How many people use AI search
These are the numbers the companies publish themselves, in earnings calls and product announcements, which makes them the most reliable figures here and the least independent.
2 billion
monthly users of AI Overviews, the AI-written summary above Google's normal results, across more than 200 countries and 40 languages.
1 billion
monthly users of AI Mode, Google's chat-style version of search, a year after it went global. It was 100 million in the US and India twelve months earlier.
950 million
monthly users of the Gemini app, with daily users tripling over the year.
900 million
weekly users of ChatGPT when OpenAI last published a figure, in February 2026.
1 in 5
Google searches produced an AI summary, measured from the actual browsing of 900 US adults rather than from a survey.
How much traffic AI search actually sends
This is the number that gets left out of the slide deck, and the one to check before anyone sells you a strategy on the strength of the audience figures above.
0.28%
of all visits to 74,752 tracked websites came from AI chatbots in March 2026. That is 3.5 million visitors in total.
2.7 million
of those 3.5 million came from ChatGPT alone. Perplexity and Gemini sent around 230,000 each and Claude about 102,000.
63%
of 3,000 studied sites had received at least one visitor from an AI chatbot. The average site got 0.17% of its visitors that way.
under 1%
of total web traffic comes from all large language models combined, against 41.35% for Google Search.
Being cited by an AI is not the same as being visited by a human. Most of the value of an AI citation is that your name appears in the answer, whether or not anyone clicks.
What AI answers cost you in clicks
34.5%
lower click-through rate for the top-ranking page when an AI Overview sits above it, measured across 150,000 keywords with one and 150,000 without.
8% vs 15%
how often people clicked a normal search result with an AI summary present against without one. Roughly half as often.
1%
of visits to a page with an AI summary ended in a click on a link inside that summary. Being cited is not the same as being clicked.
26% vs 16%
how often people ended their browsing session entirely after a search page with an AI summary against one with only normal results.
58.5%
of US Google searches ended without a click, from clickstream panel data. Of every 1,000 searches, 360 clicks reached a site Google does not own and is not paid by.
15% to 25%
estimated reduction in organic web traffic from zero-click results. In the same survey, around 80% of consumers relied on those results in at least 40% of their searches.
Whether the traffic that does arrive converts
Both of these come from companies with an interest in the answer, so treat the size of the multiple loosely and the direction seriously.
23x
the conversion rate of ordinary organic search, measured on Ahrefs' own site: AI search was 0.5% of the traffic and 12.1% of the signups.
4.4x
the value of an ordinary organic visit, by conversion rate, across more than 500 marketing and SEO topics.
The likely reason is filtering. The AI answers the casual question, so the person who still clicks through has already read a summary and wants something specific.
Which pages AI engines actually cite
37.1%
of the URLs cited in AI Overviews also ranked in Google's organic top 10 for the same query. Another 26.2% ranked between 11 and 100.
36.7%
did not rank in the top 100 at all. More than a third of what Google's own AI quotes is content its ranked list never shows you.
88%
of the URLs ChatGPT cites are taken directly from its search index rather than from memory or any other retrieval route.
0.656
similarity between a cited page's title and the internal sub-question ChatGPT generated, consistently higher than for the pages it passed over. It rewrites your query before it goes looking.
63%
of nearly 400 million citations across six AI models pointed to listicles, ranked lists of the 'best X' kind. The most-cited pages ran 1,000 to 2,000 words and averaged 18 words a sentence.
Evertune, 25,000 URLs across six models, via Search Engine Land, May 2026
69%
of the time that Google's AI Overview cited a brand's own 'best of' list, it recommended a competitor instead of that brand. Across 100 B2B software queries on three dates.
30% to 50%
of organic visibility lost within weeks by brands publishing self-promotional 'best of' lists, after Google's December 2025 core update. Google has not confirmed a targeted action.
How fresh your content has to be
1,064 days
average age of a page cited by an AI assistant, against 1,432 days for the pages ranking organically on the same queries. That is 25.7% fresher.
Ahrefs, 16.9 million citations across 7 platforms, July 2025
458 days
how much newer the content ChatGPT cites is than what Google ranks for the same queries. It has the strongest freshness bias of the platforms measured.
Ahrefs, 16.9 million citations across 7 platforms, July 2025
How stable an AI citation is
One check of what an AI says about you is a sample, not a measurement, and these are the numbers that show why.
70%
chance that an AI Overview's text changed between one observation and the next, with no algorithm update and no change on anybody's site.
Ahrefs, 43,000 keywords with 16+ observations each over a month, November 2025
2.15 days
average lifespan of a given version of an AI Overview before it was rewritten.
45.5%
of the cited sources were entirely new between consecutive versions. Only 54.5% of the links survived.
0.95
semantic similarity between consecutive versions, where 1.0 is a perfect match. The wording and the sources churn; the meaning holds.
What correlates with being visible to AI
Correlation is not causation, and the study says so itself. It is still the largest published attempt to rank the factors.
0.664
correlation between branded web mentions, your name appearing anywhere on the web with or without a link, and visibility in AI Overviews. The strongest factor measured.
0.218
correlation for number of backlinks, the weakest of the factors near the top. Branded anchors scored 0.527 and branded search volume 0.392.
169 vs 14
average AI Overview mentions for brands in the top quartile for web mentions against the quartile below. More than ten times as many.
What the technical files are worth
97%
of valid llms.txt files received no requests at all in May 2026. Of 137,210 domains studied, 38,360 had a valid file and about 1,100 of those were ever fetched.
1.1%
the share of AI bot requests to those files that came from the retrieval crawlers behind ChatGPT, Perplexity and Claude. Most of the traffic was SEO audit tools.
55.5%
of all websites use JSON-LD, the format structured data is written in. Structured data is a label in your page's code that states what the page is about.
38,065 to 1
pages crawled by Anthropic's bot for every visitor it sent back in July 2025. OpenAI's ratio was 1,091 to 1, Perplexity's 195 to 1 and Google's 5.4 to 1.
We put llms.txt in what we deliver and give it credit in the AEO score. The honest reading of that 97% is that the file is cheap insurance rather than a lever, and we would rather you saw the number than found it later.
What help with this costs
$2,917
average monthly SEO retainer across 439 providers. Agencies averaged $3,209 and freelancers $1,348.63.
63%
of quotes fell between $500 and $5,000 a month. The average hourly rate was $111.