Most business owners have heard of robots.txt, the small file that tells Google's crawler which parts of a site it can index. Far fewer have heard of llms.txt, its newer cousin, which does something more useful: it gives AI models a short, plain language summary of what your business actually is, before they even start crawling the rest of your site.
It is not yet a formal, universally adopted standard the way robots.txt is. But it costs nothing to add, carries no downside, and in every AI Score test run completed so far, the businesses tested had no llms.txt file at all. That is a genuinely open door right now, and it will not stay open forever.
What llms.txt actually does
An AI crawler visiting your site has to work out what you do, where you operate, and who you serve, by parsing whatever HTML it finds. That works reasonably well for clean, text-based pages. It works badly for sites that lean on images, JavaScript-rendered content, or marketing copy with no concrete facts in it.
llms.txt sidesteps that problem. It is a plain markdown file placed at the root of your domain, at yourdomain.com/llms.txt, written specifically for a language model to read rather than for a human browsing your homepage. Think of it as a one page briefing document you would hand a new member of staff on their first morning: who you are, what you offer, where you are, and which pages matter most.
The structure to follow
The emerging convention is simple and forgiving. There is no strict schema to break, which makes this low risk to get started on today.
- An H1 with your business name. Plain and unambiguous, exactly as you want it quoted back.
- A one line summary underneath. What you are, where you are, and who you serve, in a single sentence a model could lift directly into an answer.
- A short "About" section. Two or three sentences of real, specific detail: founding date, key facilities, notable credentials, what makes you different from the obvious alternative down the road.
- A list of key pages with links. Pricing, opening hours, services or courses, booking, and contact, each with a one line description of what a model will find if it follows the link.
- An optional FAQ section. Straight answers to the questions your customers actually ask, in the same plain phrasing they would use talking to a person.
What good content looks like inside it
The file only earns its place if what's inside it is specific. "We are a premier destination offering excellent service" tells a model nothing it can use. "18 hole championship course, four hole short course, green fees from £45 weekday and £65 weekend, driving range open 7am to 8pm" gives it something to quote with confidence. The same logic applies whether you run a golf club, a garage, or a dental practice: facts beat adjectives, every time, because facts are what a model can safely repeat without risking an inaccurate answer.
Where to put it, and how long it should be
It sits at the root of your domain, not inside a subfolder, so it is discoverable the same way robots.txt is. Keep it genuinely short. A few hundred words covering the essentials outperforms a long document trying to cover everything on the site, because the whole point is a fast, dense summary rather than a duplicate of your homepage copy.
What it will not do
llms.txt will not override weak underlying signals. If your reviews are thin, your schema markup is broken, or your site genuinely doesn't answer the buyer's question anywhere, a summary file at the root of your domain won't fix that on its own. It is one input among several, most usefully paired with clean crawlable content, accurate structured data, and a plain language FAQ addressing the specific questions your customers actually ask an AI assistant.
Where to start
Write the summary sentence first. If you can't describe your business in one specific, factual sentence, that's worth fixing on the homepage as well as in llms.txt. From there, list your five or six most important pages, write one honest line about each, and publish the file. It's a same afternoon job with no technical dependency, which makes it one of the easiest fixes on any AI Score report.