
llms.txt is a proposed convention (from Jeremy Howard at Answer.AI, 2024) that's been adopted by a growing list of developer-facing and content-heavy sites: a plain markdown file served at /llms.txt that gives language models a curated map of your site — who you are, what you offer, and links to your most citable pages, each with a one-line description.
Why it exists
HTML pages are written for browsers: navigation, cookie banners, hero carousels, and somewhere in the middle, the content. A model with a limited attention budget has to dig. llms.txt inverts that — it's the site explaining itself in the format models parse most naturally, with the noise already removed.
What goes in one
- An H1 with your name and a one-paragraph description of what you do, for whom.
- Sections of links to your canonical pages — services, guides, FAQ, key articles — each with a one-line summary.
- Business facts worth getting right: contact, location, hours, service area.
Ours is live at /llms.txt — generated from the same data as the site, so it never drifts out of date. That's the implementation detail that matters most: a hand-maintained llms.txt rots quickly.
Honest expectations
llms.txt is a proposal, not a ratified standard, and no major platform has committed publicly to consuming it. So why ship one? Because the cost is nearly zero, the downside is nonexistent, several AI tools already read it opportunistically, and everything in a good llms.txt — canonical URLs, clean descriptions, consistent facts — is content you should have straight anyway. It's the cheapest insurance in AEO: if engines standardize on it, you're early; if they don't, you've lost an afternoon.
What it will never do
llms.txt doesn't unblock crawlers (that's robots.txt and your CDN), doesn't substitute for structured data, and doesn't fix content that never states an answer. It's a map — the destinations still have to be worth citing.
The free audit checks whether you have an llms.txt, whether it parses, and whether it agrees with what your pages actually say.