There is a file called llms.txt doing the rounds. It is worth understanding precisely, because the claims around it have outrun the evidence.
What it is
llms.txt is a proposed convention: a plain-text markdown file at the root of a domain, describing the site in a form a language model can read quickly.
A minimal version looks like this:
# Example Ltd
> A two-sentence description of what this site is and who it is for.
## Key pages
- [Services](https://example.com/services): what we do and who for
- [Pricing](https://example.com/pricing): ranges, not a quote form
- [About](https://example.com/about): who works here, where we are
The idea is a curated map. Instead of parsing navigation, ads and cookie banners, a model gets a short, structured summary with links.
What it is not
It is not a standard. No major model provider has committed to reading it. It is not a ranking factor, because there is no ranking. It is not a replacement for robots.txt: robots controls fetching, llms.txt describes content.
It is also not a way to "feed" a model your preferred version of the truth. Nothing in the file changes what a model believes if the rest of the web disagrees.
Why it might still be worth five minutes
Three reasons.
Agents are becoming readers. As more agents browse on a user's behalf, a clean machine-readable summary is a sensible thing to provide, in the same way a sitemap was sensible before everyone agreed it mattered.
It forces clarity. Writing the file means answering: what is this site, in one sentence, and which five pages matter? Most sites cannot answer that, which is a useful diagnostic.
It costs nothing. One static file, no maintenance beyond occasional edits.
How to write a good one
- Lead with the summary. Two sentences, plain language, no marketing adjectives.
- Link only canonical, live pages. A 404 in llms.txt is worse than no llms.txt.
- Group by section. Tools, guides, articles, contact.
- Say what you do not do. It prevents misclassification.
- Keep it short. Under 100 lines is plenty.
The honest position
Add it. It takes minutes and might help with agent-driven traffic. Do not put it in a strategy deck as though it were a lever. The things that actually make you visible to AI systems are the same things that make you credible to humans: clear entities, quotable content, third-party corroboration, and technical access.
You can generate a starting file with our schema and llms.txt generator.