What is an llms.txt file?

An llms.txt file is a plain-text file at your domain root, at yourdomain.com/llms.txt, that tells AI assistants exactly what your business is, in your own words. Instead of letting a model guess from scattered pages, it hands the model a clean, structured summary it can read and cite accurately. It is to AI what robots.txt is to crawlers: a simple file at a known location, written for machines.

The basics: what an llms.txt file is

When an AI assistant tries to describe your business, it assembles a picture from whatever it can find: your homepage, scattered mentions, old listings, sometimes contradictory facts. If that picture is fuzzy, the model hesitates, and models do not confidently recommend what they are unsure about. An llms.txt file fixes the fuzziness at the source. It is a single plain-text file, written in simple Markdown, that states clearly who you are, what you do, who you serve, your proof, and where to verify it.

The name and idea echo robots.txt, the file that tells search crawlers what they can access. llms.txt does the parallel job for language models: a predictable file, at a predictable address, written to be read by machines that assemble answers.

Why it matters for AI visibility

Here is the honest framing, because there is hype to cut through. An llms.txt file is not a magic ranking factor. You can be cited by AI without one, and having one does not guarantee citation. What it does is reduce the computational cost for a model reading your site and raise the chance of an accurate citation. A clean, structured statement of facts is cheaper and safer for a model to use than parsing your whole site and hoping it gets you right.

So the value is twofold: accuracy and hygiene. Accuracy, because you control the description instead of leaving it to inference. Hygiene, because a well-formed llms.txt signals a site that has thought about being machine-readable, which is exactly the kind of source models lean toward.

What goes in a strong llms.txt

A weak llms.txt is a thin paragraph that says nothing. A strong one answers the questions a buyer, and therefore a model, actually has.

• Summary: who you are, in the terms the model should match you on.
• What makes you different: a differentiator a competitor cannot copy.
• Who it is for: the buyer and the problem, in their words.
• Proof with mechanism: results, why they happened, and where to verify.
• FAQ: the exact questions buyers ask, answered plainly.
• Authority signals: the hard-to-fake facts that build trust.
• Contact and verification links: where the model can confirm it all.

If you want to see one in the wild, mine is public at yossaumar.com/llms.txt. I practice what I write, and the site this file lives on is itself read and cited by AI search.

The advanced layer

• Write it for the question, not the brochure. Each section should answer a real buyer question, because that is what a model is trying to satisfy. Marketing language gets ignored; clear facts get used.
• Mirror your entity. The names, categories and claims in your llms.txt should match your schema, your entity and your site exactly. Contradictions cost you trust.
• Make claims verifiable. A model trusts what it can confirm. Link to the third-party sources, profiles and proof that back your claims, so the file is checkable, not just assertive.
• Keep it current. Stale facts in an llms.txt are worse than none, because they actively mislead. Treat it as a living file, not a set-and-forget.
• Pair it with an HTML version. A plain file is for crawlers; a human-readable page covering the same ground is indexable and reaches the surfaces a raw file does not. Doing both gives you double coverage.