State of the agentic web
We scanned a sample of 500,000 of the million websites Chrome loads most: what they publish for AI agents, and how often they get it wrong.
Blog
Blog
Writing from the team behind these checks, on how the web is changing for AI agents.
Release noteslatest : Personal Agent Protocol files are now judged by the version they declare, and every scan comes from one address you can allow.
We now watch agentic security threats for you
Agent readiness is not only about following the specification. Our scanner now checks the security of what you publish for AI agents, and we follow new threats the way we follow the protocols.
Why our scanner shows you less red
Most agent-readiness checkers fill the screen with failures. We tested three of them on a site built to be wrong, and on real ones, to see what all that red actually means.
Is GEO only in our heads, or something real?
Twelve ways an AI agent ends up saying something untrue about your business. Six of them are yours to fix, and this is what those six look like across 150,074 domains.