What it is — and isn't
robots.txt says who may crawl; sitemap.xml lists what exists; llms.txt says what matters. Written in Markdown, it gives an AI system landing on your domain a curated briefing: a one-paragraph statement of what the site is, then organized links to the pages worth reading, each with a one-line summary. It is a convention, not a standard enforced by anyone — treat it as a strong hint to cooperative crawlers, and as documentation discipline for yourself.
Format contract
- H1: site/brand name. Blockquote: what the site is, in one or two sentences, with the core facts an AI should state about you.
- Sections as H2s (Products, Documentation, Guides, Company), each a list of
- [Title](URL): one-line summary. - An "Optional" section for secondary material. Keep the whole file curated — it is a highlight reel, not a dump of every URL.
- Place at
/llms.txt; optionally publish/llms-full.txtwith full clean text of key pages.
Honest expectations
What llms.txt will not do: it will not force any crawler to visit, will not override weak content, and is not yet universally read. What it will do: cost an afternoon, help cooperative crawlers orient instantly, align your own team on what the site's canonical story is, and signal to AI-native buyers that your house is in order. Ships in every site we build — including this one.
llms.txt for multi-language sites
Lead with the site's default language; add a labeled section for other languages (as this site's llms.txt does with its 中文 section). Keep page counts honest — the links listed must exist. A broken llms.txt link is worse than a missing file.
Frequently asked questions
Do ChatGPT or Google actually read llms.txt?
The convention emerged from the AI community (not a standards body) and crawler adoption is partial and evolving. Some AI crawlers and tools honor it; others ignore it. The cost-benefit is lopsided enough — an afternoon of work, zero risk — that we treat it as default infrastructure.
Should llms.txt list every page?
No. It is a curated briefing, not an inventory — that's what sitemap.xml is for. List the pages an AI should read to understand you: pillar content, key product pages, documentation, about. Dozens of links, not thousands.