Skip to content
ceeme

Does llms.txt work, or is it a file nobody reads?

llms.txt is a text file on your domain summarising in plain language what your site is and where the main pages live, meant for language models. No engine today is obliged to read it and there is no evidence it leads to more mentions by itself — what it does do is force you to write down in twenty lines what you are, and that exercise improves the rest of your site.

By Njusja Orban, founder of ceeme4 min read

Measure your own company → Rather talk it through first

In short

What it is
A readable text file at a fixed address, with your key pages and one sentence per page.
What it is not
Not an instruction to a model and not a replacement for your sitemap. Expecting behaviour from it is expecting too much.
Why have one then
Because it is free, because it sharpens your own summary, and because in a year it may well be read.

How does this file relate to the two that already existed?

Three files on your domain that all look like the same kind of thing and do three different things. Only the third is a proposal with nothing enforcing it, and that deserves to be stated plainly.

FileWhat it is forIs it read?
robots.txtsays who may come in and who may notyes, by practically every crawler that behaves properly
sitemap.xmllists which pages exist and when they changedyes, and it is the fastest route to a complete index
llms.txtsummarises in plain language what your website isno engine commits to it — it is a proposal, not an agreement

Is there evidence it yields anything?

No evidence that the file by itself leads to more mentions, and anyone claiming otherwise should supply the measurement. What is established: it takes half a day, it cannot break anything, and the summary you have to write for it is usable in five other places.

We have one ourselves, in four languages, and it comes from the same source as the rest of the site — the amounts in it come literally from the same file as the pricing page, and the build fails the moment the two diverge. The file is not the argument; the fact that it cannot drift is.

How do you write one that is worth having?

The first two sentences say what you do and for whom, without a single adjective you cannot prove. Then a list of your main pages, each with one sentence saying which question that page answers — not what is on it, but which question.

What happens if I let it go stale?

Then it becomes a source of wrong information about yourself, and that is worse than no file. An expired file at a fixed address keeps getting fetched; a price or a service that no longer exists travels from your old site into an answer.

The lesson goes wider than this file: everything you place by hand at a fixed address ages. What comes out of a generator changes with its source — and that is the only difference that counts in the long run.

Frequently asked questions

Do I need this before doing anything else?

No. The order is: first open access for crawlers, then get your structure right, and only then this. Creating this file while your site is blocked means writing a tidy note for a door that is shut.

Can I put instructions in it about how AI should describe me?

You can type them and nothing happens with them. A model does not follow orders from a file on a stranger's domain, and it would be a serious security problem if it did. What does work is factually describing what you are; that is not an instruction but information.

Read on

llms.txt generatorMake one yourself, no account robots.txt checkThe file that does enforce Which AI comes pastThe crawlers, and which are verifiable AlternativesOur own files, with their address
Measure your own company → Rather talk it through first

Free, no account and no card.