Do engines actually read llms.txt?

There is no public confirmation that any engine uses it as a ranking input, and we do not claim one. What it does do is give you a machine-readable place to state what your company is and what it is not, and that has changed answers for us where a name collision was the problem. Treat it as cheap insurance, not a lever.

Where does the file go and what should it return?

At the root of the domain, served as text/plain with a 200 status. Check it the same way a crawler would rather than trusting the CMS: request the URL directly and read the response. In our dataset 87% of sites had no file at all, and several of the ones that did returned a 404 to a plain request while looking fine in a browser.

How long should it be?

Short enough to be read end to end and specific enough to be useful. A description of the company, the pages that matter with one line each, the key facts with dates, a disambiguation paragraph, and contact. A dump of every blog URL is the most common bad version and it is worse than a short honest file.

How to create an llms.txt file

llms.txt is a plain-text file at the root of your domain that tells AI systems what your site is, which pages matter and how to describe your company. It takes about an hour to write properly, it cannot hurt you, and it is the only place where you get to state in your own words which of the things sharing your name are not you.

In our 100-audit dataset 87% of sites had no llms.txt at all, and where the file existed it was usually a dump of blog links with no description. So the bar is low.

what goes in it

The name and one-sentence definition. Written so it can be quoted verbatim. This is the sentence you would want an engine to repeat.
What the company does, in a short paragraph. Concrete, not adjectival.
Disambiguation. The section almost everyone skips. Name every entity that shares your string and state plainly that it is not you.
Key facts with dates. Founded, size, coverage, any number you would want cited — each carrying the date it was true.
The pages that matter, each with one line explaining what a reader would get from it.
Contact and canonical URLs.

the disambiguation section, and why it matters most

Name collisions are the most common entity failure we see and no SEO tool reports them. A model that finds three different companies behind one string either hedges or picks the wrong one.

Our own file carries this section for a practical reason: our domain string collides with a betting brand, an anime series, a music project and two unrelated agencies, and engines were repeating the wrong one back at us. The correction is four lines of text. Writing them is the highest-value hour in this whole exercise.

a skeleton you can copy

Structure, not syntax — the format is markdown-ish and forgiving:

# Company Name then a blockquote with the one-sentence definition.
## Disambiguation — “X is not related to any of the following, which share the name: …”, then the canonical name, domain and contact.
## Key facts — dated bullets.
## Pages — URL plus one line of what is on it.
## Contact — website, email, legal pages.

five ways to get it wrong

A link dump. Every blog URL, no descriptions. Adds nothing a sitemap does not already say.
Marketing adjectives. “Leading, innovative, world-class” is unquotable and unverifiable.
Undated numbers. A figure with no date cannot be trusted or refreshed.
Letting it go stale. A file listing pages that no longer exist is a signal of neglect.
Serving it wrong. A 404, a redirect chain, or HTML instead of plain text. Verify with a direct request, not in a browser tab.

after you publish it

Request the URL directly and confirm a 200 with plain text. Then check that it agrees with the rest of your machine-readable surface — the schema, the profiles, the page copy. A file that contradicts your own structured data creates the confusion it was meant to fix.

Ours is live at 009.agency/llms.txt if you want a working reference; the surrounding technical layer is described in technical GEO.

how to create an llms.txt file

what goes in it

the disambiguation section, and why it matters most

a skeleton you can copy

five ways to get it wrong

after you publish it

questions about the file.