An honest assessment of the format
It is worth being precise about what is known. llms.txt is a community proposal. No major search or assistant provider has publicly committed to reading it as a retrieval or ranking input. Publishing it will not, on its own, produce citations, and any vendor claiming otherwise is overselling.
What it does provide is a curated, machine-readable statement of what your site is and which pages matter, at effectively zero cost, in a format that agents and scrapers can consume without parsing your navigation. As agentic browsing becomes more common, a clean manifest is a reasonable hedge. It also has an underrated internal benefit: the exercise of choosing thirty representative pages exposes content sprawl that no analytics dashboard surfaces.
Treat it as hygiene alongside robots.txt and sitemap.xml, keep it accurate, and set expectations accordingly with stakeholders.
Keeping it correct over time
The failure mode is staleness. A manifest listing retired URLs, renamed tools, and outdated pricing actively misinforms any agent that reads it, which is worse than publishing nothing. Regenerate it as part of the same routine that regenerates your sitemap, and validate that every listed URL returns a 200 rather than a redirect chain or a soft 404.
Never list authenticated or gated routes. A manifest is a public document, and enumerating private dashboard paths hands an attacker a map while offering no discovery benefit, since those pages should be disallowed anyway.
Keep the descriptions honest and specific. One clause explaining what the page does is more useful to a retrieval system than a marketing sentence, and it is far easier to keep accurate.