ARD: 355 websites tell agents what resources they expose

What is ARD?

Agentic Resource Discovery is a proposed standard, published 26 Aug 2026 as v0.91 by authors at Google, Microsoft and Hugging Face, that lets a website list the resources it exposes to agents and, for each, two to five example questions it answers. A site publishes it by any of three mechanisms: a JSON file at /.well-known/ard.json with an entries array, an Agentmap: directive in robots.txt pointing at an entries file, or a <link rel="ard"> in the page head, with rel="ai-catalog" kept as the predecessor for compatibility. Each entry carries an identifier, a name, a media type, a URL or inline data, and the representativeQueries field.

How does Known Good check it?

From what the probe already holds. Every probe stores the site's robots.txt and the head of its homepage, so the count reads the Agentmap: directive and the link relations from storage, and records whether either points at /.well-known/ard.json.

How many websites publish ARD?

This is the first measured count of ARD adoption we know of; the standard is 4 weeks old on the measurement date.

Does ARD make a site agent-ready?

ARD tells an agent what to ask; it does not answer the question. A declaration is not one of the signals a listing counts. A site whose ARD entries point at resources that answer (a markdown page, an API, a server) is listed for those; the ARD file is the map.

One implementation note

At least one widely used readiness scanner checks the predecessor rel="ai-catalog" and not the current rel="ard". Serve all three mechanisms and both link relations; knowngood.sh does, at /.well-known/ard.json and /.well-known/ai-catalog.json.

How do I add ARD to my site?

Publish /.well-known/ard.json with an entries array, add Agentmap: to robots.txt pointing at it, and put <link rel="ard"> in your page head. Write two to five representativeQueries per entry in your own words: that field is a site telling an agent what to ask it. Then ask for a probe at /submit.

Frequently asked

Is ARD an API? No. It is a discovery file; the resources it lists are what an agent calls.

Which mechanism should I use? All three; they cost nothing and different agents read different ones.

Counts are of sites (hostnames), measured by Known Good's probe on 28 Sep 2026. Could-not-ask is counted beside the total, never inside it. Method: /bot. Rubric: /benchmark.

All standards · How we grade