Verified AI Search Statistics, and the Source Behind Each Number

Which AI search statistics can you actually verify?

The AI search statistics worth citing in 2026 come from named organizations that published them themselves: OpenAI, Google, Pew Research Center and Cloudflare. Every figure in the table below carries the organization that published it and the date it was published, because a number without both is a number nobody can check. Read the source note before you quote any of them, since what was measured matters as much as the result.

Statistic	Source	Published	What it measures
Around one in ten adults worldwide had used ChatGPT by July 2025	OpenAI, in research with Harvard economist David Deming (NBER working paper w34255)	September 2025	Adoption against world adult population
Non work messages rose from 53 percent to over 70 percent of ChatGPT messages	OpenAI, same research	September 2025	Share of total messages by purpose
AI Overviews reached over 1.5 billion users a month	Google, Alphabet Q1 2025 earnings call, Sundar Pichai	24 April 2025	Monthly users of the feature
Users clicked a traditional link on 8 percent of visits with an AI summary, against 15 percent without	Pew Research Center	22 July 2025	68,879 searches by 900 US adults, March 2025, results collected April 2025
Users clicked a link inside the AI summary on 1 percent of visits	Pew Research Center	22 July 2025	Same study
18 percent of the Google searches studied produced an AI summary	Pew Research Center	22 July 2025	Same study
34 percent of US adults have used ChatGPT, and 58 percent of adults under 30	Pew Research Center	25 June 2025	Survey of 5,123 US adults, February to March 2025
Anthropic's Claude reached a crawl to refer ratio as high as 500,000 to 1, later settling between roughly 25,000 to 1 and 100,000 to 1	Cloudflare Radar, 2025 Year in Review	Data through early December 2025	HTML page requests by Claude's crawler versus visits it referred
Google Search's crawl to refer ratio ran from about 3 to 1 up to 30 to 1 over the same year	Cloudflare Radar, 2025 Year in Review	Data through early December 2025	Same measurement, Google's own crawler and referrals

How many people use AI assistants?

OpenAI put ChatGPT adoption at around one in ten adults worldwide by July 2025, in research it published in September 2025 with Harvard economist David Deming through the National Bureau of Economic Research. The same research found non work messages rose from 53 percent to over 70 percent of all ChatGPT usage over that period, which is the number to reach for when somebody asks whether the tool moved beyond writing help. Cite the paper directly rather than a summary of it, since the working paper itself states the sample and the method.

Pew Research Center measured the United States separately and found 34 percent of US adults have used ChatGPT, about double the 2023 share, in a survey of 5,123 adults published on 25 June 2025. Among adults under 30 that figure was 58 percent. The two sources use different methods, product research against a probability sample, and they point the same way without depending on each other.

ChatGPT adoption by population

Adults worldwide1 in 10

US adults34%

US adults under 3058%

How big is Google AI Overviews?

Sundar Pichai told Alphabet's Q1 2025 earnings call on 24 April 2025 that AI Overviews had passed 1.5 billion users a month, and that usage kept growing nearly a year after the US launch as people learned Search handled more of their questions.

Treat this as a company reported figure with a stated scope rather than an audited measurement. It is the right size for arguing that AI answers are now a mainstream surface, and the wrong size for estimating what any single site loses, since Google has not published a country or language breakdown alongside it.

What happens to clicks when an AI summary appears?

Pew Research Center found that users clicked a traditional result link on 8 percent of visits where an AI summary appeared, against 15 percent of visits where none did. The study, published on 22 July 2025, tracked 68,879 searches by 900 US adults in March 2025, and found that 18 percent of the searches produced an AI summary at all. Clicks on links inside the summary itself reached 1 percent of visits.

That is the cleanest independent measurement of the effect everyone argues about. Roughly half the click through, from a named research organization, with the sample and the window published alongside it. The behavior it describes is the subject of zero click search, and the response to it is what answer engine optimization exists to organize.

Checking an AI search statistic before you publish it

Find the organization that first published the number

Open that page, not the roundup quoting it

Write down what was measured and over what window

Check whether it is telemetry, a survey or a company statement

Publish the source and the date next to the number

Does traffic from AI assistants convert differently?

The industry-wide conversion percentages that circulate for AI referred traffic mostly trace back to a vendor's own aggregate dashboard rather than a page you can open and check, so treat any single number here with more caution than the figures above. What is verifiable instead is the shape of the traffic itself: a visit that arrives after an assistant compares options for someone has already had the comparison conversation that a first touch search visit has not, so the two are not the same kind of visit even before either converts.

That difference is the argument for tracking where assistants name your brand at all, since the volume is too new and too vendor reported to benchmark against yet. RedReplier tracks where ChatGPT, Claude and Gemini cite a brand in their answers, alongside monitoring Reddit, X, Bluesky, Facebook and Hacker News. See LLM brand monitoring for AI search for that setup, and AI visibility and brand monitoring tools for how the category compares.

What do AI crawlers take compared with what they send back?

Cloudflare measures a crawl to refer ratio for each platform, dividing how often its crawler requests a page by how often it sends a human visitor to that same page through a referral link, and published the figures across 2025 in its Radar Year in Review. Anthropic's Claude carried the highest ratio of any platform Cloudflare named, reaching as much as 500,000 to 1 before settling into a range of roughly 25,000 to 1 to 100,000 to 1, while Google's own search crawler, measured the same way, ran from about 3 to 1 up to 30 to 1 across the year.

That gap between Claude and Google is the clearest available measure of what an AI platform takes against what it sends back, and Cloudflare's own AI Insights page on Radar keeps the ratio updated per platform rather than freezing it at one snapshot. Publishers deciding what to allow should check the current numbers there before setting a policy, and robots.txt covers the mechanics of acting on that decision. Cloudflare's pay-per-crawl program gives a publisher a third option beyond allow or block: charge a crawler per request instead.

How should you use these numbers?

Use each figure for the question its method can answer, and say who published it every time you repeat it. A telemetry number describes the product that reported it, a survey number describes the population it sampled, and a company statement on an earnings call describes what that company chose to disclose. None of the three is interchangeable with the others.

Then measure your own position rather than arguing from the aggregate, because whether an assistant names your brand is something you can check directly. That practice sits inside generative engine optimization, and the strategic split between the two disciplines is covered in AEO versus SEO.

Frequently Asked Questions

What is the most cited AI search statistic right now?

Google's own count of over 1.5 billion monthly users for AI Overviews, disclosed by Sundar Pichai on Alphabet's Q1 2025 earnings call on 24 April 2025. It is the largest verifiable figure tied to any single AI search surface. Quote it as a company reported monthly user count, since Google has not published an independent audit behind it.

Do AI Overviews reduce clicks to websites?

Pew Research Center measured clicks on traditional result links at 8 percent of visits when an AI summary appeared, against 15 percent when none did, in a study published on 22 July 2025. That is roughly half. The study covered 68,879 searches by 900 US adults in March 2025, with results collected that April.

How many people use Google AI Overviews?

Over 1.5 billion a month, according to Sundar Pichai on Alphabet's Q1 2025 earnings call on 24 April 2025. That is a company reported figure with no independently audited breakdown behind it.

Is traffic from AI assistants worth anything?

Nobody has published an audited, independently checkable comparison of AI referral conversion against other channels yet, so treat any specific percentage you see quoted as a vendor's own number until you can verify it. Measure your own AI referred traffic against your own other channels instead of borrowing an industry figure. That is a data problem you can solve with your own analytics regardless of what the wider industry reports.

Why do AI search statistics vary so much between articles?

Because the underlying methods differ and most roundups copy each other rather than the source. A telemetry count, a survey estimate and an earnings call disclosure measure different things over different windows. Two articles quoting different numbers for the same idea are frequently both correct and describing different measurements.

Which AI search statistics should I avoid quoting?

Any figure whose original publisher you cannot name, and any figure you found only in a roundup that links to another roundup. Numbers about AI assistant user counts age fast, so check the publication date before reusing one. If the source page no longer states the figure, stop quoting it.

What is the source of the ChatGPT adoption statistic?

OpenAI published the one in ten adults worldwide figure in research it conducted with Harvard economist David Deming, released through the National Bureau of Economic Research as working paper w34255. The paper appeared in September 2025 and measures adoption against the world's adult population by July 2025. Read the working paper itself rather than a summary, since it states the sample and method directly.

Has ChatGPT usage shifted from work tasks to personal use?

OpenAI's research with David Deming found non work messages rose from 53 percent to over 70 percent of all ChatGPT usage, measured over the same window as the adoption figure above. That's the number to cite when someone asks whether ChatGPT moved beyond office tasks and writing help.

How high is Claude's crawl to refer ratio?

Cloudflare Radar's 2025 Year in Review put Claude's crawl to refer ratio as high as 500,000 to 1 at its peak, before it settled into a range of roughly 25,000 to 1 to 100,000 to 1. That measures how often Claude's crawler requests a page against how often it sends a human visitor there through a referral link. It's the highest ratio Cloudflare named for any platform, well above Google Search's 3 to 1 to 30 to 1 range over the same year.

What percentage of Google searches trigger an AI summary?

Pew Research Center found that 18 percent of the Google searches it studied produced an AI summary. The figure comes from the same study of 68,879 searches by 900 US adults, conducted in March 2025 and published on 22 July 2025. It's the base rate behind the click through comparison: an AI summary appeared on fewer than one in five searches, yet it changed click behavior on those it did.