AI search statistics, checked at the source.

Google says AI Overviews has more than 2.5 billion monthly active users.

Many of the same people use AI Mode, so the two counts shouldn’t be added.

Google says AI Mode passed 1 billion monthly active users within a year of launch.

In July 2026 Google said it had merged AI Overviews and AI Mode into one search experience.

Google says the Gemini app has 950 million monthly active users.

The app alone, not AI Overviews or AI Mode. Google says its daily users tripled in a year.

OpenAI says more than 1 billion people use ChatGPT each week.

It counts all use of ChatGPT, not only search.

About four in ten U.S. adults say they use AI chatbots to search for information.

Self-reported, and the base is all U.S. adults. About half (49%) say they use chatbots at all.

Six in ten U.S. adults say they read the AI summaries at the top of search results; three in ten say they don’t.

Reading a summary says nothing about clicking anything afterwards.

Across six countries, 54% of people say they saw an AI-generated answer to one of their searches in the past week.

Lowest in France (29%), where Google’s AI Overviews hadn’t launched when the survey ran.

When an AI Overview appeared, users made 39.8% fewer organic clicks to other websites, in an experiment that randomly hid Overviews from half of them.

A working paper, not yet peer-reviewed. Searches with no click at all rose 34.5%.

Sending all of people’s searches to Google’s AI Mode cut their click-through to websites by 18.8 percentage points, compared with standard Google Search.

Forced full adoption: before the study, participants used AI Mode for 0.6% of their searches.

Google users clicked a traditional search result on 8% of visits to results pages with an AI summary, and on 15% of visits to pages without one.

An association, not a measured effect. They clicked a link inside the summary itself on 1% of visits.

The top-ranking page’s average click-through rate was 58% lower on keywords that showed an AI Overview.

A correlation, as Ahrefs says, for the first position on desktop only.

AI Overviews appeared on 36% of informational queries, 8% of commercial queries and 5% of transactional ones.

An agency’s client query sets, not a sample of all searches.

68.01% of U.S. Google searches in January–April 2026 ended without a click.

Up from 60.45% in 2024, though the two years come from different panels, which SparkToro flags.

AI referrals made up 1.08% of all website traffic across 1,215 enterprise sites in ten industries; information technology sites were highest, at 2.8%.

Enterprise customers’ sites, not the whole web.

Visits to U.S. retail sites that came from AI tools converted to purchases at a 60% higher rate than other traffic in July 2026.

Retail only. On U.S. travel sites the same month, AI visits converted 1% lower than other traffic.

On Ahrefs’ own site, AI search visitors were 0.5% of traffic but brought 12.1% of signups over 30 days.

One company’s site, and signups aren’t revenue.

OpenAI’s crawlers requested about 364 pages from sites on Cloudflare for every visit OpenAI sent back to them; Anthropic’s ratio was about 568 to 1, Perplexity’s about 2,200 to 1 and Google’s 4.6 to 1.

Sites on Cloudflare only, and the ratios swing from month to month.

Roughly 57% of 11.84 billion AI citations pointed to company websites: any company’s own site, competitors’ included.

Measured on the prompts Profound tracks, not a random sample of what people ask.

ChatGPT drew 47% of its citations from company websites, the lowest share of the major engines Profound tracked; on Gemini, company websites made up 69%.

Company websites here are any company’s own site, competitors’ included, on the prompts Profound tracks.

For early-stage B2B software questions, 85% of brand mentions in AI answers came from third-party sites; 13.2% came from the brand’s own site.

AirOps doesn’t say when the queries ran.

38% of the pages cited in Google’s AI Overviews also ranked in the top 10 results for the same query.

Ahrefs’ July 2025 study found 76%, but it counted only the top three citations in each Overview, so the two aren’t a like-for-like drop.

ChatGPT cited 15% of the pages it retrieved to answer 15,000 prompts, and left 85% uncited.

AirOps doesn’t say which ChatGPT model it used, or when.

How often a brand was mentioned across the web correlated with how often AI answers mentioned it: 0.709 in AI Mode, 0.664 in ChatGPT and 0.656 in AI Overviews.

A correlation, not a cause, as Ahrefs says itself.

More than 87% of ChatGPT search’s citations matched Bing’s top organic results for the same question; 56% matched Google’s.

A small, early sample, and ChatGPT’s search has changed since.

Asked for brand recommendations 100 times, ChatGPT and Google’s AI had less than a 1 in 100 chance of giving the same list in any two answers.

Run with a researcher from Gumshoe.ai, an AI-tracking startup, which SparkToro discloses. Claude repeated itself slightly more often.

Google’s AI Mode replaced 56% of its cited sources every week; ChatGPT’s search replaced as much as 74%.

Measured by domain. AI Overviews were steadier: their sources stayed the same for just over half of prompts.

Measuring a site’s share of citations to within a five-point confidence interval took about 40–50 queries on Gemini, about 100 on Perplexity and 150 or more on SearchGPT.

The author works at IQRush, which sells measurement verification for AI search.

In repeated runs of the same query, the sites two answers cited overlapped by a median of about 0.3 on Gemini, 0.33–0.40 on SearchGPT and 0.5 on Perplexity, on a scale where 1 means the same sites.

Measured by domain (the Jaccard index). The author works at IQRush, which sells measurement verification for AI search.

The IAB’s measurement framework treats fewer than 50 queries per measurement program as exploratory rather than directional.

It sorts AI visibility into presence, prominence, portrayal and persuasion.

Journalists found at least one significant issue in 45% of AI assistants’ answers about the news.

News questions only. Sourcing was the biggest problem: 31% of answers had serious sourcing issues.

About 91% of Google’s AI Overviews answered a benchmark of factual questions correctly, but only 39% were both correct and fully supported by the sources they cited.

A benchmark of factual questions, not a sample of real searches. Oumi makes models that check AI answers.

82% of B2B software buyers sourced software recommendations from an AI chatbot in the last 24 months.

In the same survey, review sites (38%) edged past AI chatbots (37%) as the top source shaping shortlists.

51% of B2B software buyers start their research with an AI chatbot more often than with Google.

Starting there isn’t replacing Google: 80% still use Google somewhere in their buying journey.

94% of B2B buyers used large language models at some point in their buying journey.

Use peaked mid-journey: 57% used them in the middle, 45% at the start and 15% or fewer near the end.

The winning vendor was already on the buyer’s Day One shortlist 95% of the time.

Four in five deals still went to the buyer’s favorite before first contact.

72% of B2B tech buyers who used AI in a purchase fact-checked its answers very often or always; 94% did at least some of the time.

63% of buyers used AI at some point in the purchase.

69% of B2B buyers prefer to validate AI-generated insights with a sales rep.

A stated preference. 45% used generative AI in a recent purchase, mostly to gather information on vendors and products.

None of the seven major AI crawlers Vercel and MERJ named ran JavaScript: OAI-SearchBot, ChatGPT-User, GPTBot, ClaudeBot, Meta-ExternalAgent, Bytespider and PerplexityBot.

Google’s Gemini, through Googlebot, and AppleBot did render JavaScript.

ChatGPT spent 34.82% of its fetches on pages that returned a 404 error; Googlebot spent 8.22%.

On the sites Vercel monitored, not the whole web. Claude’s rate was 34.16%.

52% of crawler requests on Cloudflare’s network were for AI training in June 2026.

A share of requests by stated purpose, not of all traffic. Cloudflare Radar showed 44.1% for September 1–28, 2026.

AI bots made 5.2% of requests for HTML pages on Cloudflare’s network in September 2026; people made 39.3%.

Other bots made 52%, and mixed-purpose crawlers another 3.5%.

17% of sites on Cloudflare switch on some way to block AI training; fewer than 1% block search crawlers.

Cloudflare doesn’t say how many sites it counted, or on what date.

3.9% of the most-visited sites served Markdown when an AI agent asked for it, and 4% declared their AI-use preferences in robots.txt.

78% had a robots.txt file, mostly written for search crawlers.

Rewriting a source with quotations, statistics or citations raised its share of a generated answer by up to 40% on the study’s own benchmark.

A maximum, on a simulated engine that already had the page in front of it, measuring share of answer text rather than citations or clicks.

Keyword-stuffed sources scored 19.8 against a 19.8 baseline in the GEO study’s appendix table: no measurable gain, and no measurable loss.

The appendix table, with its error bars across five runs. Quotations scored 27.1 and statistics 25.5 in the same table.

With the passage holding the answer in the middle of its input, GPT-3.5-Turbo did worse at multi-document questions than its 56.1% with no documents at all.

Models use what sits at the start or end of their input best. Position here is in the prompt, not on your page.

In 252,000 controlled trials across six AI models, topic match and being listed first mattered most to which source got cited first; a stated price and a recent date helped too, and formatting-only edits had little effect.

No live search: the two sources were handed to the model directly.

In February 2024, Gartner predicted traditional search engine volume would drop 25% by 2026.

2026 has arrived. In July 2026 Alphabet said Google Search usage hit an all-time high; we found no independent measure of total search volume to check against.

Gartner predicts organic search traffic will fall 50% by 2028 as answer engines take its place.

The prediction opens the summary of a hiring template for answer engine optimization managers.

McKinsey projects that $750 billion in U.S. revenue will flow through AI-powered search by 2028.

The same article estimates unprepared brands could lose 20–50% of their traffic from traditional search, with no timeframe given.

Semrush projects that AI search could send more visitors than traditional search to digital marketing and SEO topics by early 2028.

For marketing and SEO topics only, and Semrush sells AI visibility tools.