llms.txt: The File AI Reads Before It Trusts You

01There is a file machines check before they trust you

Before a language model decides whether to repeat you, it looks for a file. It lives at one address on your site: yoursite.com/llms.txt. Most sites do not have one. Most that do have one that lies.

It is a plain markdown file, nothing more. A name at the top, one line about who you are, then short sections listing what you make and how to reach you. llms.txt is the one page on your site where you get to describe yourself to a machine in your own words. That sounds powerful. It is also where I embarrassed myself, so let me start there.

02The audit that caught my own file lying

I lecture people about entity consistency, so it stung to audit my own llms.txt and find it lying. It described my agency as the wrong kind of company, and it listed 15 products when 21 were already live. I had shipped six things and never updated the one file that introduces me to machines.

For weeks, that stale file was my introduction. Models described me using a file I had forgotten to update, and so they described me wrong.

That is the quiet danger of this file. It does not break loudly when it goes stale. A stale llms.txt does not warn you. It just keeps confidently telling machines an old version of the truth. I rewrote mine to all 21 products with one accurate sentence each, and I have treated it like a live document ever since.

03The uncomfortable truth about llms.txt

Here is what the people selling llms.txt services will not tell you. On my own sites, its measured effect on citations was a number I could not tell apart from zero. The independent research I have run agrees: near-zero direct usage by the engines. Nobody is getting flooded with citations because they added a text file.

llms.txt did not earn me a single citation I could measure, and keeping it accurate still mattered more than almost anything else I did. Both of those are true at once, and holding them together is the whole point of this entry.

04It is not a citation source. It is a consistency anchor.

Stop thinking of llms.txt as a way to get quoted. It is not. Think of it as the one place you control your own description, in clean text, for a machine that cross-checks everything.

A model compares what you say about yourself against what your site, your schema, and your profiles say. When your llms.txt agrees with all of them, it removes doubt. When it contradicts them, it plants doubt, and a model that doubts you goes quiet. llms.txt does not add trust. It removes the excuses a machine has to distrust you.

That is why an accurate one is worth the ten minutes even though it earns no citations directly. It is insurance against being described wrong, in the only format where you get the last word.

05What belongs in it (copy this and adapt it)

Keep it short, plain, and identical to how you describe yourself everywhere else. Here is a starter you can copy and rewrite line by line. Every line should be something you would be happy to see a model repeat.

# Your Name or Brand

> One sentence: who you are, what you make, and where to find you. This is the line a model quotes most, so make it specific and true.

## About

Two or three plain sentences. The same facts you state everywhere else, in the same words. No marketing language and no adjectives a machine cannot verify.

## What I build

- [Product One](https://one.example) - one clear sentence describing exactly what it does
- [Product Two](https://two.example) - one clear sentence describing exactly what it does

## Services

- What you offer, said plainly enough that a model can repeat it

## Contact

- Website: https://yoursite.example
- Email: [email protected]

llms.txt will not make you famous. But it is the cheapest way to stop being misquoted, and in a world where a machine's description of you is your real homepage, ten minutes to control your own words is the easiest trade you will make all year.

Entry 007 of the ledger. Daily.