llms.txt: What It Is, Who Reads It, and Whether You Need One

What llms.txt actually is

llms.txt is a proposed convention: a Markdown file at the root of your domain that gives large language models a curated, plain-language summary of your site and links to its most important content.

The reasoning behind it is sound. When a model or an AI crawler encounters your site, it has to work out what the site is from navigation, boilerplate, cookie banners and whatever HTML it can parse. A hand-written summary is a cleaner signal than an inferred one. The format is Markdown because that is close to what models consume most comfortably.

A minimal example:

# Example Studio

> Example Studio is the design practice of Jane Smith, based in
> Bristol, working with early-stage software companies.

## Key pages

- [About Jane Smith](https://example.com/about/): background and work history
- [Services](https://example.com/services/): what we do and how engagements run
- [Case studies](https://example.com/work/): named projects with outcomes

## Entity facts

- **Person:** Jane Smith
- **Role:** Product designer
- **Practice:** Example Studio
- **Based:** Bristol, United Kingdom

That is the whole idea: an H1 name, a blockquote summary, and linked sections pointing at what matters.

The honest state of adoption

This is where most articles on the subject become unreliable, so plainly:

• It is not a standard. No standards body has adopted it. It is a proposal that gained traction in SEO circles.
• No major AI provider has committed to reading it. Not OpenAI, Anthropic, Google or Perplexity. Some may fetch it opportunistically; none document it as a supported input.
• There is no verification. Unlike robots.txt, which crawlers demonstrably request, you cannot confirm that anything consumed your llms.txt or acted on it.
• Claimed results are almost entirely anecdotal. “We added llms.txt and our AI citations went up” is not attributable — those sites usually changed several things at once.

Compare this with robots.txt, which is a genuine convention that named crawlers respect and which has observable effects in your logs. llms.txt is a reasonable idea whose adoption has not yet happened.

Anyone telling you llms.txt is required for AI search visibility is describing a hope, not a mechanism.

What it is not

Three confusions worth clearing up, because they lead to real mistakes.

It is not access control. llms.txt grants and denies nothing. It cannot stop a model training on your content or instruct a crawler to skip a section. If you want to influence crawler access, that is robots.txt, and it is the only file with any traction there.

It is not a substitute for structured data. Schema is machine-readable, standardised, actually consumed by search engines, and it powers real features. llms.txt is prose. If you have limited time, write the schema.

It is not a ranking factor. It has no bearing on Google rankings. It is not a “new SEO signal”, whatever your feed says.

Why it may still be worth writing

Given all that, I still think it is worth an hour, for three reasons that do not depend on adoption.

First, the clarifying effect is real. Writing a plain summary of what your site is, who is behind it and which pages matter is a genuinely useful exercise. Most people find they cannot do it crisply, which is itself the finding. What you produce tends to improve your about page, your meta descriptions and your positioning — all of which do have measurable effects.

Second, it is cheap and reversible. One Markdown file, no dependencies, no risk. If adoption arrives, you are ready. If it does not, you have lost an hour and gained a summary.

Third, it is a single place to state entity facts. A short block naming the person, the brand, its alternate spellings and the canonical URLs is useful to anything that reads it, standardised or not — including a human trying to understand your site quickly.

What I would not do is build a content strategy around it, pay for one, or let it displace schema and content work.

How to write one that is useful

1. Start with an H1 that is your name or brand. No taglines.
2. Follow with a blockquote summary. Two or three sentences: what this is, who runs it, who it serves. Write it so that quoting it verbatim would be accurate.
3. Add an entity facts section. Person, role, brand, alternate spellings, location, canonical URLs, contact. Facts, not adjectives.
4. Link the pages that matter, with descriptions. Not your whole sitemap — the ten to twenty pages you would want cited.
5. Group by intent. “Services”, “Guides”, “Tools”, “About” are more useful than a flat list.
6. Say what you do not cover. A short “out of scope” note reduces the chance of being cited for something you do not do.
7. Keep it current. A stale llms.txt is worse than none, because it contradicts your live site.

llms.txt and robots.txt together

They do unrelated jobs and you want both.

One useful link between them: reference your llms.txt from a comment in robots.txt. Anything already fetching your robots file then has a pointer to the summary, without you relying on it guessing the path.

The verdict

Write one if: you can spare an hour, you want a single canonical statement of what your site and your entity are, and you are comfortable with an uncertain payoff.

Do not bother if: your schema is missing, your about page is thin, or your content is not worth citing. Those are all higher-value work, and none of them are substituted by this file.

Never: pay a premium for it, use it as a substitute for robots.txt, or let anyone tell you it is required for AI visibility. The genuinely useful version of this work is making your site clear, well-structured and worth quoting. llms.txt is a small, optional expression of that — not a shortcut to it.

FAQ