Generative Engine Optimisation · Data review
AI search in 2026: the statistics that hold up — and the ones that don’t
The short version
The load-bearing numbers: 68% of US Google searches now end without a click; when an AI Overview appears, organic click-through drops roughly 60%; about 83% of AI citations come from earned media; and Google’s rankings overlap with AI citations by less than 20%. AI-referred visitors convert at a multiple of organic — but the exact multiple varies so much by study that you should measure your own. Every figure below carries its source. The unsourced ones you keep seeing on LinkedIn are in their own section, with reasons to stop repeating them.
Every GEO deck opens with statistics. Most of them are recycled fourth-hand, stripped of their sample sizes, and a few are simply invented. Since we spend our weeks measuring this channel for clients, we keep a working list of which numbers actually survive contact with their primary sources.
This page is that list, published. The rules: every statistic carries who measured it and on what scale. Where good studies disagree, I say so instead of picking the flattering one. And the last section covers the numbers I’d retire entirely. It’s the same standard we applied to tactic claims in the GEO Claims Audit — this is the market-data companion to it.
The click economy is genuinely shrinking
The most consequential shift isn’t happening inside ChatGPT — it’s happening inside Google, where AI answers now sit on top of the results your SEO worked for.
Read together: the surface your traffic used to come from is being replaced by a surface where you’re either named in the answer or invisible. That’s the entire strategic case for GEO in three rows of data.
Who actually gets cited
The second cluster of numbers explains which brands and pages end up inside AI answers — and it’s here that the old SEO intuitions break most completely.
The through-line is one sentence long: engines trust what independent sources say about you, roughly three times more than anything the link graph says. The mechanism behind that — and what a DTC brand does about it — is the subject of why ChatGPT recommends your competitor instead of you.
Do AI visitors actually buy? Yes — with an asterisk
This is the cluster where I most want you to hold the numbers loosely, because the studies agree on direction and disagree wildly on magnitude.
The direction: AI-referred visitors convert meaningfully better than organic visitors. One B2B study across 312 firms measured 14.2% conversion from AI referrals against 2.8% from Google organic. Per-engine breakdowns from other datasets show ChatGPT referrals converting in the mid-teens and Perplexity around ten percent. The mechanism is at least plausible: a visitor who arrives because an engine recommended you has already been qualified and pre-sold in the conversation.
The asterisk: those exact multiples come from different industries, different attribution methods, and in several cases from vendors selling AI-visibility tooling. I’d confidently tell a client “this traffic converts at a multiple of organic.” I would not put “5x” on a slide without measuring their own funnel first — which takes one custom channel group and a checkout survey field. Here’s exactly how to set that up in GA4.
The volume caveat cuts the other way, too: AI referral volume is still small — low single-digit percentages of most sites’ traffic — but it’s the fastest-growing channel in most analytics accounts, and undercounted in all of them, because most AI visits arrive without a referrer and hide in Direct.
Small channel, absurd conversion rate, systematically undercounted. That combination is what “early” looks like.
The engine landscape is consolidating and fragmenting at once
ChatGPT is still the front door of AI search — and its lead is shrinking fast enough that single-engine strategies age badly.
Two practical consequences. First, referral mixes are shifting under everyone’s feet — a strategy tuned only to ChatGPT was already leaking share this year. Second, and less obvious: because engines share only about 11% of their citation domains, the fragmentation multiplies the measurement work rather than the content work. The same authority-building serves every engine; you just have to score each one separately. Our measurement protocol runs four engines for exactly this reason.
The numbers I’d stop repeating
An honest stats page owes you this section. Three families of figures circulate constantly and deserve retirement.
The prophecy stats
“Traditional search will drop 25% by 2026.” “50% of search will be AI by 2027.” These began life as analyst predictions — Gartner’s among them — and got laundered into present-tense facts by repetition. Predictions aren’t measurements. The measured numbers above are dramatic enough without borrowing from the future.
The suspiciously clean conversion multiples
“AI traffic converts 9x better.” Which industry? Whose attribution? What sample? A multiple quoted without those three answers is marketing, and — as covered above — the honest version (“a multiple, measure your own”) is still a strong story.
The vendor coincidences
Any statistic that happens to prove you need the vendor’s dashboard, quoted only by that vendor, sourced to their own “internal data.” Not always wrong — but unverifiable, and the incentive structure writes the headline before the data does. We ran 24 of the most-repeated tactic claims through primary sources in the GEO Claims Audit; fifteen were false. The pattern generalises: be most suspicious of the numbers that sell the easiest.
How to actually use this page
If you’re building a case internally: the click-economy cluster makes the argument that waiting is a decision, the citation cluster tells you where the work is (third-party surfaces, not your homepage), and the conversion cluster justifies the effort per visitor. Cite the sources, not this page — that’s the standard we’d want applied to us.
If you’re deciding what to do next: your own numbers beat all of these. An afternoon of running real buyer prompts across four engines tells you more about your situation than any industry aggregate — that’s the first thing we do in every engagement, and the free AI visibility audit is us doing it for you. This page gets updated as better data lands; the date at the top is the promise.
FAQ
What percentage of Google searches end without a click in 2026?
Around 68% of US Google searches now end without a click to an external site, up from about 60% two years earlier. Less than a third of searches still send a click to the open web.
How much do Google AI Overviews reduce click-through rate?
When an AI Overview appears, organic click-through drops by roughly 60%. Organic CTR on queries with AI Overviews bottomed around 1.3% in late 2025 and recovered only to about 2.4% by early 2026, versus roughly 3.3% on queries without one.
Where do AI engines get their citations from?
Overwhelmingly from earned media — coverage written by third parties — which accounts for roughly 83% of AI citations in Muck Rack’s 25-million-link analysis. Branded mentions in authoritative publications correlate with AI visibility at 0.664 in Ahrefs’ 75,000-brand study, versus 0.218 for backlinks.
Does AI traffic convert better than organic search traffic?
Multiple independent studies point the same direction: AI-referred visitors convert at several times the rate of organic visitors, because they arrive pre-sold by a recommendation. The exact multiples vary widely by study and industry, so treat any specific figure as directional and measure your own funnel.
Is ChatGPT still the biggest AI search engine in 2026?
Yes, but its dominance is eroding: ChatGPT’s share of generative-AI web traffic has fallen from roughly three-quarters to around half over the past year, while Gemini has climbed past a quarter. The engines cite substantially different sources — roughly 11% domain overlap — so visibility must be measured per engine.
Which AI search statistics should I be skeptical of?
Unsourced prediction stats (“X% of search will be AI by 2027”), precise conversion multiples quoted without sample sizes, and any vendor statistic that happens to justify the vendor’s product. Ask three questions of every number: who measured it, on what sample, and when.