AI search term

Crawl-to-refer ratio

The number of pages an AI operator's crawler fetches for every visitor its assistant refers back to a site. A high ratio (thousands-to-one for pure-AI bots) is the normal shape of AI discovery, not a failure.

Also known as: crawl-to-referral ratio, crawl:refer ratio, crawl-to-visit ratio, crawl to refer

Updated 2026-07-28

The crawl-to-refer ratio is the number of pages an AI operator's crawler fetches from a site (or the web) for every visitor its assistant refers back. You compute it by dividing total crawl requests by total referral sessions over the same window. A ratio of 217:1 means the bot crawled about 217 pages for each click it sent.

The ratio exists because AI operators crawl for jobs that mostly precede any visit: building the training data a model learns from, and maintaining the retrieval index that grounds live answers. Neither guarantees a referral. So a high ratio is expected, not a malfunction.

The number splits cleanly by operator type. As of a 28-day window ending 21 July 2026 (Cloudflare Radar data, single-source and directional), pure-AI crawlers such as Anthropic's ClaudeBot sat around 2,237:1 and OpenAI's GPTBot around 217:1, while search-backed Google sat near 4.6:1, close to parity because it has always paired crawling with a click-sending search engine. The pure-AI ratios have fallen sharply since early 2026 as assistants began sending more traffic.

Read the ratio as evidence a crawler can reach and learn from you, not as a measure of success. Because much AI-referred traffic hides as direct traffic, the referral side is usually undercounted, making real ratios lower than raw logs suggest. And because reading rarely turns into a click (the Dark Library Effect), the right scorecard is citations, not referrals.