AI search statistics 2026: what the data says about ChatGPT, Perplexity and Google AI Overviews

When a Google results page showed an AI summary, US users clicked a regular result link on 8% of visits, versus 15% without one (Pew Research, July 2025). Ahrefs measured 58% lower clicks for the top result (Ahrefs, February 2026). ChatGPT reached 900 million weekly users (TechCrunch, February 2026).

This page collects 15 sourced statistics on AI search. Each one links to the page we opened to check the number. Where sources disagree, we say so. Near the end is a small dataset of our own.

The statistics at a glance

Google AI Overviews

Pew tracked 900 US adults and 68,879 Google searches in March 2025. About one in five searches (18%) produced an AI summary. When one appeared, people clicked a regular link on 8% of visits. Without one, 15%. Clicks on links inside the summary were rare: 1% of visits. Sessions ended on 26% of pages with a summary, versus 16% without (Pew).

Ahrefs compared 300,000 keywords (150,000 with an AI Overview, 150,000 informational keywords without) and found the top result's click-through rate was 58% lower with an AI Overview. Position 2 lost 50.8%. Position 10 lost 19.4% (Ahrefs).

How often do they appear? Sources disagree, because they count different things. Semrush saw AI Overviews on 15.69% of keywords in November 2025, after a peak near 25% in July (Semrush). Similarweb reports them in 43% or more of US search sessions (Similarweb). Keywords and sessions are not the same unit. Treat both as directional.

Semrush also found zero-click rates fell from 33.75% to 31.53% on keywords after an AI Overview appeared. That runs against the Ahrefs result. Different keyword sets, different method.

On citations: in July 2025 Ahrefs found 76.1% of cited pages ranked in the top 10 (Ahrefs). By its March 2026 study of 4 million AI Overview URLs, that was 37.9%. Another 31.2% ranked 11 to 100 and 31.0% ranked beyond 100 (Ahrefs). Page-one rank is a weaker gate than it was.

Google zero-click and AI Mode

SparkToro and Similarweb found 68.01% of US Google searches ended without a click in the first four months of 2026, up from 60.45% in 2024. SparkToro notes the 2024 figure used a different data panel, so part of the jump is measurement. Only 0.34% of searches moved into AI Mode in that window (SparkToro).

ChatGPT

OpenAI said ChatGPT had 900 million weekly active users in February 2026, up from 800 million in October 2025 (TechCrunch). Pew's February 2026 survey found 44% of US adults use ChatGPT, up from 34% in 2025. Pew changed its question wording in 2026 to cover other chatbots too, so some of the rise may be wording (Pew).

Share is moving. Similarweb says ChatGPT's share of generative AI website visits fell from about 76% in June 2025 to about 53% in May 2026, with Gemini gaining (Similarweb).

Perplexity

Perplexity's CEO said the service handled 780 million queries in May 2025, growing more than 20% a month (TechCrunch via NetChoice). That is a company claim from 2025, and we found no newer first-party figure we could open. On referrals, Cloudflare measured Perplexity's crawl-to-refer ratio under 200:1 from September 2025, the lowest of the major AI platforms (Cloudflare Radar).

AI crawlers

Cloudflare found AI bots made up 4.2% of HTML requests across 2025. Googlebot alone made 4.5%. Crawl-to-refer ratios show how little traffic comes back: Anthropic ran at 25,000:1 to 100,000:1 after May, OpenAI peaked at 3,700:1 in March, and Perplexity stayed under 400:1 (Cloudflare Radar).

On blocking, Cloudflare's July 2025 look at top domains found GPTBot disallowed in 7.8% of robots.txt files, and only 37% of the top 10,000 domains had a robots.txt at all (Cloudflare). Our own sample, below, is very different. That gap is the point.

What this means for a small business site

None of this guarantees traffic or citations. It does suggest three plain steps.

1. Let AI crawlers in, unless you have a reason not to. If you want to appear in AI answers, a robots.txt that blocks GPTBot, ClaudeBot or PerplexityBot works against you. Our guide covers how to block (or allow) ChatGPT in robots.txt. Run the free AI visibility tool to see what your site currently says.
2. Write answer-first pages. Ahrefs found cited pages often sit outside the top 10, so a clear page can be picked up without a page-one rank. Put the direct answer in the first lines, with the number and the source beside it.
3. Publish steadily and keep claims sourced. Check each stat against its source before it goes live. That is how we build articles: researched, then checked against saved sources. If you want that handled, see our notes on an AI blog writer for small business.

Our own data: who blocks AI crawlers

The full per-site table and a CSV download are on a separate page: Who blocks AI crawlers? A dated robots.txt snapshot, fetched October 26, 2026.

On 2026-10-25 Paperboy fetched robots.txt from 60 well-known sites. 55 returned one (HTTP 200). Five did not: one timed out, three returned 403 and one returned 418. All percentages use the 55.

27 of 55 sites (49.1%) blocked at least one of the three bots. 16 (29.1%) blocked all three. Five sites block ClaudeBot but not GPTBot.

Read this caveat first. The sample is 55 sites, mostly media and large sites, picked by Paperboy. It is not representative of small business sites. It says nothing about the web as a whole. Compare it with Cloudflare's 7.8% figure above, which covers top domains broadly. Other limits: robots.txt is a request, not enforcement. Our parser counts a bot as blocked only on a Disallow: / rule in a group naming it. It does not apply wildcard inheritance, so "no specific rule" does not prove access. This is one fetch per site, from one network, on one day. The method uses public data only and is repeatable with curl https://SITE/robots.txt.