AI search statistics 2026: only the ones that check out
AI search statistics for 2026, each checked against its primary source on July 2, 2026 — 68.01% zero-click, 2B AI Overviews users — plus the ones that fail.
Every statistic on this page was checked against its primary source on July 2, 2026. The three most load-bearing: 68.01% of US Google searches now end without a click, AI Overviews reach 2 billion monthly users, and 52% of crawler requests on Cloudflare’s network are for AI training. The famous numbers that failed checking are at the bottom.
Roundup pages list a hundred stats with no dates or methodology. This page runs the other way: fewer numbers — source, date, caveat attached — plus a folklore file of famous figures that fail checking.
How many people use AI search in 2026?
The major AI surfaces now count their users in the hundreds of millions to billions — but user counts say nothing about search volume, so read the last row before quoting the first.
| Stat | Source | Date | Method caveat |
|---|---|---|---|
| ChatGPT: 900M weekly active users | OpenAI, via TechCrunch | Feb 27, 2026 | Weekly actives — not monthly |
| 49% of US adults use AI chatbots, 44% use ChatGPT, 60% read AI search summaries | Pew Research | June 17, 2026 | Self-reported survey, n=5,119 |
| AI Mode: 1B+ monthly active users; queries “more than doubled every quarter since launch” | May 19, 2026 | Google’s own figure; no methodology published | |
| AI Overviews: 2 billion monthly users, driving over 10% more queries for the types of searches that show them | Pichai, Q2 2025 earnings, via SEJ | July 23, 2025 | Earnings-call reported speech; no dataset |
| Gemini app: ~900M monthly active users, up from 750M+ in February | EMARKETER compilation, via ppc.land | May 2026 | Medium confidence — not a Google primary |
| AI Mode’s share of US Google searches: 0.34% (Jan–Apr 2026) | SparkToro | June 9, 2026 | Similarweb US web panel; excludes the Google app |
The last pairing is the finding: AI Mode passed a billion users and was 0.34% of US search volume. Quoting one without the other is half the story — and note the two 900M figures differ: ChatGPT’s is weekly, Gemini’s monthly. Collapsing them is how folklore starts.
How often do AI Overviews appear?
Anywhere from about 16% to “nearly half” of queries — the number depends entirely on whose keyword panel you use.
| Stat | Source | Date | Method caveat |
|---|---|---|---|
| AIOs on 6.49% of tracked queries (Jan 2025) → 24.61% peak (July) → 15.69% (Nov 2025) | Semrush | Dec 15, 2025 refresh | 10M+ US keywords, Datos clickstream |
| AIOs on 20%+ of searches, Jan–Apr 2026 | SparkToro | June 9, 2026 | Similarweb US web panel |
| AIOs on nearly half of tracked keywords after 58% YoY growth; healthcare 88%, restaurants 78% | BrightEdge, via SEJ | Mar 1, 2026 | Commercial tracked keywords, not all searches; medium confidence (secondary) |
These studies sample different query populations — that’s nearly the whole spread. Anyone quoting a single prevalence figure without describing the panel is guessing, which is exactly how the folklore version of the BrightEdge number was born (see below).
What do AI Overviews do to clicks?
Every independent panel measures fewer clicks when an AI Overview is present; Google says clicks are “relatively stable” and has published no data supporting its side.
| Stat | Source | Date | Method caveat |
|---|---|---|---|
| 68.01% of US Google searches ended without a click (Jan–Apr 2026), up from the 60.45% the post cites for 2024 | SparkToro | June 9, 2026 | Similarweb panel, no Google app; Fishkin: cross-year chart “apples and oranges” |
| Position-1 CTR 34.5% lower with an AIO, widening to 58% lower in the re-run | Ahrefs | Apr 2025 → Feb 4, 2026 | 300K keywords, desktop GSC; correlational |
| Users clicked a result on 8% of visits with an AI summary vs 15% without | Pew | July 22, 2025 | 900 US adults, March 2025; Google disputes it |
| Counter-evidence: same-keyword zero-click rates slightly decreased, 33.75% → 31.53% | Semrush | Dec 15, 2025 refresh | 200K+ keyword subset, Jan–Oct 2025; zero-click share, not position-1 CTR |
| Google’s position: click volume “relatively stable year-over-year,” “slightly more quality clicks” | Google, Liz Reid | Aug 6, 2025 | No charts or numbers in the post |
| Our data, n=1: 160 impressions, 4 clicks, position 59.1 in our first two GSC weeks | HZ Signal | June 19–29, 2026 | One site’s experience, not a universal law |
Ahrefs and Semrush don’t contradict — position-1 CTR and zero-click share have different denominators; both verify. Corroborating studies in Ahrefs’ update range from 47.5% to 90% lower. Google’s reply to Pew: “a flawed methodology and skewed queryset” — no counter-data published. The full study-by-study reconciliation lives on our great decoupling page.
Which pages do AI engines actually cite?
Not simply the organic top 10 — and Google’s two AI surfaces don’t even agree with each other: AI Overviews and AI Mode cite the same URLs just 13.7% of the time.
| Stat | Source | Date | Method caveat |
|---|---|---|---|
| AIOs and AI Mode cite the same URLs 13.7% of the time (16.3% top-3), despite highly similar answers (86% average semantic similarity; ~90% of pairs above 0.8) | Ahrefs | Dec 15, 2025 | Sept 2025 data — predates 2026 model changes |
| ~17% of AIO-cited sources rank in the organic top 10 | BrightEdge, via SEJ | Mar 1, 2026 | Tracked commercial keywords; medium confidence |
| 44.2% of ChatGPT citations come from a page’s first 30%; cited content is 2x likelier to include a question and averages 20.6% proper nouns vs a typical 5–8% | Kevin Indig, via SEL | Feb 18, 2026 | 18,012 verified citations, ChatGPT only; correlations, not causes |
| Being cited in an AIO = ~120% more organic clicks per impression; organic CTR 2.36% with an AIO vs 3.82% without | Seer Interactive | Apr 24, 2026 | 5.47M queries, 53 brands, all intents, Feb 2026 |
The top-10 overlap question has its own trajectory and counting-method traps — reconciled in full in our deep dive. For the mechanics, see how AI engines source and cite content and query fan-out.
What are the AI crawlers doing?
Crawling far more than they send back: 52% of crawler requests on Cloudflare’s network are now for AI training, and over half of all internet traffic is non-human.
| Stat | Source | Date | Method caveat |
|---|---|---|---|
| 52% of crawler requests are for AI training (up from 22% in Spring 2025); over 50% of internet traffic is non-human | Cloudflare | July 1, 2026 | Cloudflare’s network — 20%+ of the web |
| Training was nearly 80% of AI-bot crawling; crawl-to-refer ratios: Anthropic ~50,000:1, OpenAI 887:1, Perplexity 118:1 | Cloudflare | Aug 28, 2025 | July–Aug 2025 data; platform-level; different base from row above |
| 35.7% of the top 1,000 websites block GPTBot | Originality.ai | As of Aug 2024 | Deliberate robots.txt blocking; tracker ongoing |
| Reputable news-site AI blocking rose from ~23% (Sept 2023) to nearly 60% (May 2025) | arXiv 2510.10315 | 2025 | Academic; news-site population |
| 7.9M spoofed ChatGPT-User requests — second-most impersonated agent — of 7.9 billion AI-agent requests | DataDome | Jan–Feb 2026 data | One vendor’s network telemetry |
| Our data, n=1: Google indexed 10/16 new pages (median 3 days); Bing crawled 0/16 in 11 days despite IndexNow pings on 8 | HZ Signal | July 2, 2026 | One new domain; live experiment, outcome pending |
The 52% and 80% rows use different denominators — all crawler requests versus AI-bot crawling — never chain them. Blocking is a deliberate publisher choice; we weighed the trade-off with our own robots.txt open.
How much traffic does AI search actually send back?
Growing fast from a small base: total AI referral visits more than tripled in a year — yet Google still accounts for roughly 88% of referral traffic on Cloudflare’s network.
| Stat | Source | Date | Method caveat |
|---|---|---|---|
| Total AI referral visits grew more than 3x, Sept 2024 → Sept 2025 | Similarweb | May 28, 2026 | Clickstream panel; referrals without a referrer header are undercounted |
| Visits to the assistants, Sept 2024 → Mar 2026: ChatGPT +84%, Gemini ~9x, Claude ~+770% | Similarweb | May 28, 2026 | Traffic to the platforms — often misquoted as referrals from them |
| ChatGPT referral traffic converts at 7.1%, second only to paid search’s 7.8% | Similarweb | Apr–May 2026 | One panel, one window |
| After OpenAI’s May 7, 2026 links update: ChatGPT referrals +157.7% week-over-week | Similarweb | May 28, 2026 | Product-change step event — numbers move overnight |
| Google still accounts for ~88% of referral traffic | Cloudflare | July 1, 2026 | Cloudflare network view |
The headline is the tension between rows one and five: AI referrals are tripling and still a rounding error next to Google. Strategies built on only one of those fail differently. To see your own numbers, segment AI traffic in GA4; engine profiles: ChatGPT, Gemini.
The folklore file: famous numbers that fail verification
None of these survived a hunt for a primary source — several never had one. Every roundup we checked repeats at least one.
| The claim | What verification found |
|---|---|
| ”FAQ schema increases AI citations by +28%“ | No named study, dataset, or methodology anywhere — origin untraceable. The only traceable data cuts the other way: SE Ranking (via ZipTie, Mar 2026) found FAQ-schema pages averaged fewer ChatGPT citations (3.6 vs 4.2). Google also removed FAQ rich results on May 7, 2026. We’re testing structure vs citations ourselves — Experiment 02, running now |
| ”Content under 30 days old gets 3.2x more AI citations” | No primary source, and circulating variants contradict each other — a competing figure says AI-cited content averages ~2.9 years old. What’s documented: Google’s date guidance — dates must describe the page’s real publication or update. That’s it |
| ”Query fan-out generates 8–12 sub-queries per search” | Not from Google, which has never published a sub-query count. Its official language is qualitative: “a multitude of queries,” and “hundreds of searches” for Deep Search only. The 8–12 range generalizes single demo anecdotes |
| ”IndexNow gets you indexed in 5–30 minutes” | Vendor marketing. IndexNow’s own FAQ: “Submitting a URL does not guarantee immediate indexing”; Bing’s page promises immediate notification, not crawling. Our data, n=1: 8 pinged URLs, zero Bing crawls in the first 11 days |
| ”AI engines read llms.txt” | No engine has confirmed using it. John Mueller, April 2025: “AFAIK none of the AI services have said they’re using LLMs.TXT (and you can tell when you look at your server logs that they don’t even check for it)”. Google’s AI-features doc, checked July 2, 2026, never mentions it. We publish one anyway, labeled speculative |
| ”AI Overviews appear on 48% of all searches” | Real number, wrong denominator: nearly half of BrightEdge’s tracked commercial keywords, not all searches. BrightEdge’s own May 2025 all-queries figure was over 11%; Semrush’s panel peaked at 24.61% |
| “99% of AI Overview citations come from the top 10” | Citation drift: a per-AIO statistic retold as per-citation. The per-citation figure we could verify runs far lower — ~17% of AIO-cited sources rank in the organic top 10, per BrightEdge. Full reconciliation |
Two more drift cases: Ahrefs’ 34.5% still circulates although its own re-run widened it to 58%, and a “~276 clicks per 1,000 searches” figure attributed to SparkToro appears nowhere in its post. Catch one of ours drifting? Tell us — the correction is the product.
How we keep this page honest
We re-verify quarterly, and dateModified moves only when numbers were actually re-checked or replaced. The cadence tracks the sources’ own cycles — Semrush ~December, Ahrefs ~February, Google I/O in May, Pew and SparkToro ~June, Cloudflare’s bot report ~July 1. Per Google’s publication-dates guidance, dates must describe a real update, not cosmetic freshening. We’ll re-run our own probes too: as of July 2, 2026, the GSC searchAppearance API returns zero rows for our property — no AI-surface breakdown in our own data yet. If a number dies at re-verification, it moves to the folklore file with a note. If the data surprises us, that’s the finding — we report it, we don’t move the line.
Frequently asked questions
What percentage of Google searches show AI Overviews?
How much traffic do AI assistants actually send to websites?
Is the 'FAQ schema gets you 28% more AI citations' stat real?
Does any AI engine read llms.txt?
Sources
- TechCrunch: ChatGPT 900M WAU
- Pew: Americans and AI 2026 · Pew: AI summaries and clicks
- Google: AI Mode insights · Liz Reid on clicks · AI Mode I/O post
- Google Search Central: AI features · publication dates · updates changelog
- SparkToro: 2026 zero-click study
- Ahrefs: AIO CTR update · AIOs vs AI Mode
- Seer Interactive: AIO impact on CTR · Semrush: AI Overviews study · Similarweb: Gen AI stats
- Cloudflare: agentic internet bot report · AI crawler traffic by purpose
- SEJ: Pichai Q2 2025 · BrightEdge data · Mueller on llms.txt
- SEL: Kevin Indig citation study
- Originality.ai tracker · arXiv 2510.10315 · DataDome threat research
- IndexNow FAQ · Bing IndexNow · ZipTie: FAQ schema · ppc.land: Gemini 900M
- HZ Signal first-party: first weeks in Search Console · Google fast, Bing crawled nothing · live index-lag data