a stats round-up, don't put it in front of a CFO who can Google it.

A third correction matters more than either: zero-click is not one number. iPullRank's analysis of 13.1 billion search events from 9.1 million opted-in panel users found 41% zero-click in Japan against 54% in the Philippines, and 46% on desktop against 66% on US mobile — while Similarweb's panel reads device the other way round, 79.6% desktop versus 54.8% mobile. Two credible panels, opposite conclusions, different definitions. So never report a blended company-level rate: segment by device, market and query type, or you cannot say which part of your funnel moved. None of this started with AI, either — the same series shows roughly 45% zero-click in 2016 and 64.82% in 2020. AI accelerated a decade-old erosion, which is a far better story for a review than "AI broke our channel."

The clicks did not move to ChatGPT

The most common strategic error right now is assuming displaced clicks resurface as AI referral traffic, then setting a KPI on it.

They don't. In March 2026, across those 74,752 sites, all AI chatbots together sent 3.5 million visitors — ChatGPT 2.7 million of them, with Perplexity, Gemini and Claude in the hundreds of thousands or lower. Over the same ten months, Google's share of those sites' referrals fell from 35.11% (June 2025) to 30.53% (March 2026). Roughly four and a half points of Google referral share left the building; AI referrals picked up a fraction of one point.

The surfaces themselves are enormous, which is what makes this counterintuitive. Google told investors on its Q2 2026 earnings call that AI Mode had crossed a billion users and the Gemini app was at 950 million monthly users. Yet AI Mode was just 0.34% of searches in the SparkToro window, and Similarweb estimates it refers a click on 1.6–2.5% of queries against Google's traditional 17–19%. Huge audience, tiny click yield — which is why 0.28% cannot be your headline KPI. So if clicks are the wrong number to lead with, what is CTR telling you?

Your CTR chart is mostly measuring its denominator

Seer Interactive's 2026 CTR study — 53 brands, 5.47 million queries, 2.43 billion organic impressions, January 2025 through February 2026 — documented informational AI Overview brand-cited CTR "collapsing" 61% quarter over quarter. Then it showed the working: "clicks were essentially flat in October, moving from 398K to 400K. Impressions more than doubled, from 15.8M to 33.1M. CTR compressed because the denominator grew far faster than the numerator."

Read that as a business event and nothing happened. Read it as a KPI and you have a 61% decline in a red cell, in a deck, in front of a CMO. CTR is now a ratio between two series that move independently: impression inflation can crater it while clicks hold, and impression deflation can improve it while clicks fall. The prescription is one line long — never ship a CTR trend without the impression trend beside it, on the same chart. CTR becomes a diagnostic you explain, not a headline you defend.

Search Console makes that harder than it should be. An AI Overview occupies a single SERP position and every link inside it shares that position; a link must also be scrolled or expanded into view to register at all. There is no "AI Overview present" filter in the main performance report, so when impressions jump and average position holds, you cannot attribute the change from that report alone. Say so, rather than guessing.

What actually changed — and the honest counterweight

The strongest evidence here is behavioural rather than modelled. Pew Research tracked real browsing for 900 US adults across 68,879 Google searches in March 2025. Users clicked a traditional result link on 8% of visits to pages with an AI summary, versus 15% without — roughly half. They clicked a link inside the summary on just 1% of those visits, and 26% ended the session entirely, against 16% on conventional results pages. That is not a CTR model; it is people not clicking.

The trend line comes from Ahrefs running the same 300,000-keyword study twice: the first pass found position-1 CTR down 34.5% on AI Overview keywords (March 2024 vs March 2025), and the February 2026 update found 58%, with position-1 CTR falling from 0.073 to 0.016 between December 2023 and December 2025. Same method, two readings, worsening.

Now the counterweight, which belongs in this section rather than a footnote. Google's position is that the premise is wrong: Liz Reid, VP and Head of Search, wrote in August 2025 that "total organic click volume from Google Search to websites has been relatively stable year-over-year" and that "average click quality has increased." No data was published alongside it. Lily Ray of Amsive answered publicly: "Do the hundreds of thousands of Google Search Console screenshots showing impressions remaining flat (or increasing) this year, while clicks dramatically decline… count as 'flawed methodologies' or 'isolated examples'?"

Both statements can be true at once, and that reconciliation is what to bring to your review: Google's claim is about aggregate clicks web-wide, while every third-party study measures per-query CTR. If query volume is growing, total clicks can hold steady while the clicks any single page earns per impression fall. Your site-level report lives in the second world.

One more piece of honesty, because a straight line down is already contradicted by the best-instrumented dataset available: the curve bent. Seer found organic CTR on AI Overview queries recovering from a 1.3% floor in December 2025 to 2.4% in February 2026, an 85% rebound that beat their own forecast two months running. Their published caveat adds that commercial no-AIO CTR also declined meaningfully in 2025 without significant AI Overview exposure — so attributing every organic decline to AI is a claim the data doesn't support.

The traffic you lost probably wasn't your buyers

Seer's AI Overview prevalence by intent is the reframe: 36% of informational queries, 8% of commercial, 5% of transactional. The compression concentrates where in-market buyers are least likely to be. Kerry Cunningham, Chief Research Officer at 6sense, puts the implication plainly: "For companies experiencing a decline in web traffic attributable to LLMs, it is likely the traffic that has gone missing is not from in-market buyers."

There is a genuine tension in the same dataset, though. Sort by query shape rather than intent and the picture inverts: comparison queries ("X vs Y") trigger an AI Overview 95.4% of the time, reviews 86.3%, question formats 85.9%, "best of" 81.3%; single-word queries only 27.3%. Those high-trigger shapes are exactly what a buying committee types at shortlist stage — so low-intent traffic is thinning while your most decision-adjacent queries are the ones most often answered above the links.

Which matches what buyers say they do. 6sense's Buyer Experience Report — roughly 4,000 buyers across North America, APAC, Europe and the UK, median deal value $300–400K — found 94% of B2B buyers use LLMs during the buying journey, and TrustRadius data cited in the same report has 90% clicking through to verify what AI tells them. They use the answer layer as a filter, then verify on your site.

If that is what's happening, sessions should get fewer and better — and where the visits are visible, they do. Adobe Analytics' May 2026 read on US retail found AI-referred visitors converting 54% higher than non-AI traffic, generating 53% more revenue per visit and bouncing 36% less often, with AI-sourced traffic up 138% year over year. So the honest line for your next review is not "traffic is down, ignore it." It is that fewer, better-qualified sessions is a different business than fewer sessions — and you will be asked to prove which one you're in.

Being in the answer still pays — and it is measurable

If the answer layer only destroyed clicks, the rational response would be to disengage. The data says otherwise. Seer quantified the citation premium on identical queries: "Being cited in the AIO delivers +120% more organic clicks per impression versus when you are not cited. But it still underperforms No AIO by -38% in informational queries." In volume: one million informational impressions produced roughly 33,500 organic clicks with no AI Overview, about 20,743 when cited inside one, and about 9,445 when not cited. You cannot control whether an AI Overview appears; you can influence whether you're in it, and that influence is worth about eleven thousand clicks per million impressions. Seer's caveat travels with the number — the "not cited" aggregate leans heavily on one account at 47% of impressions — so treat the ratio as directional.

What correlates with being cited looks unfamiliar to a link-building budget. Ahrefs' Spearman analysis of 75,000 brands found branded web mentions correlating with AI Overview visibility at 0.664, branded anchors 0.527, branded search volume 0.392 — and backlinks last at 0.218, with Ahrefs' own caveat that correlation is not causation. The distribution is the part worth taking to planning: 26% of brands studied had zero AI Overview mentions, and the top quartile by branded web mentions averaged 169 against 14 in the quartile below. That is winner-take-most, which makes share of AI answers a competitive metric rather than a hygiene checkbox.

Baselines vary enormously by sector, so calibrate before anyone calls a low number a failure. Similarweb's May 2026 read on ChatGPT citation presence puts Travel & Hospitality around 23% of relevant answers and Automotive near 20%, while Professional Services sits under 4%. A professional-services brand at 5% and a travel brand at 5% are in completely different competitive positions.

The 2026 measurement stack: what belongs in the monthly report

Two pieces of instrumentation shipped this year that most existing advice predates, and neither is sufficient alone.

Search Console's Generative AI performance report arrived June 3, 2026. It covers AI Overviews and AI Mode across Pages, Countries, Devices and Dates — and per Google's documentation it reports impressions only: no clicks, no CTR, no queries, in limited rollout to a subset of site owners. The data was already inside your aggregate report; this separates it out. Useful, and strictly a Google-surfaces impression counter.

GA4's native "AI Assistant" default channel landed around May 14, 2026, recognising ChatGPT, Gemini and Claude among others under medium ai-assistant. Two limits belong in any report that uses it: it depends on a referrer header, so in-app browsers, copy-pasted links and referrer-stripped sessions still land in Direct, and the reclassification is not retroactive. Report AI referrals as a floor. Our guide to connecting an AI visibility score to GA4 signups covers the channel groups and the attribution gap in detail.

Neither report answers the question your executives are actually asking: does the brand show up when a buyer asks? That third leg has to be built as a poll, not a rank check. Search Engine Land's recommended method is to define 250–500 representative high-intent queries and re-run them repeatedly — "a polling-based model inspired by election forecasting" — so nondeterministic answers average into a stable estimate, then track share of voice against named competitors on that fixed set. Smaller teams can start well below 250; what matters is that the set is fixed. Our method for choosing the 25 prompts you track maps a starter panel to funnel stages rather than to a keyword list — which matters, because 43% of B2B marketers in one 2026 survey said they don't know which buyer prompts to target.

That polling layer is what Geoptimizer was built for: it runs your buyer-intent prompts live on ChatGPT, Gemini, Claude and Grok with web search enabled, then reports mention rate, citation rate, prominence and sentiment as one 0–100 AI Visibility Score with a per-engine breakdown. All four engines are included on every plan, including the free one, so you can put a real number beside your Search Console impressions before deciding whether the category deserves budget.

Two items finish the report. Add self-reported attribution — a "how did you hear about us?" field with explicit AI options — because influence you can't see in GA4 sometimes announces itself on the form. And read branded search and direct traffic as leading indicators, since an uncited answer often resurfaces as someone typing your name a week later. Semrush's ROI sequence also offers a lag expectation worth agreeing with your board up front: 0–30 days for a visibility change, 30–60 for traffic, 60–90 for leads, 90–180 for revenue.

The objections you will get, answered

"Google says clicks are stable." True, and unfalsifiable from outside, since Google publishes no click data. Answer on scope: their claim covers aggregate clicks web-wide, your report covers clicks per impression for your pages. Bring the Pew behavioural data, which measures people rather than models.

"These tools disagree with each other." They do. Digiday collected the complaint in May 2026, including Paul Dyer of /prompt: "If you use three different tools and give them the same prompts, you get three different answers." That's fair, and it has a mechanical explanation: different prompt sets, model versions, grounding configuration and time windows, plus genuine nondeterminism. The response is methodological transparency — a single scan is a snapshot, a defensible headline number is a rolling window with a confidence band, and the weighting should be published so a moving score can be explained. Audit any vendor's method before trusting its number, starting from what generative engine optimization actually measures.

"This is a vanity metric." It is, if you report the score alone. It stops being one when it sits beside impressions, branded search, self-reported attribution and pipeline, with the citation premium as the connecting mechanism. The demand is real, too: in a May 2026 survey of 150 B2B tech marketing leaders, 88% said leadership had asked about AI visibility while only 34% had a defined strategy, with 31% naming measuring success and tracking deal flow as a top hurdle.

"Analysts said search volume would collapse and it didn't." Correct. Gartner predicted in February 2024 that search engine volume would fall 25% by 2026; as of April 2026 it hadn't, with Google still above 90% share. The shift landed in click behaviour, not query volume — which is the argument for instrumenting your own funnel rather than planning off forecasts.

Page one of Monday's report

Five metrics, five sources. Everything else is an appendix.

Metric	Where it comes from	Healthy trend
Search impressions (total + AI surfaces)	GSC performance report + Generative AI report	Flat or up; always charted next to clicks
Search clicks	GSC performance report	Flat or up; read only alongside impressions
Share of AI answers	Fixed prompt panel across ChatGPT, Gemini, Claude, Grok	Rising mention and citation rate vs named competitors
Branded search + direct sessions	GSC queries + GA4	Rising — the leading indicator of uncited influence
Self-reported attribution mentioning AI	Form field / sales-call notes	Rising share of new pipeline

CTR is not on that list. Keep it as a diagnostic you produce on request, with its impression series attached, and explain the denominator every time it moves.

FAQ

Is zero-click really 65%, or 83% with AI Overviews? The best-sourced current figure is 68.01% of US Google searches ending without a click in January–April 2026 (SparkToro, using Similarweb clickstream data, browser-based searches only — the Google mobile app is excluded, so the real rate is likely higher). The widely circulated "83% with AI Overviews" figure does not appear on the primary pages it is usually attributed to, so treat it as unverified. If you need an AI-specific number, use Pew's behavioural finding: 8% of visits produced a result click when an AI summary was present, versus 15% without.

Does AI referral traffic replace the organic clicks we lost? Not at current volumes. Across 74,752 sites in March 2026, all AI chatbots combined were 0.28% of traffic against Google Search's 28.12%. AI referrals are a small, high-quality stream — Adobe measured AI-referred retail visitors converting 54% higher — but they cannot serve as a replacement KPI, and referrer-less sessions mean what you see is a floor.

Can I isolate AI Overview performance in Search Console? Partially. The Generative AI performance report, launched June 3, 2026 and in limited rollout, gives impressions inside AI Overviews and AI Mode by page, country, device and date — impressions only, no clicks, CTR or queries. The main report still has no "AI Overview present" filter, an AI Overview counts as one shared position, and a URL appearing in both the overview and the blue links counts as a single impression.

What do I say when clicks are down but impressions are up? Show both series and explain the denominator — Seer's 61% CTR "collapse" involved clicks moving from 398K to 400K while impressions went from 15.8M to 33.1M. Then add the intent split: AI Overviews appear on about 36% of informational queries versus 5% of transactional ones, so the thinning traffic skews research-stage. Support that with conversion rate and pipeline, not sessions.

The short version

Zero-click is not a story that began in 2024 — it is a decade-old erosion that AI answers accelerated, and 68% of US Google searches now end without a click. Those clicks did not migrate to a channel you can report on, which is why a traffic-only organic report increasingly misrepresents your own performance. Being cited in the answer still pays roughly 2.2x over being absent from it, and the metric to stop leading with is the one that moves when its denominator does.

Rebuild the report around five numbers, label every limitation, and get a real reading of where you appear in AI answers before the next review. Check your visibility across ChatGPT, Gemini, Claude and Grok for free, and if the data earns it, the tracked-prompt plans scale from there.