The 7 signals AI agents check before they trust your site

Agents do not read your homepage the way a person does. They look for specific, machine-readable signals that tell them what exists, how to read it, and what they are allowed to do. Here are the seven that matter most — what each is, why an agent cares, and how to add it.

1. Robots rules and a content policy

Your /robots.txt is the first file an agent checks. Beyond crawl rules, the newer Content-Signal directive lets you state your policy for AI training, search and input use. Missing or vague rules make an agent cautious; clear ones make you predictable.

2. A reachable sitemap

A valid /sitemap.xml (or one referenced from robots) hands an agent the map of your site instead of making it crawl link by link. It should be real XML — a <urlset> — not an error page returning a 200.

3. An llms.txt

/llms.txt is a short, Markdown guide that points agents to your most useful pages and resources. Think of it as a concierge note: here is what we are, here is where the important things live. It is the single highest-leverage file most sites are missing.

4. Markdown when asked

When an agent sends Accept: text/markdown, can your important pages return clean Markdown instead of HTML? This strips the navigation, pop-ups and layout an agent has to fight through, and gives it the content directly. Serving Markdown on request is one of the strongest readability signals there is.

5. An API catalog and OpenAPI description

If you have any public API — even a simple one — a /.well-known/api-catalog and an /openapi.json let an agent discover its operations without guessing endpoints. This is the bridge from "read my site" to "use my site".

6. An MCP server card

The Model Context Protocol is becoming the common language for describing tools an agent can call. A /.well-known/mcp.json advertises what tools you expose — once you genuinely have them. Publishing real tools here is how you move from being read to being used.

7. An agent-skills index

A /.well-known/agent-skills/index.json lists stable, callable skills an agent can rely on. It is the most forward-looking signal on this list, and the clearest statement that your site is built to be operated, not just viewed.

Grade yourself

You do not need to audit these by hand.

AgentVisible checks all of these (and a few more), scores your site, and generates draft files for the ones you are missing.

Most sites pass one or two and miss the rest. That is the normal starting point — and the fastest list of wins you will find this quarter.