llms.txt: The New File "for AI". Does a Small Business Website Need It?

Another miracle on offer: this time it’s called llms.txt

Sooner or later an email will land in your inbox: “We’ll set up llms.txt for you so artificial intelligence can see your company. Your competitors already have it.” It sounds urgent — AI search is all anyone talks about and nobody wants to be left behind. That feeling is exactly what the offer is aiming at.

If you’ve ever paid for SEO that ended with a report on rankings and zero enquiries, you have every right to be suspicious. So this article is a fair description: what llms.txt actually is, what it demonstrably does today, what it doesn’t do yet — and when an hour spent on it makes sense. So you can answer the offer yourself, without feeling that the train is leaving without you.

What llms.txt is: a website signpost written for machines

llms.txt is an ordinary text file stored in the root of your website — at an address like yourdomain.com/llms.txt, right next to the robots.txt and sitemap.xml files you most likely already have. There’s no code inside: just the company name, a few sentences about what you do, and an organised list of your most important pages with a short description for each. The abbreviation LLM stands for large language model — the technology that ChatGPT, Copilot or Gemini run on.

The thinking behind it is sound. An ordinary web page is a mess to a machine: menus, footers, pop-up bars, scripts — the useful text itself is a fraction of what a crawler downloads. So the llms.txt proposal from autumn 2024 says: let’s give AI tools a clean text map of the website that immediately tells them what the company does and where the substance is. There’s also an extended variant, llms-full.txt, which contains the full text of the pages themselves.

The important word, though, is proposal. llms.txt wasn’t devised by Google, OpenAI or Microsoft — it came from an independent developer in the AI community, and the big players haven’t signed up to it so far. That makes it fundamentally different from the files you already have on your website:

What it actually does — and what it doesn’t yet

Now the thing the miracle salesman won’t tell you: not one of the major AI tool operators has officially confirmed that it reads and uses llms.txt. Google has even said openly that its systems don’t work with the file — one of its representatives compared it to the old “keywords” meta tag, into which websites used to list their keywords and which search engines learned to ignore. And in its official documentation on AI features, Google says the same thing in other words: no special file or tag is needed to appear in AI Overviews, the ordinary rules of search apply.

With ChatGPT the situation is much the same. Its search runs on the Bing index and OpenAI’s own crawlers — and you won’t find any binding mention of llms.txt in OpenAI’s guides. Websites that have deployed the file and watch their crawler access logs report the same so far: AI crawlers ask for it only rarely.

And one common misunderstanding, straight away: llms.txt is not a tool for forbidding crawlers anything. If you don’t want AI tools using your website, that’s handled in robots.txt. llms.txt is an invitation, not a lock — it blocks nothing and enforces nothing.

Does that mean it’s useless? Not quite. It’s just honest to know that today you’re buying a ticket to a performance nobody is sure will actually be staged.

Why to consider it anyway

Three reasons why an hour with llms.txt need not be wasted.

It costs you nothing and harms nothing. The file contains only public information you already have on the website. The worst possible outcome is that nobody ever reads it — and you’ve lost an hour.

If the proposal catches on, you’re ready. AI tools evolve fast, and what nobody reads today may be standard equipment in a year. It won’t be any kind of head start, but you won’t be playing catch-up.

It’s a useful exercise even without AI. Writing down on a single page what your company does, for whom, and which five pages of your website matter most — that’s a test many websites fail. If you find you have nothing to link to, because your services are described by one generic paragraph and there’s no pricing at all, you’ve just found the real problem. And no file will fix that one.

One sober note on top of that: when a language model loads your website directly — say at the moment a customer asks it about your company specifically — a clean text signpost can make its job easier. That’s the most likely scenario today in which the file gets used for anything. Whether models are being asked about your company at all, and what they say about you, you can check using the method in the article What ChatGPT says about your company — and how to find out.

How to set up llms.txt in an hour

You don’t need a programmer, just access to the website (or someone who will upload the file).

1. Write the content. The company name, two or three sentences about what you do and for whom, then sections with links: services, pricing, frequently asked questions, contact, and three to five of your best articles. One sentence of description for each link.
2. Save it as plain text named llms.txt and upload it to the root of the website — where robots.txt sits.
3. Check in a browser that the file opens at yourdomain.com/llms.txt.
4. Update it whenever your services, prices or key pages change. An out-of-date signpost does more harm than none at all.

Here’s what the skeleton might look like for a trades business:

# Novák Plumbing, Brno

> Plumbing work for households and apartment buildings in Brno
> and the surrounding area. Repairs, pipework refits, emergency
> call-outs.

## Services
- [Repairs and emergencies](https://www.priklad.cz/opravy/): what we
  handle and how fast we get there
- [Pipework refits](https://www.priklad.cz/rekonstrukce/):
  the process, indicative scope of work

## For customers
- [Pricing](https://www.priklad.cz/cenik/): indicative prices for work
- [FAQ](https://www.priklad.cz/faq/): answers to the most
  common questions
- [Contact](https://www.priklad.cz/kontakt/): phone, address

The file should contain only what you’re happy to show anyone — it’s public and competitors will read it too. And notice what that skeleton assumes: that your service pages, pricing and answers to questions exist and are worth linking to. Which brings us to the heart of the matter.

How to tell when someone is selling you a miracle

When the offer arrives, measure it against three questions.

Does it promise a result it doesn’t control? Nobody can honestly guarantee that AI Overviews will show you or that ChatGPT will recommend you — just as nobody can guarantee a search ranking. Anyone promising that is selling something that isn’t theirs.

Do they want a monthly retainer for a static file? llms.txt only changes when the website changes. A one-off job measured in hours, not a service invoiced every month.

Is it the first and only step they propose? That’s the most reliable signal. Whether AI mentions you is decided by content it has something to quote from, and by a verifiable trace of your company — reviews, mentions, consistent details in directories. How to write text an AI can pull an answer out of is covered in the article How to write content so AI quotes you. Anyone selling you a file instead of that work is selling you a sign — and doesn’t care that it points at an empty shop.

A sign belongs on a full shop

Let’s put the priorities in the order that makes business sense:

1. Service pages that answer the customer’s questions — what you do, for whom, how it works and roughly what it involves. What such a page looks like is shown in the guide How to write a service page that brings in enquiries.
2. Answers to frequently asked questions — exactly the format AI likes to pull from most. That’s covered in the article FAQ on your website: how to answer customers and AI alike.
3. A verifiable trace off your website — reviews, profiles, directories, the same name and phone number everywhere.
4. And then, that same evening if you like, llms.txt. An hour of work that adds a clean signpost to the finished arrangement.

If you don’t know where in that queue your company currently stands — whether AI tools know you at all, whether your website has anything worth quoting, and what specifically is holding visibility back — it’s faster to have it measured than to guess. That’s exactly what the initial analysis does: it goes through your website, the data from search engines and what AI tools say about your company, and returns a list of steps ordered by benefit. llms.txt will probably be on it. Just not on the first line.

And if you find that what’s missing is precisely the content AI could draw on, that’s a job for content creation and management — the indicative scope and prices are in the pricing.

A conclusion with no fog: go ahead and set up llms.txt, it costs an hour and one day it may come in handy. Just don’t let it be sold to you as a ticket into AI. The ticket is content — the file is only an arrow pointing the way to it.