GEO Technical Note

What is LLMs.txt

“llms.txt is robots.txt for language models — a plain text file that says which parts of a site are open to them and what they may do with what they find.”

01

What is llms.txt?

It exists so a site owner can state intent rather than guess at it: whether training, summarising, rewriting and quoting are welcome. It is not a ratified standard, and saying otherwise would be overselling it — but the need behind it is real, and it is the clearest articulation so far of how sites want AI to treat what they publish.

Against robots.txt: what is actually different

Dimension robots.txt LLMs.txt
Who it addresses Search crawlers (Googlebot) Language models — training and answering
What it governs Fetching. Allow or disallow Use. Training, rewriting, quoting
What it is for Managing what gets indexed Rights, and the terms of AI use

02

What goes in it

A minimal example

# Title
> Optional description goes here

## Section name
- [Link title](https://link_url): Optional details

## Optional
- [Link title](https://link_url)

Beyond the basics

  • Who you are. State the core service in a sentence a model can quote back.
  • Notes for the model. Where the content applies, which version it describes, and that attribution is expected.
  • Structure. Group pages — guides, services — so a model can tell what kind of page it is reading.

03

Why bother

Rights over your own work

Original work absorbed into training and regenerated elsewhere is hard to notice and harder to undo.

Privacy and exposure

Declare the areas that should stay out — user-generated content, anything sensitive.

Being quoted correctly

Clear guidance reduces the odds of a model paraphrasing you into something you did not say.

Crawl load

A place to signal limits on large-scale automated fetching.

04

Common questions

Is llms.txt a standard yet?
No. It is a proposal in active use rather than a ratified standard, and most crawlers still take their access rules from robots.txt.
What happens if you do not have one?
Nothing immediate. Your content simply keeps being read and used on whatever terms the reader assumes.
Will models actually honour it?
That depends entirely on the operator, and it is voluntary. The firmer commitments today are to the AI user-agents declared in robots.txt.

Want your content read on your terms?

We can set up the GEO strategy and the protocol files that go with it.