AI SEO Statistics 2026: Every Figure With Its Sample, Date and Method

There is a genre of blog post that collects forty AI search statistics, attributes none of them properly, and gets cited by everyone. The numbers travel further than their sources, lose their caveats on the way, and end up in pitch decks as settled fact. Half of them contradict each other. Nobody notices, because nobody checks the sample.

This page is the opposite of that. Every figure below carries three things inline: who measured it, on what sample, and when. Where two credible studies disagree, both are shown rather than the more dramatic one. And at the end there is a section listing the widely-quoted figures we went looking for and deliberately did not publish, with the reason.

That last section is the one worth reading if you are short on time. A statistic you cannot trace is not evidence, and the AI search category is currently full of them.

How to read any AI search statistic

Before the numbers, the filter. Nearly every dispute in this category dissolves once you ask three questions, and most published figures fail at least one.

What universe was measured? “AI Overviews appear on 20% of searches” and “AI Overviews appear on 68% of searches” can both be true if one sampled a broad keyword set and the other sampled informational queries in six industries. Coverage figures are almost entirely a function of the keyword list, and the keyword list is rarely published.

Whose traffic was it? Panel-based vendors measure their panel. Analytics vendors measure their customers, who skew toward marketing-aware businesses. A “share of AI referrals” figure drawn from a sample of SEO and marketing sites will overstate AI’s role for a plumber in Ang Mo Kio, because the sample’s audience is the population most likely to be using assistants in the first place.

When was it measured, and how fast does it move? This category has metrics that change by an order of magnitude within a quarter. A ratio quoted without its measurement window is not a statistic, it is a souvenir.

Search behaviour and the zero-click question

68.01% of Google searches ended without a click. SparkToro, analysing Similarweb clickstream data covering US Google searches from January to April 2026. The study excludes searches made in Google’s mobile search app, where SparkToro said zero-click behaviour may be even higher. Its authors caution that because such studies rely on different providers, panels and methodologies, long-term comparisons are not directly equivalent.

0.34% of searches transitioned into AI Mode. Same study, same window. This is the figure that almost never travels with the one above it, and it should. Zero-click search is a decade-old trend driven by featured snippets, knowledge panels, weather boxes and Google’s own properties. In the period measured, AI Mode’s contribution to it was negligible.

Clicked a traditional search result 8% of the time when an AI summary appeared, versus 15% when none did. Session ended on that page 26% of the time versus 16%. Clicked a link inside the AI summary itself: 1%. Pew Research Center, from 68,879 unique Google searches by over 900 US adults in March 2025, collected through its KnowledgePanel. Three caveats belong with it every time: it is a US panel and not Singapore, the data is from March 2025 and the surface has changed since, and it is a correlation on pages where summaries appear, which skew informational and always had lower click-through rates.

AI Overviews appear on more than 20% of Google searches, and click-through rates drop by nearly 60% when they do. Reported by Search Engine Land alongside the SparkToro study. Note that coverage estimates for AI Overviews across 2026 analyses have ranged roughly from the low 20s to the high 40s in percentage terms depending on the keyword universe sampled; there is no single agreed figure, and we have not found one that publishes its keyword list.

Where AI Overview citations come from

Only 38% of pages cited in AI Overviews also rank in the top 10 for the same query. 31.2% rank at positions 11 to 100, and 31.0% rank beyond position 100. Ahrefs, from 863,000 keyword result pages and 4 million AI Overview URLs, March 2026.

That top-10 share was 76% in July 2025. Same source, previous edition of the same study, which analysed roughly 1.9 million citations across a smaller keyword set. The eight-month collapse from 76% to 38% is the single most consequential trend line in this category, and the accepted explanation is query fan-out: the system is running its own related searches rather than citing what is on the visible results page.

88% of AI Overviews cite three or more sources; 1% cite a single source. Ahrefs, same March 2026 dataset.

The top 1% of domains, roughly 12 sites, capture 47% of all citations. Ahrefs, same dataset. This is the concentration figure that matters most for a small business: citation supply is extremely top-heavy, and the realistic goal for most sites is the long tail of specific queries rather than the head.

YouTube is the most-cited domain at 20.9% citation share, up 34% in six months. Ahrefs, same dataset. Worth noting for anyone weighing whether video belongs in a content plan.

Platform share and usage

Gemini app: 950 million monthly active users. Alphabet, Q2 2026 results, 22 July 2026. A primary source, and one of the few genuinely audited numbers in this entire post.

Google Search and other revenue: USD 63.3 billion, up 17% year on year. Total Alphabet revenue USD 119.8 billion, up 24%. Alphabet Q2 2026. This matters for a reason beyond the number itself. Gartner’s widely-circulated prediction that search engine volume would fall 25% by 2026 has now been overtaken by events: a full year into the AI Overviews and AI Mode era, Google’s search advertising revenue grew 17%. If AI answers were destroying search as a commercial channel, this is the line where it would appear first.

ChatGPT’s share of worldwide generative AI web traffic fell from roughly 76% in June 2025 to about 53% in May 2026, while Gemini grew from under 9% to roughly 27-28% and Claude from about 2% to roughly 9%. Similarweb worldwide traffic panel. Note carefully that this measures visits to the assistants, not referrals from them, and panel-based measurement of app-heavy products has known limitations.

Average monthly web visits across AI platforms grew about 70% year on year to 9.5 billion, with unique visitors up around 57% to 655 million. Similarweb, same panel, covering June 2025 to May 2026.

Google holds 92.46% of Singapore search, against Bing at 3.37%, Yahoo at 0.88% and DuckDuckGo at 0.83%. StatCounter, July 2026. StatCounter measures page views on sites carrying its tracking code, which is a real methodology with real bias, but the direction here is not in dispute and the practical implication is: in Singapore, whatever Google does to its results page happens to almost all of your search demand simultaneously.

Citation rates and what they are worth

The rate at which AI answers include a citation rose from 1.6% in June 2025 to 6.8% in May 2026, ranging by category from roughly 23% in travel and hospitality and 20% in automotive to under 4% in professional services. Similarweb. The category spread is the useful part: if you are in professional services, the base rate of getting cited at all is a fraction of what a travel brand sees, and a benchmark drawn from the wrong vertical will make your performance look broken when it is normal.

Roughly 26% of ChatGPT responses contain ads, with about a third of ad impressions appearing in the first response. Similarweb, 2026. Included because it changes the strategic picture: the surface is monetising, which historically has meant organic real estate contracts.

Crawling versus referring

The crawl-to-refer ratio is the most abused statistic in this category, so it needs unusual care.

Cloudflare defines it precisely: dividing the total requests from a platform’s user agents where the response was HTML by the total requests for HTML content where the Referer header contained that platform’s hostname. In plain terms: pages taken, divided by visitors sent back.

Cloudflare’s own published figures for 19-26 June 2025 put Anthropic at 70,900:1 and Mistral at 0.1:1. Those are Cloudflare’s numbers with Cloudflare’s window.

Third-party readings of Cloudflare Radar circulating in 2026 have reported figures such as roughly 23,951:1 for ClaudeBot over January to March 2026, improving to around 11,122:1 for a single week in late May, and around 1,276:1 or 903:1 for OpenAI’s GPTBot depending on the window. We are reporting the existence and spread of these figures rather than asserting any one of them, because they move by an order of magnitude between windows and we have not verified each against Radar directly.

Cloudflare’s own stated limitation applies to all of them: “traffic referred by Claude’s native app does not include a Referer: header, and we believe that the same holds true for traffic generated from other native apps as well”, so the calculations “may overstate the respective ratios, but it is unclear by how much”. A ratio that systematically undercounts its denominator is directional evidence of an imbalance, not a measurement of it.

The constraint that is not a statistic

One thing on this page is not a measurement but governs how you should read all of the measurements: Google’s own published position on what optimising for these surfaces requires.

From Google’s guide to optimising for generative AI features: “From Google Search’s perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO.” On eligibility: “to be eligible to be shown in generative AI features on Google Search, a page must be indexed and eligible to be shown in Google Search with a snippet, fulfilling the Search technical requirements.” On markup: “structured data isn’t required for generative AI search, and there’s no special schema.org markup you need to add.” And on content: “don’t just recycle what others on the internet have already said, or could easily be produced by a generative AI model.”

Hold those quotes next to any statistic being used to sell you a separate AI optimisation discipline. The publisher of the surface says the work is ordinary SEO work. That does not make the numbers above unimportant — the citation-source collapse from 76% to 38% is a genuine strategic shift — but it does mean the correct response is a change in emphasis, not the purchase of a new category. Our explainer on generative engine optimisation works through where that line actually falls.

Singapore-specific figures

There are far fewer of these than the market’s confidence would suggest, and that itself is the finding.

Google: 92.46% of Singapore searches, July 2026 (StatCounter). Repeated here because it is the only Singapore search figure on this page that is directly measured rather than extrapolated.

AI adoption among Singapore SMEs rose from 4.2% to 14.5% in a year, while non-SME adoption rose from 44% to 62.5%. Attributed to IMDA’s Singapore Digital Economy Report, published October 2025 and covering 2024 data. Two cautions: this measures adoption of AI in business operations generally, not AI search behaviour, and we were unable to open the primary PDF directly on this pass, so it is reported as consistently attributed rather than as verified at source. It tells you Singapore firms are adopting AI tools quickly. It tells you nothing about how Singapore consumers search.

Singapore’s digital economy reached S$128.1 billion, or 18.6% of GDP, up from 14.9% in 2019. Same report, same caution.

There is no published Singapore-specific study of AI search click behaviour that we have been able to verify. This is the most important line in this section. Every Singapore percentage you see quoted about AI Overviews, zero-click rates or AI referral share is an extrapolation from a US or global panel. That may still be the best available guide, but it should be labelled, and it usually is not.

Figures we deliberately did not publish

Every roundup should have this section and almost none do. These are claims we encountered, went looking for a source on, and left out.

If a statistic you were about to put in a client deck appears in that table, it is doing less work than you think.

What we would actually do with these numbers

Four conclusions we think the data supports, stated as conclusions rather than facts.

Ranking and being cited have come apart. With 62% of AI Overview citations now going to pages outside the top 10, rank tracking alone no longer describes your visibility. That is an argument for adding citation monitoring to reporting, not for abandoning rankings — which still drive the majority of measurable clicks.

The commercial panic is not supported. Google’s search revenue grew 17% year on year in Q2 2026. AI Mode was a fraction of a percent of searches in the largest 2026 clickstream study. Zero-click rates are high and have been climbing for years for reasons mostly unrelated to generative AI.

Category base rates vary enormously. A citation rate near 23% in travel and under 4% in professional services means cross-industry benchmarks are close to useless. Measure your own baseline before judging performance, which is the argument our guide to tracking AI search traffic makes in detail.

The work is the same work. Content genuinely worth citing, on a site that can be crawled and indexed, covering the full question space rather than one keyword. That is the conclusion Google’s own documentation points to, and it is what our AI SEO strategy guide and the fundamentals in our Singapore SEO guide are built on.

Frequently asked questions

What percentage of Google searches now show an AI Overview?

There is no reliable single answer, and the honest response is a range. 2026 analyses have reported figures from roughly 20% to the high 40s, and a local-business-focused study found much higher rates on informational queries and much lower on transactional ones. The difference is almost entirely driven by which keywords each study sampled, and most do not publish their keyword list. Treat any single confident percentage as a description of that vendor’s keyword set rather than of Google.

Is AI search actually reducing website traffic?

For queries where an AI summary appears, click-through is measurably lower — Pew found 8% versus 15% on a US panel in March 2025. At the aggregate commercial level the picture is different: Alphabet reported Google Search and other revenue up 17% year on year in Q2 2026. Both can be true, because AI summaries concentrate on informational queries that converted poorly anyway. Your own segmented data is more informative than either figure.

Which AI platform sends the most referral traffic?

ChatGPT, on every published analysis we found, but the share varies from roughly 63% to 92% depending on whose sites were measured and over what period. Note also that a large share of assistant traffic arrives with no referrer header at all and is therefore invisible to all of these studies, so the true distribution is unknown.

Do I need schema markup to appear in AI Overviews?

No, according to Google, which states that structured data “isn’t required for generative AI search, and there’s no special schema.org markup you need to add”. Schema remains worth implementing for other reasons, covered in our guide to schema markup for Singapore businesses, but it is not an AI visibility lever and should not be sold as one.

Are there Singapore statistics for AI search?

Very few. StatCounter’s search engine share for Singapore is directly measured. IMDA has published business AI adoption figures, which describe firms using AI tools rather than consumers searching with them. We have not found a verifiable Singapore study of AI search click behaviour, so figures presented as Singapore-specific are almost always extrapolations from US or global panels.

How often should these statistics be refreshed?

The citation-source figures and platform shares move meaningfully each quarter; the crawl-to-refer ratios move faster than that. Anything in this category older than about six months should be re-checked before it goes in front of a client, and anything quoted without a date should be discarded rather than updated.

The short version

The strongest numbers in AI search are narrower and less dramatic than the ones that circulate. Citations have decoupled from rankings, with only 38% of cited pages ranking top 10 against 76% eight months earlier. Google’s search business grew 17% year on year through the AI era. AI Mode was 0.34% of searches in the largest 2026 clickstream study. Referral volumes are small, category-dependent, and systematically undercounted by missing referrer headers.

Everything else is a range, a panel artefact, or an unsourced claim that has been repeated until it sounds like a fact. When you next see a striking AI search statistic, ask the three questions: what universe, whose traffic, and when. Most numbers do not survive it.

If you want a strategy built on figures that hold up rather than on the ones that make the best slide, that is how our AI SEO service works, and our case studies show what we report and how. Talk to us if that is the kind of reporting you want.