GEO statistics: figures and sources

Every figure was checked against the original publication on 4 November 2026. Each figure states exactly what was measured, because that is where GEO statistics often go wrong.

Figure	What was measured	Note	Source
37.9 percent	Share of pages cited in AI Overviews that also ranked in the first 10 search results, across 863,000 search results.	In July 2025 Ahrefs measured about 76 percent, with a smaller sample. Compare the two with care.	Ahrefs, 2026
0.664 and 0.218	Spearman correlation of visibility in AI Overviews with brand mentions and with backlinks, across 75,000 brands.	A correlation, not proof of cause. Only covers AI Overviews.	Ahrefs, 2025
About 60 percent	Share of citations in ChatGPT and Google AI Mode that had gone the next day. Ahrefs tracked about 37,500 citations in five tools over 20 days.	Across all five tools, about 75 percent of what disappeared came back within 20 days.	Ahrefs, 2026
47.9 and 46.7 percent	Share of Wikipedia within the top 10 sources of ChatGPT, and of Reddit within those of Perplexity. Data from August 2024 to June 2025.	This is a share within the top 10. Of all citations it was 7.8 and 6.6 percent.	Profound, 2025
30 to 40 percent	Gain in visibility on Position-Adjusted Word Count from citing sources, adding quotations and adding statistics.	Experiment with rewritten texts, not on real websites. The effect differs per domain. Keyword stuffing did not perform well.	Aggarwal et al., KDD, 2024
None	Number of major AI crawlers that executed JavaScript in the measurements by Vercel and MERJ.	Measurement from December 2024. Check your own logs if it concerns a newer crawler.	Vercel and MERJ, 2024
7 May 2026	From this date Google no longer shows FAQ rich results in its search results.	FAQPage markup does no harm, but no longer produces an expandable result in Google.	Google Search Central, 2026

The sources themselves

GEO: Generative Engine Optimization (Aggarwal et al., KDD, 2024)
In an experiment with rewritten texts, citing sources, adding quotations and adding statistics gained 30 to 40 per cent more visibility on Position-Adjusted Word Count, and 15 to 30 per cent on Subjective Impression. Keyword stuffing did not perform well. The effect differs by domain, and the experiment did not run on real websites.
An Analysis of AI Overview Brand Visibility Factors (Ahrefs, 2025)
Across 75,000 brands, brand mentions on the web correlated more strongly with visibility in AI Overviews (Spearman 0.664) than the number of backlinks (0.218). This is a correlation, not proof of cause and effect.
38% of AI Overview Citations Pull From The Top 10 (Ahrefs, 2026)
Across 863,000 search results, 37.9 per cent of the pages cited in AI Overviews also appeared in the first 10 results for the same query. In July 2025, Ahrefs measured about 76 per cent, with a different sample.
How Stable Are AI Citations? (Ahrefs, 2026)
Ahrefs tracked about 37,500 citations in five AI tools over 20 days. In ChatGPT and Google AI Mode, about 60 per cent disappeared the next day, and about 75 per cent of the citations that disappeared came back within those 20 days.
AI Platform Citation Patterns (Profound, 2025)
Within the 10 most cited sources per platform, Wikipedia had a share of 47.9 per cent in ChatGPT and Reddit 46.7 per cent in Perplexity. Of all citations, that was 7.8 and 6.6 per cent. Data from August 2024 to June 2025.
The rise of the AI crawler (Vercel and MERJ, 2024)
In the measurements by Vercel and MERJ, none of the major AI crawlers ran JavaScript. They did sometimes fetch JavaScript files. The measurement dates from December 2024.
Optimizing for generative AI features on Google Search (Google Search Central, 2026)
Google writes that structured data is not required for generative AI in Search, that no special schema.org markup is needed, and that Google Search does not use llms.txt or similar AI files. Structured data does remain useful for rich results.
AI features and your website (Google Search Central, 2025)
AI Overviews and AI Mode can run several related searches on subtopics (query fan-out). A page must be indexed and eligible to appear in Search with a snippet to show up as a supporting link. There are no extra technical requirements.
Documentation updates: FAQ rich result (Google Search Central, 2026)
Since 7 May 2026, Google no longer shows FAQ rich results in its search results.
Overview of OpenAI Crawlers (OpenAI, 2026)
OAI-SearchBot determines whether a site can appear in ChatGPT search results. GPTBot collects content for training. The settings are independent of each other: you can allow OAI-SearchBot and block GPTBot.
Does Anthropic crawl data from the web? (Anthropic, 2026)
ClaudeBot collects content that may contribute to training. Claude-SearchBot indexes for search results: blocking it may reduce visibility there. Claude-User fetches pages when a user asks a question.
Perplexity Crawlers (Perplexity, 2026)
PerplexityBot shows and links websites in Perplexity search results and is not used for training. Perplexity-User fetches pages when a user asks a question.
Google's common crawlers (Google, 2026)
Google-Extended controls whether content may be used for training and grounding Gemini. It has no user agent of its own, and it does not affect inclusion or ranking in Google Search.
Webmaster Guidelines (Microsoft Bing, 2026)
Bing and the search features of Copilot use the same basis for crawling, indexing and ranking. NOINDEX keeps a page out of Bing and Copilot, and NOARCHIVE keeps it out of Copilot answers.
RFC 9309: Robots Exclusion Protocol (IETF, 2022)
The standard for robots.txt. The rules are a request to crawlers, not access control.

Figures we do not use

You often read online that citing sources yields exactly 40 percent, statistics 37 and quotations 30. We could not find that split in the text of the Princeton study, so we give the range the authors state themselves. The claim that Wikipedia supplies almost half of all ChatGPT citations is also wrong: that share applies within the top 10 sources.