llms.txt: What It Is, Why It Matters and How to Write One
llms.txt is a Markdown text file placed at the root of a website (/llms.txt). It compactly describes what the site is and which pages a language model should read. The standard was proposed by Jeremy Howard (Answer.AI) in September 2024 — analogous to robots.txt, but for understanding content rather than restricting access.
Key takeaways
- llms.txt is a community-proposed standard; the format is plain Markdown.
- Structure: an H1 with the name, a blockquote description, H2 sections with annotated link lists.
- Adopting it costs almost nothing and carries no risk, but there are no guarantees either: major providers have not officially confirmed automatic use.
- The one rule that matters: the file must be factual and in sync with the site.
The format
A minimal working example:
# Project name
> One or two sentences: what this is and who it is for.
## Core
- Home: https://example.com/ — what is on it
- Docs: https://example.com/docs/ — what is inside
## Optional
- Blog: https://example.com/blog/ — what you write about
For large sites the standard also allows llms-full.txt — an extended version with the full text of key pages in Markdown.
What llms.txt actually gives you
- It saves model context. An agent or answer engine gets a digest of the site in one request, without crawling the navigation.
- It sets the emphasis. You decide which pages are primary and how your service is phrased.
- It disciplines the team. The file forces you to state the facts about the company once — the same phrasing then propagates into answers.
An honest caveat: as of 2026, neither OpenAI, nor Anthropic, nor Google has officially stated that their systems automatically fetch llms.txt. It is a zero-downside bet on the future — which is exactly why we include it in the baseline set of GEO practices.
Common mistakes
- Marketing instead of facts. A model will not quote "the best solution on the market"; it will quote "a studio that does X".
- Drift from the site. The file describes pages that no longer exist or contradicts the content — this undermines trust in the whole source.
- A giant file. A tens-of-thousands-of-words wall is not a digest; that is what llms-full.txt is for.
- Broken or non-canonical links. Use the same URLs as in your sitemap and canonicals.
How to ship one in an hour
- Collect the list of key pages with one sentence about each.
- Phrase the company description in one or two lines — factual, no superlatives.
- Put the file at the site root and add an update date.
- Add a reminder to re-check the file against the site on every release.
Our own file is public: geo-rank.ai/llms.txt. Want one tailored to your site — get in touch.
Published: 2026-08-09 · geo-rank.ai