Guide
llms.txt: what it is, and what it is not
A convention, not a standard, and ignored by Google Search. It is still worth publishing — because it is the one place you get to write the summary rather than hope one is inferred.
Last reviewed 27 August 2026
llms.txt is a Markdown file at the root of your site that tells a language model what this site is and which pages matter. It was proposed as a convention in 2024 and has spread through developer-tooling documentation faster than anywhere else.
Two things to be clear about before you spend an afternoon on it:
- It is not a standard. No specification body has adopted it. Support is voluntary and uneven.
- Google Search ignores it. It has no effect on indexing or ranking. Anyone selling it as an SEO measure is mistaken.
So why does our rule set award it 3 points? Because a crawl gives an agent everything, and llms.txt gives it the part you would point at yourself. It costs an hour, it cannot hurt, and it is the only place where the summary of your business is written by you rather than inferred from whatever page happened to rank.
The shape
# Acme Espresso > Acme sells espresso machines, grinders and accessories, with verified > compatibility data for every portafilter, basket and tamper we stock. > Based in Berlin, shipping across the EU. Last reviewed: 2026-08-27. ## Start here - [Machines](https://example.com/machines): all current models with specs - [Compatibility guide](https://example.com/compatibility): what fits what - [Grinders](https://example.com/grinders): by burr size and use case ## Reference - [Portafilter sizes](https://example.com/ref/portafilter): 49, 51, 53, 57, 58 mm - [Shipping and returns](https://example.com/shipping) ## Optional - [Company history](https://example.com/about)
The conventions worth following: an H1 with the name, a blockquote summary, H2 sections of linked lists, and a short note after each link saying what is there. An Optional section marks what can be skipped when context is tight.
The variant files
| File | Purpose |
|---|---|
/llms.txt | The index: description plus curated links |
/llms-full.txt | Full content inlined, for models that will not follow links |
/llms-en.txt, /llms-de.txt | Per-language variants, referenced from the main file |
llms-full.txt is the one that goes stale. It duplicates your content, so either generate it at build time from the same source or do not publish it — a full text file six months out of date is worse than none, because it will be believed.
How we grade it
- Pass — the file exists, is over 200 bytes, has a heading and contains links.
- Partial — it exists but is too thin, or does not follow the heading-and-links shape.
- Fail — no file, or the request returns your HTML 404 page.
Many sites return a styled HTML 404 with status 200 for unknown paths. That means /llms.txt appears to exist and serves a web page. We check the content type and the body, so this reports as a fail — correctly, because an agent asking for the file gets a rendering of your error page.
Where it sits among your priorities
Below server rendering, structured data and form semantics — all of which affect every agent, including the ones that have never heard of this convention. Above nothing at all, and worth doing on the same afternoon you fix your robots.txt.
The genuinely useful case is documentation. If you ship an SDK or an API, a curated index of your reference pages measurably improves what assistants tell your users about your product, and that is a support-cost argument rather than a marketing one.
Sources
Primary documents, checked on 27 August 2026
- llmstxt.org — The original proposal and format description
- Google — crawlers and user agents — No llms.txt support documented
- RFC 9309 — Robots Exclusion Protocol — For contrast: what an actual adopted convention looks like
Keep reading
Check your own site against this
The Agent Readiness Score measures exactly what this article describes, and shows the evidence behind every finding.
Run the check →