Behind the build

What llms.txt is, and why your site should have one

A plain text file at your site's root stating what your business is, for AI tools to read. What goes in one, whether anything reads it, and the honest case.

The SiteRoot team7 min read

A robotic hand and a human hand reaching towards one another.

Add /llms.txt to the end of a website's address and you will sometimes get a plain text page describing what that business or product is. No design, no navigation, no pictures. Just a heading, a sentence of summary and a few short sections of facts. It is the least impressive-looking thing on the internet and it is genuinely useful.

That file is llms.txt, and the idea is simple: a small, plain, human-readable file that sits at the root of a website and states in ordinary words what the thing is, what it offers, who it is for and how to reach it, written so an AI tool reading it does not have to work any of that out from the design. It is an open convention rather than an official standard, published at llmstxt.org in 2024, and no major AI company has committed to reading it. That last part matters and we will come back to it, because the honest case for publishing one is not the one usually made.

What is llms.txt, exactly

It is one plain text file, written in simple markdown, served at yourbusiness.com.au/llms.txt.

The shape it asks for is short: a heading with the name of the thing, a one line summary written as a quote, then optional sections under ordinary headings and usually a list of links to the important pages. That is the whole specification, small enough to write by hand in twenty minutes.

It is text rather than something more structured on purpose: it is meant to be read by a language model, and a model reads prose better than anything else. There is no syntax to learn and nothing to validate.

What goes in one for a small business

The published convention is aimed at software products with documentation. For a plumber in Frankston, a bare list of links to your six pages is useless: anything that can fetch your llms.txt can fetch your pages anyway.

So the version worth publishing carries facts, not just links. For a small business that means:

  • The business name, as a heading.
  • A one line summary: what you are and where. "A family-run plumbing business in Frankston."
  • A short paragraph about what the business actually does and how it works.
  • Your services, as a list, in the words customers use.
  • Your service area, written out.
  • Who you work with, if that narrows things usefully. Homeowners, builders, strata managers.
  • Contact details: phone, email, the website address.
  • A list of the pages, with real labels.

That is a file an assistant can answer a question from directly: who does bathroom renovations near Frankston, are they open Saturdays, what is the number.

Why a separate file, when the facts are already on the site

Two honest reasons, and one of them is not about AI at all.

The first is cost of reading. Your home page is thirty kilobytes of layout, navigation, a cookie notice, a cascade of styles and a footer of links, wrapped around maybe two hundred words of actual information. Anything reading it has to find the two hundred words in all of that. An llms.txt file is the two hundred words with nothing else attached. It is cheaper to read and there is less to misread.

The second is that the facts are usually not all in one place. Your hours are on the contact page, your service area is implied by a footer list, your specialities are spread across three service pages, and the sentence that best describes your business does not exist anywhere because nobody ever had to write it. Making the file forces someone to write that sentence down, which plenty of owners find the most useful part of the whole exercise.

Is anything actually reading it

As of now, no major AI company has published a commitment to reading llms.txt. Some tools and some crawlers look for it. Plenty do not. Anyone telling you that publishing one will get you recommended by ChatGPT is making it up, and the mechanism they are describing does not exist.

So the case for having one has to stand on its own, and it does, on three grounds. It costs almost nothing: a few hundred words, published once. The downside is zero. And conventions in this area have a history of being ignored until suddenly they are not, at which point the people who already had the file simply had it.

That is a modest case rather than an exciting one, and we would rather make the modest one accurately.

Does it help me with Google

Not directly, no. Google does not use it and has not said it will, and you should treat anyone claiming otherwise the same way you would treat a promised first position.

The indirect effect is real but unglamorous: writing one forces you to state your business facts plainly in one place, and those same sentences belong on your actual pages, where search engines definitely are reading.

How is this different from robots.txt

They are both small text files at the root of a site and that is where the similarity ends.

robots.txt is an instruction: it tells crawlers which parts of your site they may look at. It is about permission, it is widely honoured, and it has been around since the nineties.

llms.txt is a description: it tells a reader what your business is. It grants nothing and blocks nothing. If you want to control which AI crawlers may read your site at all, that is a robots.txt job, and it is a separate decision from whether you publish an llms.txt.

The two get confused constantly. Publishing an llms.txt does not opt you in to anything, and not publishing one does not opt you out.

What it is not

It is not a submission form, and there is nowhere to submit it. And it is not a replacement for having readable pages: if your website's words only appear after a script runs, so that what a crawler receives is an empty shell, an llms.txt file is a good summary attached to a site nothing else can read.

How do I make one

By hand. Open a text editor, write the sections above in plain sentences, save it as llms.txt and put it in the same place as your home page. Check it by visiting yourbusiness.com.au/llms.txt in a browser: if you see the text, it is live.

The real difficulty is not writing it. It is the second year. The file states your hours, your services and your phone number, and those things change. A file that has quietly been wrong since you dropped Saturdays is worse than no file, because now you are handing a machine a confident, incorrect fact about your own business. If you write one by hand, put a reminder in your calendar to reread it every six months, or accept that it will go stale the way the copyright year does.

What SiteRoot does

Every site we build publishes an llms.txt file, written from the answers you gave when the site was built.

It carries the name, a one line summary with your industry and location, a paragraph about what you do, your services, your service area, who you work with, your contact details, and a labelled list of every page. It is rewritten every time you publish a change, so editing your hours or adding a service updates it without anyone having to remember. That is the part that solves the year-two problem.

You can look at one right now: add /llms.txt to the end of priority-plumbing.siteroot.com.au or getset-lending.siteroot.com.au and read what is there. It is public, it is meant to be, and it shows more than another page of explanation would.

Alongside it, each site publishes a file listing every page for search engines, refreshed on the same schedule, and each page carries a block of facts naming the specific business type Google publishes for that trade. Three different readers, the same facts, one source.

None of this is a promise that an AI assistant will recommend you. Nobody controls that and nobody should say they do. It means that when something goes looking for a plain statement of what your business is, there is one there, and it is current.