Think robots.txt, but for meaning rather than permission. Robots.txt tells a crawler what it may read. llms.txt tries to tell it what to understand, which is a much harder thing to ask of a text file.
The honest status.
It is a proposed convention, not a standard. No AI engine has committed to reading it, and anyone who tells you llms.txt is why a brand shows up in ChatGPT is guessing. What actually drives AI recommendations, as far as anyone can measure, is how often and how consistently your brand is described across the web.
So why ship one? Because it takes twenty minutes, it cannot hurt, and writing it forces a useful exercise: saying what your business does in one paragraph a machine could repeat back. Most businesses cannot do that on request. I have had clients find their positioning while writing theirs, which was not the point of the exercise.
That is the whole case for it, and I would rather give you the weak version of the case than the version that sells. A file that costs you one morning and might do nothing is still a reasonable bet. A file sold to you as the reason you will appear in ChatGPT is a different product, and it is being sold on a claim nobody can support.
What goes in it.
- One paragraph on who you are and what you sell, in the words a customer would use.
- The named terms your business uses, each defined in a line. If you have invented language, this is where you teach it.
- The pages that matter, each with a line on what it covers. Not your whole sitemap.
- How to get in touch, and the one action you want someone to take.
- Nothing you would not say on the page itself. Your name, address and phone number match the page copy and the schema character for character.
How to write one in twenty minutes.
The trick is to stop writing marketing copy, which is the first thing everybody does. Nothing in this file has to persuade anyone. It has to be repeatable by a machine with no stake in how good you sound.
- Open with the sentence you would say on the phone. Who you are, what you do, who for, where. If you cannot write that sentence, you have found something more valuable than the file.
- Name your services the way a customer names them. Not your internal product names. Somebody searching does not know you call it a Brand System.
- List the pages a stranger would need, each with a line on what is on it. Pricing, the thing you sell, how to get in touch, and whatever you publish. A handful, not forty.
- Define your own vocabulary. Any term you invented, any term you use differently from the rest of your industry, gets a line. This is the part most files skip and the part most likely to be useful.
- Read it back as a stranger. Every claim in it should be checkable on the site itself. If a line only exists in this file, delete the line or put it on a page.
The one thing that can hurt you is a file that disagrees with your pages. If llms.txt says one thing, your schema says another and your footer says a third, you have taught anything reading all three that your site is an unreliable source about itself. Consistency is the entire mechanism being proposed here. Contradicting yourself in the file meant to fix it is worse than not having one.
How to tell whether anything is reading it.
Your server logs, and only your server logs. A request for /llms.txt with an AI crawler in the user agent is the one piece of hard evidence available, and it tells you the file was fetched, not that it was used. Nobody outside those companies can tell you more than that, and a dashboard that claims to is inferring.
So do not put llms.txt on a report as a thing that worked. Put the file live, check it is reachable, and judge the programme on what you can actually measure: whether your name comes back when you ask an assistant the questions your customers ask, over time, against a written list of prompts you keep.
The part that actually matters.
Answer engines are the 2026 problem. An AI overview sits above the results and takes the click you used to get. The question stops being where you rank and becomes whether you are the thing being quoted.
An assistant reaches for names it keeps seeing described the same way in a lot of places. That comes from structured data on every page, an FAQ written in real questions, publishing regularly, and mentions on sites that are not yours.
llms.txt is the cheapest item on that list and the least powerful. Do it because it is cheap, not because it is the lever.
If you only have one morning, spend it on the first paragraph of your most important page instead and come back to this file afterwards. How to get your site cited by AI assistants is the long version, in the order I would actually do the work.
Ours is at www.5fourdigital.com/llms.txt, and it is section 7 of the 5Four Launch Checklist, alongside explicitly allowing the AI crawlers in robots.txt.
Related: llms.txt in the glossary, Answer Engine Optimization, the Site System.



