Crawl-to-Referral Ratio
Crawl-to-referral ratio compares AI pages fetched with visitors referred back in one fixed window, expressed as pages crawled per referral.
The crawl-to-referral ratio is the number of pages an AI operator fetched per visitor its products referred back, written N:1. It puts AI crawling and AI referrals in the same fixed window.
How it is calculated
Count attributable AI pages crawled, count detected AI referrals, then state how many crawls occurred for each referral. Crawls from operators that serve both search and AI are counted separately because they cannot be attributed to either side.
The referral count is a floor: AI platforms frequently strip the Referer header required for detection. The ratio is therefore an upper bound, not a precise measurement. A site can record zero AI referrals while AI tools are still linking to it.
What it can tell you
Under fixed definitions and one reporting window, the ratio supports comparison between operators or between like-for-like snapshots. It describes an observed exchange; it is not a conversion rate, a causal claim, or an industry benchmark.
WebDecoy’s 27 July–26 August 2026 dogfood snapshot recorded 13,828 attributable AI pages crawled and 5 detected referrals, producing an upper-bound ratio of 2,766:1. Referrals may be missing because of stripped headers, and one 30-day window on one small site does not establish a trend. See the public report and methodology write-up.
Related terms
- AI Referral, the undercounted visitor side of the ratio
- Generative Engine Optimization, the practice whose observable traffic can be measured
- AI Analytics, measurement of what AI crawlers take and what AI products send back
Frequently Asked Questions
Is crawl-to-referral ratio a conversion rate? +
No. It compares pages fetched by attributable AI crawlers with detected AI referrals during one window. It does not show that a crawl caused a visit, and it does not measure how often crawled content became an answer or citation.
Why is the ratio an upper bound? +
AI platforms frequently strip the Referer header used to detect referrals. The referral count is therefore a floor, so dividing crawls by that undercount makes the resulting N:1 ratio an upper bound rather than a precise measurement.
See these concepts in action
WebDecoy puts deterministic detection and rotation-proof enforcement behind a 5-minute setup.