per visitor returned Move At that volume
readers you get back
What servers actually answered
The rate is only half the trade. Across the same network, this is how sites replied when an AI crawler asked. Roughly a fifth of requests were refused outright, which is publishers repricing the deal themselves.
Who says who they are
A rate is easier to accept from an operator that identifies itself, separates its crawlers by purpose, and publishes what it does with what it takes. Cloudflare tracks all three.
Verified — the operator's crawler can be cryptographically confirmed as theirs. By purpose — separate agents for training, search and user-initiated fetches, so you can allow one and refuse another. Metrics — publishes what it does with the data.
What this board does not say
A ratio is not a theft. Some of that crawling trains a model, some answers a question in place, and some genuinely returns a reader. Reasonable people price those differently. This board shows the terms; it does not tell you whether to accept them.
One number carries an asterisk the others don't. Cloudflare notes that referrals from
Claude's native app carry no Referer header, so Anthropic's ratio is likely
overstated by an unknown amount. It sits near the top of the board, so it needs saying near
the top too.
It is a snapshot, not a feed. Rates read on 4 October 2026 over a seven-day window. On the previous read, 21 September, Perplexity sat at 2650.5:1 and OpenAI had just passed Anthropic for second; since then Perplexity has nearly doubled and Anthropic is back ahead. The transparency and response tables below are from the 30 August read and have not been re-read. Check the live board before quoting any of it.
These are network-wide figures. Your own ratio depends on what you publish and who wants it. The column above scales the published rate to your traffic; it does not measure your logs, and it cannot.