A single index card standing proud of a long drawer of tightly packed library cards.

The llms.txt Standard And Why You Should Ship One This Week

A 1KB file that tells AI crawlers what your site is about and which sections they should read first.

An llms.txt file is a plain text file at your domain root that tells AI crawlers what your business is, what it offers, and which pages matter. It takes about twenty minutes to write and almost nobody has one.

That is the whole reason to ship it this week.

What it does

AI systems building an answer about your category have to work out what you are from whatever they can parse. Left to infer, they get it from your homepage, which is written for humans and is usually the least literal page you own.

An llms.txt gives them an unambiguous statement instead: this is the business, this is what it does, these are the pages that answer the common questions.

What goes in it

A single line describing the business in plain language. Not a tagline. The literal description a stranger would need.

Then the offering, stated as what it is rather than how it is positioned. Then the service area. Then a short list of the pages that carry real answers, each with a line explaining what is in it.

Keep it under a couple of kilobytes. It is a map, not a brochure.

What to leave out

Marketing language. Anything a crawler cannot verify. Claims without specifics.

The file works because it is literal. Filling it with positioning defeats the point, because positioning is exactly the thing a model cannot check.

Why now

The window matters. Right now the file is rare enough that having one is a genuine differentiator in how accurately AI systems describe you. As adoption rises that advantage flattens into a baseline.

The cost of shipping it is an afternoon. The cost of not having one is that your category gets described by whoever did.

The pairing that actually wins

On its own an llms.txt improves how you are described. It does not on its own get you cited.

Citation comes from having pages that answer specific questions directly, in extractable form, with facts a model can attribute. The file points at those pages. If the pages are generic, pointing at them harder does not help.

Ship the file, then make sure what it points at is worth quoting.

Written by David Eid. Published .