AI Search Statistics: Numbers Worth Citing

Short answer

AI search statistics measure how many people use AI answer engines and what that does to clicks. OpenAI reports 900 million weekly ChatGPT users, Google says AI Overviews reach 2.5 billion people a month, Pew found Google clicks fall from 15% to 8% when an AI summary shows, and Conductor puts AI referrals at about 1% of site traffic.

Key takeaways

ChatGPT reached 900 million weekly users in February 2026. Google counts the Gemini app in monthly users instead: 950 million in Q2 2026.
Google says AI Overviews reach 2.5 billion people a month and AI Mode passed 1 billion monthly users (May 2026).
Ahrefs links an AI Overview to a 58% lower click-through rate for the top-ranking page (December 2025 data).
In B2B, ChatGPT's share of AI referrals fell from 89.1% to 62.6% while Claude's rose from 1.4% to 18.5% (Goodie, 2026).
Anthropic's bots crawled about 38,000 pages for every visitor they referred in July 2025. Perplexity's ratio was about 195 to 1 (Cloudflare).

How we pick statistics

Every number here links to the study or announcement it came from and carries the period its data covers. The period matters because a click-through figure from March 2025 and one from December 2025 describe two different versions of Google.

Company figures come from the company itself (OpenAI, Google, Cloudflare, Adobe). Where we couldn't reach the primary page, we name the outlet that reported it. Most click and citation research is published by firms that sell SEO or AI visibility software, so we only include a vendor study when it states its sample size and time window, and we name the vendor in the sentence so you can weigh it yourself.

Almost all of this research is observational. When a study shows a correlation, we call it a correlation. Numbers we couldn't trace to a source with a stated method were cut. Terms like zero-click search, AI Overviews, AI Mode and citation are defined in the glossary.

How many people use AI search?

ChatGPT is the biggest standalone assistant. Google still puts AI answers in front of more people, because AI Overviews sit on top of ordinary results and nobody has to opt in.

OpenAI says 900 million people used ChatGPT each week in February 2026, up from 800 million in October 2025, as reported by TechCrunch.1
2.5 billion prompts a day went to ChatGPT in July 2025, roughly 330 million of them from US users, per figures OpenAI gave Axios.2
The Gemini app had 950 million monthly users in the second quarter of 2026, Sundar Pichai said in Alphabet's Q2 2026 earnings remarks.3
2.5 billion people use AI Overviews every month, according to Pichai's Google I/O 2026 keynote in May 2026.4
At the same event, Google said AI Mode had passed 1 billion monthly active users about a year after launch.4
Perplexity handled 780 million queries in May 2025 and was growing more than 20% month over month, CEO Aravind Srinivas told Bloomberg's Tech Summit.5
34% of US adults had used ChatGPT by early 2025, about double the 2023 share, in a Pew Research Center survey of 5,123 adults.6
In the same Pew survey, 26% had used it to learn something new, against 8% in March 2023.6
McKinsey's AI Discovery Survey of 1,927 consumers, fielded in August 2025, puts AI-powered search use at about 50% of US consumers.7
Nearly 80% of ChatGPT conversations fall into three topics (practical guidance, seeking information and writing), according to "How People Use ChatGPT", an NBER working paper by OpenAI and Harvard researchers.8
By mid-2025, more than 70% of consumer ChatGPT messages had nothing to do with work, up from 53% a year earlier, the NBER paper found.8
Semrush clickstream data on nearly 69 million US desktop sessions shows AI Mode at about 1% of Google search sessions by early July 2025, up from 0.25% in early May (Semrush).9

Our take: the units don't match

The headline figures don't share a unit. ChatGPT reports weekly users, Gemini and AI Mode report monthly users, Perplexity reports queries, and Pew counts anyone who has ever tried the product. Weekly is the stricter count, so 900 million weekly ChatGPT users is a bigger audience than it looks next to Gemini's 950 million monthly.

AI Mode shows the same gap from the other side. One billion monthly users in May 2026 sits next to a 1% share of desktop sessions in July 2025, and both can be true: a monthly user can be someone who tried it once. Query counts compare more cleanly. Perplexity's 780 million queries in May 2025 is about 25 million a day, while ChatGPT took 2.5 billion prompts a day two months later, roughly a hundredfold gap. If you're testing one assistant first for a B2B brand, test ChatGPT. Google's AI reach comes bundled into Search, so you check it on the queries you already track.

How often do AI Overviews appear?

Somewhere between 15% and 25% of US searches, depending on whose queries you count and when. The three sources below are all US data and still differ by almost 10 points.

Across the more than 10 million keywords in Semrush's AI Overviews study, 24.61% triggered an AI Overview at the July 2025 peak, up from 6.49% in January. By November 2025 the rate had settled at 15.69%.10
Conductor's 2026 AEO/GEO Benchmarks Report found AI Overviews on 25% of about 21.9 million US Google searches between September 15 and October 12, 2025. In health care it was 48.75%.11
Pew Research Center followed the real searches of 900 US adults in March 2025 and found an AI summary on 18% of them (Pew).12
Query length was the strongest pattern in Pew's data: searches of 10 or more words got a summary 53% of the time, one- or two-word searches only 8%.12

Our take: why 18% and 25% are both right

These studies measure different things, so 18% and 25% can both be right. Semrush counts keywords in a tracking set, each weighted the same whether it's searched ten times a month or ten million. Pew counts what real people typed, short navigational searches included, and its data is from March 2025, four months before the peak Semrush recorded. Conductor's window is four weeks in autumn 2025, after that peak, with results broken out by industry.

The part all three agree on is the shape: coverage climbs with query length and informational intent. For planning, assume long, question-style queries in your topic show an AI Overview about half the time, and check your own tracked keywords before you trust any site-wide average.

Do AI Overviews and AI Mode reduce clicks?

Yes. Every study we found shows fewer clicks when an AI answer appears. They disagree on how many fewer, because they count different clicks over different periods, and all of them are correlational.

When a Google results page showed an AI summary, users clicked a regular result on 8% of visits, against 15% when there was no summary. That's Pew Research Center's analysis of March 2025 browsing data.12
In Pew's data, links inside the summary itself got a click on just 1% of those visits.12
Pew also found that 26% of visits to a page with an AI summary ended the browsing session, against 16% without one.12
Ahrefs' study of 300,000 keywords associated an AI Overview with a 58% lower average click-through rate for the top-ranking page, comparing December 2025 with December 2023.13
Ahrefs' first version of that study, in April 2025, put the drop at 34.5%.13
On informational queries with an AI Overview, organic CTR went from 1.76% to 0.61% between June 2024 and September 2025, a fall Seer Interactive reports as 61% (3,119 queries across 42 organizations).14
Seer found paid CTR on the same queries fell 68%, from 19.7% to 6.34%.14
Brands cited inside an AI Overview got 35% more organic clicks and 91% more paid clicks than brands that weren't, in Seer's Q3 2025 averages. Cited brands may simply be stronger brands, so read it as a correlation.14
In AI Mode, 92% to 94% of sessions ended without a visit to any outside website in mid-2025, per Semrush clickstream data.9
69% of news-related Google searches ended without a click to a news site in May 2025, up from 56% a year earlier, according to Similarweb data reported by TechCrunch.15
Human pageviews on Wikipedia were 8% lower in recent months of 2025 than in the same months of 2024. The Wikimedia Foundation puts that down to generative AI and social media.16
One result runs the other way. For keywords that started showing AI Overviews in 2025, Semrush found the zero-click rate dipped slightly, from 33.75% to 31.53%.10

Share of visits to a Google results page, with and without an AI summary. Bars to scale (full track = 30%). Source: Pew Research Center, browsing data from 900 US adults, March 2025.12

Our take: much of the decline was coming anyway

The direction is settled. The size depends on what's counted. Pew's 15% to 8% is a drop of almost half, measured across every result on the page for real users in a single month. Ahrefs' 58% is narrower: clicks on the one top-ranking result, December 2025 against December 2023, so it also absorbs two years of everything else that changed on Google. Seer looks only at informational queries, the most exposed type. (Seer labels its organic drop 61%; its published endpoints, 1.76% and 0.61%, work out to 65%.)

The most useful number for a business case is one Seer reports alongside: informational queries without an AI Overview also lost 41% of their organic CTR over the same tracking period, from 2.72% to 1.62%. A large share of the decline would have happened anyway. Semrush's small zero-click dip points the same way, since AI Overviews tend to land on queries that already produced few clicks. Our position: plan for informational traffic to keep falling whether or not you're cited, and judge success on commercial queries and on being named in the answer.

Which sources do AI engines cite?

Each engine has favorite sources, and they overlap with Google's top 10 far less than most teams assume. Watch the denominators: "share of all citations" and "share of the top 10 sources" are different measures, and Profound reports both. The mechanics are in how AI engines choose sources.

Wikipedia made up 47.9% of ChatGPT's top-10 cited sources, and 7.8% of all its citations, in Profound's analysis of 680 million citations from August 2024 to June 2025.17
In the same Profound data, Perplexity's leader was Reddit, at 46.7% of the top 10 and 6.6% of all citations.17
Profound found Google AI Overviews spread citations much more thinly: Reddit led there with only 2.2% of citations, followed by YouTube at 1.9%.17
Only 12% of URLs cited by ChatGPT, Gemini and Copilot also ranked in Google's top 10 for the same prompt, in an Ahrefs study of about 15,000 prompts published in August 2025.18
In the same Ahrefs study, Perplexity tracked Google most closely, with 28.6% of its cited URLs in the top 10.18
Ahrefs also found that about 80% of the URLs AI assistants cited didn't rank anywhere in Google for the original query.18
AI Mode and AI Overviews gave answers that were 86% semantically similar for the same query, yet only 13.7% of their cited URLs overlapped, per Ahrefs' comparison of 540,000 query pairs in September 2025.19
Ahrefs found AI Mode leans harder on Wikipedia: 28.9% of its citations, against 18.1% for AI Overviews.19
Citation mixes can swing within weeks. In Semrush's study of 230,000 prompts, the share of ChatGPT responses citing Reddit dropped from about 60% to about 10% between early August and mid-September 2025, and Wikipedia went from about 55% to under 20%.20
Branded web mentions correlated with AI Overview visibility at 0.664. Backlinks managed 0.218. That's from Ahrefs' study of 75,000 brands.21
Government sites made up 6% of the sources cited in Google's AI summaries, three times their 2% share of links in standard results (Pew, March 2025).12
When the Tow Center for Digital Journalism asked eight AI search tools to identify the source of news excerpts, they got more than 60% of 1,600 queries wrong. Perplexity's error rate was 37%; Grok 3's was 94%.22
In an AskedAbout API test of 144 answers in August 2026, Perplexity returned 20 citations in 34 of 36 answers. ChatGPT averaged 7. Claude returned none in 16 of 36.23

Our take: date the data before you quote it

Profound and Semrush look like they contradict each other on ChatGPT, and the dates explain most of it. Profound's window ends in June 2025. Semrush's starts in July and catches a sharp fall in ChatGPT's Reddit and Wikipedia citing between August and September. Treat any "ChatGPT loves Wikipedia" claim built on data from before late summer 2025 as history. The denominator matters too: 47.9% of the top 10 is only 7.8% of total citations, because the long tail is enormous.

The steadier finding is the low overlap with Google. Engines cite pages Google doesn't rank, and Google's own two AI surfaces share about one cited URL in seven. Put that next to the 0.664 brand-mention correlation and ranking work alone looks insufficient. You need to be named on the pages engines pull from: reviews, comparisons, forums and reference sites. One caution: that correlation spans 75,000 brands, and big brands have more mentions and more of everything else.

For brand-level data, Arobis AI, the agency that publishes AI Ranked, runs standardized buying-intent prompts through ChatGPT, Gemini, Claude, Perplexity and Copilot and reports which SaaS vendors get recommended by category in its State of AI Search Visibility report.

How much traffic does AI search send?

About 1% of total site traffic, but the split between engines moved fast in 2026. GA4 lumps Google AI Mode and AI Overviews visits in with organic Google traffic, so none of the shares below include them. To set up your own tracking, see how to measure AI visibility, or compare trackers in our roundup of AI visibility tools.

ChatGPT's share of B2B AI referral traffic fell to 62.6% in March and April 2026, from 89.1% in May to August 2025, in Goodie's 2026 AI search traffic report.24
In Goodie's data, Claude went the other way, from 1.4% in mid-2025 to 18.5% in March and April 2026. Goodie's panel averages GA4 data per brand.24
Goodie puts Gemini at 10.6% of B2B AI referrals in the same period, Perplexity at 7.3% and Copilot at 4.0%.24
Across 3.3 billion sessions from May to September 2025, Conductor found AI referrals made up 1.08% of all website traffic, and 2.80% in information technology.11
In Conductor's data, ChatGPT sent 87.4% of that AI referral traffic on average.11
At Ahrefs, the 0.5% of visitors who arrived from AI search over 30 days in mid-2025 produced 12.1% of signups, according to the company's own analytics. It's one site, so read it as a well-documented anecdote.25
Traffic from generative AI tools to US retail sites was up 693.4% in the 2025 holiday season over the year before (Adobe Analytics).26
Those visitors were 9% less likely to convert than other retail visitors in February 2025, a gap that had been 43% in July 2024, according to Adobe data on more than 1 trillion visits.27
Similarweb data shows ChatGPT sent news sites 25 million visits from January to May 2025, against under 1 million in the same months of 2024.15

Share of B2B AI referral traffic by engine, brand-averaged GA4 panel. Bars to scale (full track = 100%). Source: Goodie, 2026 AI Search Traffic Report.24

Our take: ChatGPT is no longer all of AI search

Conductor's 87.4% for ChatGPT and Goodie's 62.6% are both right for their samples. Conductor covers May to September 2025, before Claude's share grew, and it counts sessions, so big sites dominate the total. Goodie is B2B only, averages per brand (a small SaaS site counts as much as a large one), and its later wave is from spring 2026. B2B buyers using Claude at work will show up in a panel like Goodie's long before they move a cross-industry total.

The conversion numbers split the same way. Ahrefs' 12.1% of signups from 0.5% of visitors works out to roughly 24 times the site's average signup rate per visit, but that's free signups on a single SEO software site. Adobe measures retail purchases and still found AI visitors converting slightly worse. For B2B software, our advice is to track AI referrals separately by engine, judge them on pipeline, and stop reporting "ChatGPT" as shorthand for all of AI search.

How much do AI crawlers take, and what do they send back?

Far more than they return. The crawl-to-refer ratio counts how many pages a company's bots fetch for each visitor its products send back. Few big sites push back: in AI Ranked's study of 261 major websites' robots.txt files, 21% block at least one AI search crawler and OpenAI's OAI-SearchBot is blocked least, on 11.5%. To decide what to allow, see every AI crawler user agent, then test your file with the robots.txt AI checker.

Anthropic's ratio was 38,065 to 1 in July 2025, down from 286,930 to 1 in January, per Cloudflare's crawl-to-click analysis.28
Cloudflare put OpenAI's ratio at 1,091 to 1, a little better than 1,217 to 1 in January.28
Perplexity's ratio moved the other way in Cloudflare's data, from 54.6 to 1 in January to 194.8 to 1 in July.28
Training accounted for 80% of AI crawling in the 12 months to July 2025, Cloudflare found. Search was 18% and user-triggered actions 2%.28
OpenAI's GPTBot made up 28.1% of AI-only crawler traffic in July 2025, up from 11.9% a year earlier, per Cloudflare.28
AI bots sent 4.2% of HTML requests across Cloudflare's network in 2025, close to Googlebot's 4.5% on its own, per the 2025 Cloudflare Radar Year in Review.29
Crawling triggered by user actions, an AI agent fetching a page to answer someone's question, grew more than 15x during 2025, according to Cloudflare Radar.29
On publisher sites, AI bots made 1 visit for every 31 human visits in Q4 2025, up from 1 in 200 in Q1, according to TollBit's State of the Bots report.30

Crawl-to-refer ratio (pages crawled per visitor referred), July 2025. Bars drawn to a linear scale, which is why OpenAI and Perplexity barely register. Source: Cloudflare.28

Our take: the ratio shows why a bot crawls

A high ratio mostly tells you what a company crawls for. Training crawls never send anyone back, and they were 80% of all AI crawling, which goes a long way toward explaining Anthropic's number. Perplexity crawls largely to answer live searches, so its ratio stays in the hundreds. Anthropic's ratio also fell 87% between January and July 2025, meaning Claude was sending far more visits per crawl, and that fits Goodie's later finding that Claude's share of B2B referrals grew about thirteenfold.

The user-action number is the one to watch, because those fetches come from a person waiting on an answer. Our rule of thumb: blocking training crawlers costs you no referrals today (it may cost you presence in future models, which none of these studies measure). Blocking the search and user-triggered agents, such as OAI-SearchBot and PerplexityBot, removes you from the answers that do send visits.

How many companies invest in GEO and AEO?

Most search teams now have some budget for it. The samples are small, drawn from people who already work in search, and two come from vendors in this market, so read them as direction rather than market size.

Enterprises put 12% of digital marketing budgets into AEO and GEO on average in 2025, and 94% plan to spend more in 2026, according to Conductor's survey of more than 250 enterprise executives.31
SEOFOMO's survey of 171 professionals in 36 countries found 69% had budget for AI search optimization in 2026, up from 38% in 2025.32
SEOFOMO also found that 92% of those professionals track AI visibility and citations for at least some sites, and 98% have been asked about AI search visibility by a client or manager.32
In Minuttia and Growth Memo's State of AEO survey of 599 professionals, 82% of organizations had implemented AEO to some degree, though about 40% had no dedicated AEO budget.33
26.7% of organizations in the Minuttia survey reported a drop of more than 20% in organic traffic and click-through rate.33
Gartner predicted in February 2024 that traditional search engine volume would fall 25% by 2026 as chatbots and virtual agents take queries. It's a forecast, and nothing on this page measures total search volume.34
McKinsey projects $750 billion in US revenue will flow through AI-powered search by 2028 and warns that unprepared brands could lose 20% to 50% of their traffic from traditional search.7

Our take: "budget" means different things

SEOFOMO's 69% with budget and Minuttia's 40% with no dedicated budget aren't in conflict. One asks whether any budget is allocated, the other whether there's a line item of its own, and plenty of teams fund AEO out of the SEO budget. Conductor's 12% average comes from a company that sells AEO software and surveyed enterprise executives only, so treat it as the top of the market.

The figure we'd lean on is Minuttia's 26.7%. A quarter of practitioners reporting organic drops above 20% is about what the click studies above predict. As for Gartner's 25%, Google's own numbers suggest much of that activity moved into AI answers inside Google rather than away from it. Measure your clicks, not search volume.

What does the research say works?

Citing sources, quoting people and adding statistics, according to the best-known controlled test, the GEO paper by Princeton and IIT Delhi researchers on generative engine optimization. To apply it, read what is generative engine optimization.

Adding citations, quotations and statistics raised a source's visibility in generated answers by up to 40% on a 10,000-query benchmark, according to "GEO: Generative Engine Optimization" (KDD 2024).35
In the GEO paper, pages ranked fifth gained 115.1% from citing sources, while the top-ranked page lost 30.3% on average.35
The GEO paper scored keyword stuffing at 17.8 on its position-adjusted word count metric, below the unoptimized baseline of 19.5. It was the only classic SEO tactic tested that made results worse.35
On 200 real queries the GEO authors ran through Perplexity, quotations lifted visibility 22% and statistics up to 37%.35

Our take: GEO gains are relative

Two details get less attention than the 40% headline. The gains are relative: when the fifth-ranked page gained 115.1%, the top page lost 30.3%, because an answer only has so much room. GEO moves visibility around inside an answer, so a competitor doing the same work cancels part of yours. And the real-world check was 200 queries on one engine, so the exact percentages are directional.

The method holds up better than the numbers. Sources, quotes and figures make a page easy to check and easy to lift into an answer, and Seer's data shows brands cited inside an AI Overview get 35% more organic clicks. That's the case for doing it.

Frequently asked questions

How many people use ChatGPT in 2026?

OpenAI said ChatGPT had 900 million weekly active users in February 2026, up from 800 million in October 2025, along with 50 million paying subscribers. ChatGPT is reported in weekly users while Google reports Gemini in monthly users (950 million in Q2 2026), so the two headline numbers are not directly comparable.

Do Google AI Overviews reduce clicks to websites?

Yes. Pew Research Center found users clicked a regular result on 8% of visits to a Google results page with an AI summary, against 15% without one, in March 2025 browsing data. Ahrefs found AI Overviews were associated with a 58% lower click-through rate for the top-ranking page in December 2025. Both are correlations, but every major study points the same way.

What share of website traffic comes from AI search?

About 1%. Conductor measured AI referrals at 1.08% of all website traffic across 3.3 billion sessions between May and September 2025, with ChatGPT supplying 87.4% of it. Shares are higher in some sectors, such as 2.80% in information technology. Google AI Mode and AI Overviews visits are not counted, because analytics tools group them with organic Google traffic.

Which websites do AI engines cite most?

Wikipedia and Reddit, though each engine has its own favorite. Profound's analysis of 680 million citations found Wikipedia made up 47.9% of ChatGPT's top-10 sources, while Reddit made up 46.7% of Perplexity's and was the most-cited domain in Google AI Overviews. Engines also cite pages outside Google's top results: Ahrefs found only 12% of URLs cited by ChatGPT, Gemini and Copilot rank in Google's top 10.

Do AI crawlers send traffic back to websites?

Very little compared with how much they crawl. Cloudflare measured Anthropic's bots making about 38,000 crawls per referred visitor in July 2025, OpenAI about 1,091 and Perplexity about 195. Over the 12 months to July 2025, 80% of AI crawling was for training, which sends no visitors at all, while 18% was for search.

Does generative engine optimization actually work?

The best controlled evidence comes from the GEO paper by Princeton and IIT Delhi researchers, published at KDD 2024. On a 10,000-query benchmark, adding citations, quotations and statistics raised a source's visibility in generated answers by up to 40%, and the gains were largest for lower-ranked pages. Keyword stuffing performed worse than doing nothing.

Sources

Cite this page

AI Ranked Editorial. "AI search statistics: the numbers worth citing." AI Ranked, October 11, 2026. https://airanked.ai/research/ai-search-statistics