Live data · recounted hourly · since —
Most of the “AI agents” visiting this site were not AI agents.
A User-Agent header is a claim, not a fact — anyone can send one. So we check every request against the operator’s own published IP ranges, or against reverse DNS confirmed in the forward direction. These are our own server logs, recounted from scratch on every run. Nothing here is modelled or estimated.
The count never resets. Server logs here are deleted after fourteen days, so every counted day is written to a ledger and kept. Otherwise this number would quietly shrink instead of grow — and a shrinking counter is worse than no counter, because it looks like an answer.
Loading…
Who actually came
Requests per user-agent, split by whether the source address could be verified.
The same numbers as a table
For screen readers, print, and anyone who prefers digits to bars.
How this is counted
Two methods, chosen per operator by what that operator actually publishes:
- Published IP ranges. The source address must fall inside the operator’s own JSON feed of prefixes — Anthropic, OpenAI, Perplexity, Google, Microsoft.
- Forward-confirmed reverse DNS. Where an operator publishes hostnames instead of ranges — Amazon, Apple, Yandex, DuckDuckGo — the address is resolved to a name and the name resolved back, because a reverse record alone can be forged.
- Neither available. Some crawlers publish no verification method at all. They are counted in their own column and never added to the verified total.
Raw data, machine-readable, same numbers as above: /agent-traffic.json.