Statistics about AI search are everywhere, and many are quoted without their context. A number that is true for one dataset gets repeated as a universal fact, then rounded, then turned into a headline. If you are deciding where to spend limited time, that is a poor foundation.
By the iSuggest.ai Team ยท Updated for 2026
This post collects the main figures we rely on, names where each comes from, and says plainly what each one can and cannot tell you. We have kept the framing of every figure intact and avoided inventing any number of our own. Where a source is a vendor or industry compilation, treat the result as direction rather than precision. If you are new to the field, what GEO is gives the background.
How to read a statistic honestly
Before the numbers, four questions are worth asking about any statistic you meet, including ours.
- What was measured? A percentage of queries, of clicks, of citations or of sessions are four different things.
- In which sample? A panel of tracked queries, a set of large sites, one country and one industry can each give different results.
- Over what period? AI search is moving fast; a figure from a year ago may already be stale.
- Who published it, and what do they sell? A source can be accurate and still have a reason to emphasise a result. That does not make it wrong, but it justifies a second look.
With that in mind, here are the figures, grouped by what they say.
Statistic 1: Zero-click searches
One compilation reports that roughly 68% of US Google queries end without a click to an external website (Digital Applied). In plain terms, most searches are answered on the results page itself, or abandoned, and never reach a site.
What it tells you
That clicks are scarcer than rankings suggest. Many queries are resolved by an answer box, a map, a knowledge panel or an AI summary, so a first-page position is worth less than it used to be for informational questions.
What it does not tell you
It does not say that your traffic fell by 68%, or that every kind of query behaves the same way. Transactional and local queries behave differently from simple factual ones, and "zero-click" includes people who got what they needed and people who gave up. Our post on zero-click searches and how to still get traffic goes into what to do about it.
Statistic 2: AI Overviews coverage
An industry compilation reports that AI Overviews appear on about 48% of tracked queries (SEO Sherpa). The same area of research reports that click-through rates drop by roughly 34% to 61% when an AI Overview is present, depending on the study.
What it tells you
That a large share of the queries people care about now have a generated answer at the top, and that where one is present, fewer people click the classic results beneath it.
What it does not tell you
The 48% refers to tracked queries, which are chosen by the vendor and do not represent every search. The share varies by topic, intent and region. And the range of 34% to 61% is wide because studies differ in method; none of them can tell you the effect on your particular pages. See how to get featured in Google AI Overviews for the practical side.
Statistic 3: AI referral traffic growth
Industry analysis reports that traffic referred by AI search engines grew about 16x between 2024 and 2026 (SEO Sherpa). That is dramatic, but it starts from a small base.
What it tells you
That visitors arriving from assistants are a real and growing channel, not a rounding error.
What it does not tell you
Growth of 16x from a small base can still be a small share of total traffic. For many sites, AI referrals are still a modest slice. The sensible reading is "this is growing fast and worth preparing for", not "this will replace your other channels next quarter". If you want to see your own numbers, our guide to measuring AI search traffic in Google Analytics shows how.
Statistic 4: The assistants are not growing equally
A PPC Land analysis reports that ChatGPT's absolute traffic still grew 27% year on year while Gemini grew 231% and Claude grew 320%, with ChatGPT's share of AI referral traffic falling as others rose (PPC Land). Separately, SEO Sherpa reports that in B2B, ChatGPT's referral share fell from 89% to 63% while Claude's rose from 1.4% to 18.5%.
What it tells you
That the market is diversifying. Different audiences use different assistants, and B2B audiences in particular appear to be spreading across more of them.
What it does not tell you
Percentage growth from a small base is easy to inflate; 320% growth of a small number is still a small number. And these figures describe referral traffic, meaning clicks that reached sites, not how many people use each assistant overall. The practical lesson is to avoid tuning for one assistant, a point we make in how to get recommended by Claude and how to rank in ChatGPT.
Statistic 5: Brand mentions versus backlinks
One analysis reports that brand mentions correlate with AI search visibility about three times more strongly than backlinks, with correlation coefficients of 0.664 and 0.218 respectively (Media Officers).
What it tells you
That being talked about, across the web, tracks with being surfaced by AI systems more closely than the traditional link count does.
What it does not tell you
It is a correlation. It does not prove that adding mentions will cause visibility, and well-known brands tend to have both. It also does not mean backlinks are worthless; they still help conventional search. Treat it as a reason to invest in credible coverage alongside, not instead of, everything else. The deeper discussion is in brand mentions versus backlinks.
Statistic 6: Where AI citations come from
A Meltwater study covering more than 8 million citations across eight language models reports that earned and news media accounted for 37.6% of citations, while press releases accounted for only 0.2% (Meltwater).
What it tells you
That independent journalism and editorial coverage are a major source of what language models cite, and that self-published distribution barely registers by comparison.
What it does not tell you
It says nothing about your category specifically, and a vendor of media monitoring has an interest in the value of media. It also does not say that you should stop publishing on your own site; first-party pages still matter, particularly for facts only you can state. See earned media versus press releases.
Statistic 7: Crawler blocking
As reported by LLM Pulse, over 35% of the top 1,000 websites block at least one major AI crawler.
What it tells you
That blocking is common, even among large and sophisticated sites. Some are choosing to opt out; others are probably blocking by accident through defaults and old rules.
What it does not tell you
It does not say which of those blocks are intentional, and it covers only the biggest sites. Your own situation can only be learned by checking, and our guide on how to check if GPTBot can crawl your website shows how.
What the numbers add up to
Put the figures together and a consistent picture emerges, even allowing for the caveats on each one.
- More queries are answered before anyone clicks.
- Where an AI answer appears, fewer people click the standard results.
- AI-referred traffic is growing quickly, and spreading across several assistants.
- Independent mentions and editorial coverage line up with visibility more than self-published announcements do.
- Many sites are unreachable to AI crawlers, sometimes unintentionally.
None of this means search engine optimisation is dead. It means that ranking is no longer the only thing to optimise for, a theme we develop in GEO versus SEO: ranking is not enough.
What to do with this information
Statistics are useful only if they change what you do. These are the actions that follow most directly from the numbers above, in order of how much they tend to unblock.
- Make sure you can be read. Check robots.txt, firewall settings and that your main text is in the HTML. Access comes before everything else.
- Answer questions directly. Put clear answers near the top of your pages so that both people and models can use them.
- Describe yourself consistently. Use one name and one description everywhere so that mentions add up to one entity.
- Earn independent mentions. Offer genuine data, expertise or stories to people who cover your field.
- Measure your own traffic. Your analytics are a better guide to your business than any industry average.
- Keep re-checking. These numbers will shift, and so will your own results. See how often to re-audit your website.
A worked example: turning a statistic into a plan
Suppose you run a small company that publishes buying guides. This is an illustration, not a case study. You read that AI Overviews appear on a large share of tracked queries and that clicks fall when they do. Rather than panic, you ask a narrower question: for my ten most important queries, does an AI answer appear, and am I named in it?
You check by hand and record the results in a sheet with the date. Three queries show an AI answer that mentions a competitor but not you. For each, you read the cited page and note what it does that yours does not: a clearer opening, a comparison table, or a stated update date. You revise your own pages accordingly, check that crawlers can reach them, and re-test in a month.
That is the whole method: use the statistic to choose where to look, then use your own evidence to decide what to change.
Common misreadings to avoid
"68% zero-click means my traffic will fall by two thirds"
It describes the share of searches that end without a click across a large sample. Your traffic depends on which queries you win and what you offer beyond a quick answer.
"16x growth means AI is already my biggest channel"
Fast growth from a small base is still small in absolute terms. Check your own analytics before reallocating anything.
"Mentions beat backlinks, so stop link building"
A correlation is not a switch to flip. Both help; the study suggests mentions deserve more attention than they usually get.
"My competitor appears, so a trick must exist"
Usually the difference is plain: a clearer page, easier access, or more independent coverage. Be suspicious of anyone who sells a secret.
Where to go from here
The most reliable statistic is the one you collect about your own site. A free iSuggest.ai audit checks the technical and structural factors behind many of the figures above, such as access, rendering, markup and identity, and you can read what an audit measures for the full list. Audits are free and not metered, so you can record a baseline today and repeat it as often as you like.
When your pages are in good shape, you can publish a factual snapshot to the iSuggest.ai AI directory, with pages for ChatGPT, Claude and Gemini. Publishing spends credits, which are non-refundable, and improves findability without guaranteeing a citation. See how it works, run an audit on your homepage, and sign in at your account page when you are ready.