AI visibility
AI search statistics, each with its source and sample
These are the figures on AI search we were able to trace to a source that states its sample and date. Each row says what was counted, how many, when, and who published it. Most come from companies that sell AI visibility software, ours included, so read each as that company's measurement of its own data and not as a law.
How to read this page
A statistic is only as good as its denominator. "71% of citations" and "71% of cited domains" are different claims, and we found the first form quoted for a study that says the second. Where the way a figure is usually repeated differs from what its source says, the row gives what the source says.
We read every source on 14 November 2026.
What AI answers cite
| Figure | What it counts | Sample and date | Source |
|---|---|---|---|
| 74% | Share of cited sources that were blog content (listicles 30%, blog posts 29%, comparisons 14%) | 2,671 sources from 96 questions put to ChatGPT, Perplexity, Gemini and Claude, 20 October 2026 | Outrank |
| 17% to 94% | The same share by assistant: ChatGPT 17%, Gemini 74%, Perplexity 85%, Claude 94% | Same study | Outrank |
| 53.7% | Share of citations going to company-owned pages. News and media: 16.9% | 6,095,520 citations, 17 US industries, 7 engines, July to August 2026 | Otterly |
| 38% | Share of Google AI Overview citations that point at the first 100 words of a page | 100,000 citation placements from 10,000 responses, 2 June 2026. Stated as correlational | Surfer |
The spread in the second row matters more than the average in the first. "AI answers cite blogs" is true of Claude and Perplexity in that test and not of ChatGPT.
Our own check of Google's AI Overviews, on a different set of questions, found a different mix:
| Figure | What it counts | Sample and date | Source |
|---|---|---|---|
| 29% | Share of AI Overview citations that were articles and guides | 356 citations from 42 AI Overviews for 48 buyer questions, US, 14 November 2026 | Our study |
| 22% | Share that were YouTube videos | Same study | Our study |
| 44 of 81 | Citations that were comparison pages, for "X or Y" questions | Same study | Our study |
| 195 of 212 | Sites cited only once | Same study | Our study |
Do the assistants agree with each other?
| Figure | What it counts | Sample and date | Source |
|---|---|---|---|
| 71% | Share of cited domains that appeared in only one of five models. For individual URLs: 89% | 2.04 million citations from 134,673 responses, published 24 March 2026. ChatGPT, AI Overviews, Gemini, Grok and Copilot; not Claude or Perplexity | Temso |
| 14.4% | Share of cited domains that an average pair of models has in common | Same study | Temso |
| 89.1% | Share of pages Claude cited that were also in Brave Search's top 20 for the query | 1,000 prompts, 8,500 URLs, published 28 October 2026. Not confirmed by Anthropic | Peec AI |
How long a citation lasts
| Figure | What it counts | Sample and date | Source |
|---|---|---|---|
| 11 days | Median time from a page's citation peak to half of that peak share. Not how long a citation lasts | 883,000 pages, 7 engines, 8 September 2025 to 8 September 2026 | Profound |
| 42 days | Median time from a page's first citation to half of its peak share | Same study | Profound |
AI crawlers and robots.txt
| Figure | What it counts | Sample and date | Source |
|---|---|---|---|
| 19 of 56 | Large sites that block GPTBot, the training crawler | 60 well-known sites, 56 readable, 14 November 2026 | Our study |
| 8 of 56 | Large sites that block OAI-SearchBot, the crawler behind ChatGPT search | Same study | Our study |
| 0 of 98 | B2B software vendors that block an AI search crawler site-wide. 2 block a training crawler | 98 vendors from 24 US buyer searches, checked 9 October 2026 | Frase |
| None | Major AI crawlers that ran JavaScript | Requests seen on Vercel's network, published 17 December 2024. Two years old | Vercel |
llms.txt
| Figure | What it counts | Sample and date | Source |
|---|---|---|---|
| 19 of 48 | Large sites with an llms.txt. 5 of the 19 follow the format on every point | 60 well-known sites, 48 clear answers, 14 November 2026 | Our study |
| 58 of 93 | B2B software vendors with an llms.txt | Checked 9 October 2026 | Frase |
| 84 of 62,100+ | AI crawler visits that went to the llms.txt on a test site | One site, 90 days; page last updated February 2026 | Otterly |
AI-written content and Google
| Figure | What it counts | Sample and date | Source |
|---|---|---|---|
| 86.5% | Top-20 pages containing some AI-generated text, by Ahrefs' own detector | About 600,000 pages for 100,000 keywords, July 2025 | Ahrefs |
| 0.011 | Correlation between the share of AI text on a page and its ranking position | Same study | Ahrefs |
| 2 of 2 | Test sites of 1,000 AI posts each that Google removed from its index (within 8 days and 12 weeks) | One experiment | Otterly |
| 2.0x | Median growth in Google clicks by days 197 to 224 for sites using one AI SEO product | 2,119 sites, data to 25 September 2026. No control group | Outrank |
| 24.4% | Sites using that product that had fewer clicks than before they joined | 1,383 sites followed for 90 days or more | Outrank |
How writers use AI
| Figure | What it counts | Sample and date | Source |
|---|---|---|---|
| 95% | Creators who use AI somewhere in their blogging | 80 of 81 respondents, 8 April to 18 May 2026, mostly the publisher's own audience, about ten points of margin | RightBlogger |
| 50% | Creators who never tell readers they used it | Same survey | RightBlogger |
What is not here
Figures we could not trace to a stated sample, which is most of what circulates. If a number you have seen quoted is missing, that is usually why. Forecasts are left out too: a prediction is not a measurement.
Corrections
Before this page was written we checked our own notes against each source, and eight figures needed correcting: a share of domains noted down as a share of citations, a half-life noted down as a lifespan, a sample attached to the wrong finding. The rows above give what each source says. If you find one that no longer matches, the source wins.
Questions
- Can I quote these figures?
- Quote the original source, which is linked in each row, and keep the sample and date with the number. For our own two studies, the data and a citation line are on the study pages.
- Why are so many from companies that sell AI visibility tools?
- Because they are the ones with the data. A tracker sees millions of answers; nobody else does. It also means each study measures that company's customers and prompts, so two studies can disagree and both be right.
- How often is this page updated?
- The date at the top is when every source was last read. Figures on AI search go stale within months, so treat anything older than a year as history.
Sources
Each one was opened and read on 14 November 2026. Figures in the text say what was counted, how many and when.
- 1.Outrank (/ai-citations) · outrank.so
- 2.Otterly (/blog/ai-search-index) · otterly.ai
- 3.Surfer (/blog/ai-citations-first-100-words) · surfer.ai
- 4.Temso (/data/Same-question-different-sources) · temso.ai
- 5.Peec AI (/blog/how-to-optimize-content-for-claude-search) · peec.ai
- 6.Profound (/blog/the-half-life-of-an-ai-citation-is-11-days) · tryprofound.com
- 7.Frase (/blog/is-your-website-blocking-ai-search) · frase.io
- 8.Vercel (/blog/the-rise-of-the-ai-crawler) · vercel.com
- 9.Otterly (/blog/the-llms-txt-experiment) · otterly.ai
- 10.Ahrefs (/blog/ai-generated-content-does-not-hurt-your-google-rankings) · ahrefs.com
- 11.Otterly (/blog/how-to-appear-in-ai-search-engines) · otterly.ai
- 12.Outrank (/case-studies/do-ai-seo-agents-work) · outrank.so
- 13.RightBlogger (/blog/ai-blogging-statistics) · rightblogger.com
Content Autopilot team
We make Content Autopilot, an SEO and AI visibility product
Our guides are drafted with AI from the sources listed above, and every figure is checked against its source before it is published. Where we sell something a guide discusses, the guide says so.