security.txt and an AI policy page: the identity files agents look for

Why a contact file is an agent-readiness signal

Agents and the people who run them look for a clear contact and a clear policy before they trust a site. AI assistants now answer the question before anyone clicks a result, so trust signals help a model treat you as a reliable source.

Agent traffic breaks in ways a human visitor never reports. A crawler gets rate-limited, a fetcher hits a challenge page, an assistant's tool call returns a malformed response. The operator on the other side wants to tell someone, and to ask whether they may keep going. If your domain publishes a machine-readable contact, they can. If it does not, they log the failure and move on, and your pages quietly stop appearing in that platform. The same file also carries a Policy link, which is where an operator expects to find your rules for automated access.

security.txt, per RFC 9116

RFC 9116 defines a plain-text file at /.well-known/security.txt, served over HTTPS. Two fields are mandatory: at least one Contact (a mailto:, tel: or https: URI) and an Expires timestamp after which the file should be considered stale; the RFC recommends keeping expiry under a year. Optional fields include Policy, Canonical, Encryption, Preferred-Languages and Acknowledgments. The file was designed for vulnerability reports, but it is the only standardised machine-readable contact a domain has, which is why agent-readiness checks reuse it.

Contact: mailto:[email protected]
Contact: https://www.brand.ch/contact
Expires: 2027-07-01T00:00:00Z
Policy: https://www.brand.ch/ai-policy
Canonical: https://www.brand.ch/.well-known/security.txt
Preferred-Languages: de, fr, en

• Roll `Expires` forward automatically. A stale file is worse than none; generate the date at deploy time or serve the file from a small route, as qomvia.com does.
• Point `Policy` at your automated-access page, not only at a vulnerability disclosure policy. Operators read it for both.
• Keep it plain text, text/plain; charset=utf-8, no HTML wrapper.

The AI policy page

robots.txt and Content Signals say what a crawler may fetch and what it may do with the copy. Neither answers the question a cautious platform asks before it acts for a user on your site: may an agent place an order, submit a form, create an account here? A short page states the answer in plain language. It does not need legal weight to be useful; it needs to exist at a findable address and to be linked from the places agents look. Qomvia's own is at /ai-policy.

• What is allowed: reading and quoting public pages with attribution; using the API within the published limits; acting on behalf of a signed-in user.
• What is not: training on paywalled content; bulk copying; bypassing rate limits or bot protection.
• How to identify yourself: a user-agent string with a contact URL, or a signed request.
• Where to ask for more: the same contact as in security.txt.
• Links to the machine files: robots.txt, llms.txt, the sitemap, the MCP or API documentation.

Link the page from the homepage footer under a name that contains 'policy' and 'AI' or 'automated', reference it in llms.txt, and add a comment line in robots.txt pointing at it. Discoverability, not eloquence, is what the check measures.

What Qomvia checks

Two checks under Identity. Machine-readable contact (3 points) passes when /.well-known/security.txt returns 200. Stated automated-access policy (4 points) passes when a homepage link matches a policy-style path such as /ai-policy, /automated-access or /bot-policy; a robots.txt that merely mentions a policy earns 2. Together they are 7 of the 12 identity points, alongside Organization structured data, and all three can be finished in one afternoon.

How Qomvia checks your trust files

Qomvia's score checks for security.txt and a published AI policy and gives you a template.

The free scan runs all 26 checks of the agent-readiness score in about a minute and lists every failing check with a copy-paste fix; the free Site monitor re-scans weekly, no signup for the score. The AI monitor then asks your buyers' questions to ChatGPT, Gemini and Grok every day and shows whether you are named, which rival is named instead and which sources the models used. Qomvia Market and the MCP server take it one step further, letting agents act on the recommendation. Sign up free or see plans and pricing.

Questions

Does security.txt attract spam or vulnerability reports we cannot handle?

It attracts reports you would otherwise receive nowhere. Use a dedicated mailbox and set expectations in the Policy page; RFC 9116 was designed for exactly this.

Is an AI policy page legally binding on crawlers?

That depends on jurisdiction and how it is incorporated into your terms. Its readiness value is separate: it is a clear, discoverable statement that lets platforms decide to act rather than abstain.

Can the policy live in our existing terms of service?

It can, but the check and most operators look for a dedicated page with a recognisable path. A short page that links to the relevant terms section is the practical answer.