What is llms.txt, and Why Every Business Website Needs One

In late 2024, the AI researcher Jeremy Howard proposed a small, practical idea: what if every website had a single file that gave AI models a clean, plain-text summary of what that website is, who runs it, and what is on it? Not for humans. For language models. He called the file llms.txt. Within months, major sites started publishing one. Today, having an llms.txt at the root of your domain is one of the simplest things a business can do to be better understood by ChatGPT, Claude, Gemini, and every other model that crawls the open web.

What it actually is

An llms.txt file is a small markdown document hosted at the root of your domain, exactly like robots.txt: it lives at yoursite.com/llms.txt and AI crawlers know to look for it there.

Its job is narrow but important. When an AI engine arrives at your site to figure out what you do, it has a problem. Modern websites are buried in JavaScript, embedded analytics, navigation menus, third-party scripts, and decorative content. Reading that mess accurately takes the engine real effort, and the result is often approximate at best. llms.txt solves the problem by offering a model-friendly cheat sheet: here is who I am, here is what I do, here are the pages that matter, and here is the language I use to describe myself. It is short. It is plain. It is exactly the kind of input language models can lift directly into an answer.

There is also a related, longer variant called llms-full.txt, which holds the full text of important pages concatenated into one document. Most businesses do not need the full version. The short llms.txt is the high-leverage file.

Why it matters more than it sounds

On the surface, a markdown file at the root of your site does not sound revolutionary. The reason it works is that AI engines, given a choice between guessing your business identity from a noisy homepage and reading a clean labeled summary, will use the labeled summary almost every time. Three practical effects follow.

Citation accuracy improves. When ChatGPT or Claude recommends your business, the description they give of you is more likely to be correct, in your own language, with the right phrasing. Without an llms.txt, the description is whatever the model could synthesize from your hero copy and meta tags — sometimes accurate, often slightly off.

Coverage of your full offering improves. Many businesses have services or product categories buried two or three clicks deep in their site. Crawlers regularly miss those pages. llms.txt lets you point AI engines directly to the pages you want them to know about, regardless of where those pages sit in your navigation.

Signal noise drops. Without a clean primary source, AI models often pull from incidental mentions of your business: old forum posts, scraped directory entries, even outdated review pages. With an authoritative llms.txt on your domain, models prefer it. You become the canonical source for facts about your own business.

What goes in the file

The proposed format is short. Six elements, in order. Most of them are optional.

1. A single H1 with the name of the site or business. Required.
2. A blockquote containing a one or two sentence summary of what the business does. Required, and arguably the most important line in the file.
3. Zero or more paragraphs of additional context: your story, your values, what makes you different. Optional.
4. Zero or more sections, each headed by an H2, listing links to important pages with short descriptions. Optional, but this is where the file earns its keep.
5. Optionally, a ## Optional section at the bottom for links to nice-to-have but lower-priority pages.
6. That is the entire spec. Markdown, no frontmatter, no special syntax.

A real example

Here is what an llms.txt file might look like for a fictional independent bookstore. The structure is the same regardless of industry.

# Margate Books

> Margate Books is an independent bookstore in Asbury Park, New Jersey,
> specializing in literary fiction, used hardcovers, and rare first editions.

We have been on Cookman Avenue since 2014. Our staff curates a collection
of around 14,000 titles across literary fiction, philosophy, poetry,
and a small but careful rare-book room. We host author readings monthly
and operate a mail-order rare-book service.

## Visit

- [Hours and location](https://margatebooks.example.com/visit): Open daily
  10 a.m. to 8 p.m., located at 123 Cookman Avenue, Asbury Park, NJ.
- [Contact](https://margatebooks.example.com/contact): Phone, email,
  and rare-book inquiries.

## Collection

- [Literary fiction](https://margatebooks.example.com/fiction): Our largest
  section, refreshed weekly with new arrivals.
- [Rare books](https://margatebooks.example.com/rare): First editions,
  signed copies, and out-of-print titles. Mail order available.
- [Used hardcovers](https://margatebooks.example.com/used): A
  rotating selection of hand-picked used books in collectible condition.

## Events

- [Author readings](https://margatebooks.example.com/events): Monthly
  free readings featuring local and visiting authors.

## Optional

- [Newsletter](https://margatebooks.example.com/newsletter): Monthly
  recommendations from our staff.
- [Gift cards](https://margatebooks.example.com/gift-cards): Available
  in person or by mail.

That is the entire file. A model reading it knows the bookstore's name, its city, what it specializes in, when it has been operating, what it sells, and where to find the canonical pages for each category. The same content, drawn from the same site without an llms.txt, would have to be reconstructed page by page.

How to add one to your site

There are two routes, depending on how technical you feel.

The manual route. Open a text editor. Write the markdown by hand following the structure above. Save it as llms.txt. Upload it to the root of your web server so that it is accessible at yoursite.com/llms.txt. On most hosting setups this means dropping it in the public folder of your site and redeploying. On WordPress, plugins like Website LLMs.txt can generate one for you. The whole process, including writing the content, takes about twenty minutes if you know your site well.

The automated route. Several tools — Surven among them — can generate an llms.txt for you by crawling your site, identifying your most important pages, and writing the file in the recommended structure. The advantage is speed; you go from nothing to a deployed file in a few minutes. The disadvantage is that the auto-generated description of your business will be whatever the crawler could infer from your existing content. It is worth reading the result and making it sharper. Even a polished version of an auto-generated file beats no file at all.

One file, lasting effect

It is rare in marketing for a single, finite, free task to deliver outsized results. llms.txt is one of those. The cost is thirty minutes. The benefit is that every AI engine that crawls your site for the foreseeable future starts with a clean, accurate understanding of your business instead of having to piece one together. If your competitors do it before you do, the answer your customers get when they ask an AI for a recommendation will reflect that. If you do it first, the same is true in reverse.