AI assistants like ChatGPT, Claude, and Perplexity increasingly answer questions about products by reading those products' documentation. The problem for them is the same as for a new human reader: a docs site is a sprawl of navigation, marketing, and content, and it is not obvious which pages are the canonical answer.
An llms.txt file is a small, deliberate map that solves this. It is a proposed standard (see llmstxt.org) for a single Markdown file that tells AI systems which of your pages matter and what they contain.
Direct answer
An llms.txt file is a plain Markdown file hosted at your site root (/llms.txt) that gives AI assistants a curated list of your most important pages. Write it as an H1 with your product name, a one-sentence summary in a blockquote, then H2 sections containing Markdown links to canonical pages, each with a short description. Keep it short, link only pages that are current, and update it when your docs structure changes.
The format
The spec is intentionally minimal. It is readable by a machine because it is readable by a person, using ordinary Markdown:
- An H1 with the name of the product or site. This is the only required line.
- A blockquote with a one-sentence summary of what the product is.
- Optional prose paragraphs with any context an assistant needs, kept brief.
- H2 sections (for example, Docs, API, Guides), each a Markdown list of
[page](url): short descriptionlinks. - An optional
## Optionalsection for pages that can be skipped when the assistant has a limited context budget.
A worked example
The shape is the point. Each link is a canonical page, and each description tells the assistant when that page is the right answer:
# Acme-- the H1.> Acme is a scheduling API for developers.-- the summary blockquote.## Docsthen- [Quickstart](/docs/quickstart): get a first request working in five minutes.- [Authentication](/docs/auth): API keys, scopes, and token rotation.## Optionalthen- [Changelog](/changelog): dated record of API changes.
The file is only as good as the pages it points at
An llms.txt is a routing layer, not a content layer. If it faithfully points an AI assistant at a help article that describes last quarter's behavior, you have not helped the assistant cite you accurately. You have made it confidently wrong faster.
Frequently asked questions
Where do you put an llms.txt file?
At the root of your domain, served at /llms.txt, the same way robots.txt is served at /robots.txt. Some sites also publish an llms-full.txt with the full text of key pages inlined for assistants that want the content directly.
Is llms.txt the same as robots.txt?
No. robots.txt tells crawlers what they may access. llms.txt tells AI assistants what is worth reading and where the canonical answers are. One is about permission, the other is about curation. They complement each other.
Does an llms.txt file improve AI visibility?
It helps non-Google AI assistants find and cite the right pages, and it costs almost nothing to add. Google has said it does not require special files for AI Overviews. Treat llms.txt as low-effort insurance for the assistants that do parse it, on top of writing clear, current, well-structured content.
Conclusion
- An llms.txt is a curated Markdown map of your docs, hosted at /llms.txt.
- Write it as an H1, a summary blockquote, and H2 sections of linked canonical pages.
- Keep it short and point only at pages that are actually current.
- A map to stale docs makes AI answers wrong faster, which is why DocsKoala keeps both the file and the articles current.