What Is llms.txt?

The short definition

llms.txt is a Markdown-formatted plain-text file served at yourdomain.com/llms.txt. It states, in language a model can parse in one pass, who you are, what you do, who you serve, and where your most authoritative pages live. Think of it as a curated table of contents written for a machine reader rather than a human browsing your navigation.

It was proposed in 2024 by Jeremy Howard as a convention for making websites legible to large language models operating under tight context limits. It is a community standard, not a W3C or IETF specification — no AI provider is contractually obliged to read it.

How it differs from robots.txt and sitemap.xml

• robots.txt is permission. It says which agents may crawl which paths. It is enforcement-oriented and says nothing about meaning.
• sitemap.xml is inventory. It lists every URL you want discovered, with no hierarchy of importance and no explanation of what any of them contain.
• llms.txt is meaning and priority. It states what the organization is, and points to the handful of pages that actually define it — the summary a model would need if it could only read 2,000 words of your site.

They are complementary, not substitutes. A site can have a perfect sitemap and still leave a model guessing about what the business does.

What actually belongs in it

The structure that holds up in practice:

• An H1 with the entity name. The exact, canonical name — not a tagline. This is the single most important line in the file.
• A blockquote summary. One or two sentences defining the organization in the terms you want a model to repeat back. Write the sentence you want quoted.
• Disambiguation. If your name collides with other entities, say so explicitly. This is the highest-leverage line for any brand with a common name.
• Grouped links with descriptions. Sections like Services, Proof, About, with each link followed by a short clause explaining what it contains.
• Verifiable facts. Founding date, contact, service area, certifications. These must match your JSON-LD exactly — conflicting facts are worse than absent ones.

Some sites also publish llms-full.txt, a longer expansion containing the full text of key pages. Useful when your core content is short and stable; a maintenance burden when it is not.

What llms.txt will not do

Being honest about this matters, because the file is frequently oversold. It will not force an AI system to cite you. It will not repair a site that blocks GPTBot at the firewall. It will not manufacture authority you have not earned, and it is not read by every model or on every request. Adoption across providers remains partial and largely undocumented.

What it reliably does is remove ambiguity. When a model does encounter your site, the file determines whether it forms a confident, specific understanding or a vague one. That difference is what separates being named in a recommendation from being described generically.

The most common mistakes

• Contradicting your schema. If llms.txt says one founding year and your JSON-LD says another, you have actively reduced model confidence. Consistency beats completeness.
• Writing marketing copy. Superlatives and slogans do not survive extraction. Write declarative, checkable statements.
• Listing every page. That is what sitemap.xml is for. Fifteen well-chosen links outperform two hundred.
• Letting it go stale. An llms.txt describing services you no longer offer is a live source of wrong answers about you.
• Treating it as the whole strategy. It is one input among several, and not the heaviest one.

Where it sits in the wider picture

llms.txt is a BotIQ improvement — it makes your site more legible to the machines doing the reading. It does nothing for CommunityIQ, the independent corroboration a model looks for elsewhere on the web. A site with an immaculate llms.txt and no third-party evidence still gets skipped, because the model has nothing outside your own claims to lean on.

This is also why a file alone is not remediation. It belongs alongside consistent JSON-LD, server-rendered pages crawlers can actually read, and content that answers the questions buyers put to assistants. See why measurement tools cannot fix invisibility for the fuller argument.

How Monic AI Systems handles it

We treat llms.txt as a derivative of the entity model, never as a standalone asset. The canonical facts live in structured data; the file is generated to match. Our own is public at monicaisystems.com/llms.txt — worth reading as a working example rather than a template to copy, since the disambiguation section is specific to our name collision problem and yours will differ.

What to do next

• Check what AI already says about you. Run the AI Visibility Checker before writing anything — the gaps tell you what the file needs to correct.
• Audit crawler access first. Read What Is BotIQ. A file no bot can fetch is decoration.
• Talk it through. Book an AI Visibility Review.