llms.txt: a discovery aid, not a substitute for useful pages

What is being proposed

The llms.txt project proposes a concise Markdown description and links to useful material for agents. It is a discovery convention rather than an authentication system, universal indexing protocol or guaranteed ranking signal. A vendor publishing such a file on its own documentation site does not establish that every search product uses third-party files to choose citations.

Make the maintenance decision explicit

The file is most useful when someone owns a compact, accurate description and the linked resources remain available. Record which client or documented integration benefits and what maintenance is required. If your site already generates a public catalogue and Markdown pages, use the same source of truth rather than hand-maintaining contradictory descriptions. Do not put credentials, private prompts or confidential client documents into a public discovery file.

Test what actually exists

Fetch the file and check status, content type, correct business identity and every destination. Fetch a linked Markdown representation and compare it with the public HTML’s substantive facts. An HTML fallback with a 200 status is not a working Markdown resource. Check behaviour after service changes and deployment. Keep the file’s purpose distinct from robots access preferences.

Do not make the file the strategy

Google says its AI search features do not require new AI text files or special schema. If a key service page is blocked or fails to explain scope, fix that before treating discovery conventions as urgent. Where a partner requires a particular format, implement and test that requirement as an integration task. Report it as implemented discovery support, not proof of future model recommendations.

Continue with the related decision

• AEO learning paths
• AI-referred 404s: recover the right destination without misleading redirects
• Discuss the scope of AEO and AI-search work