Keep agents and crawlers separate
Crawler volume is scheduled and usually far larger than agent volume. Adding the two together hides how much on-demand agent use you really have.Don’t count every markdown fetch as an agent
Scripts, documentation mirrors and integrations fetch.md and llms.txt files constantly. A markdown fetch by curl, axios or a custom script is a bot, unless it identifies itself or asks for markdown in Accept.
Say what’s verified
Without a real client IP, every identity is self-declared. Report verified numbers only when you classified at request time with IP ranges loaded.Replacing an existing metric
If you already show an “AI traffic” number, a stricter definition will usually shrink it. Instead of changing the headline overnight, report two numbers:- Broad:
agentplusai-crawler, plus automated reads of.mdandllms.txt. - Narrow:
agentonly, the traffic where a person is waiting.