llms.txt for Your Website: What to Put in It, and What It Does Not Do

Site audit tools have started to mark a missing llms.txt file, and owners are asking whether they need one. The file is easy to write. The harder part is knowing what belongs in it and what to expect from it afterwards.

This article covers both. It is written on 13 October 2026, and the evidence in it is the evidence available on that date.

What llms.txt is

llms.txt is a Markdown file that sits at the address /llms.txt on your site. It gives AI assistants and agents a short summary of the site and a list of links to its most useful pages.

Jeremy Howard proposed it on 3 September 2024, and the format is documented at llmstxt.org. The reasoning is simple. A web page carries menus, scripts, banners and footers, and an assistant has limited room for text. A plain file that says "this is who we are, and these are the pages that matter" is cheaper to read than the whole site.

It is a proposal. No standards body has adopted it, and no AI company is obliged to read it.

What it is not

• It is not robots.txt. It allows nothing and blocks nothing. Access for AI crawlers is still set in robots.txt.
• It is not a sitemap. A sitemap lists every page. llms.txt lists the few pages you would hand to someone who has ten minutes.
• It is not a ranking signal. Nothing in it moves a page up in Google.
• It is not a place for sales copy. An assistant needs facts it can repeat, and slogans give it none.

Does anything read it yet

Mostly no, and it is better to know that before you spend time on it.

• Ahrefs looked at about 137,000 domains for May 2026. 97% of the llms.txt files it found received no requests at all that month.
• Of the requests that did arrive, about 1% came from AI retrieval bots linked to ChatGPT and Perplexity. About 10% came from coding agents.
• Ahrefs counted requests. It did not measure whether anything was done with the file after it was fetched.
• John Mueller of Google wrote in June 2025 that no AI system was using llms.txt, and compared it to the keywords meta tag.

There is one sign of movement. Lighthouse 13.3 includes an llms.txt audit in an experimental category called Agentic Browsing. As reported by DebugBear, it checks three things: the file has a top-level heading, it is longer than 50 characters, and it contains at least one Markdown link.

What to put in it

The format has a fixed order. Only the first item is required, but a file with nothing else is of little use.

1. The name. One top-level heading with the name of the site or company. This is the only required part.
2. A short summary. One or two sentences in a blockquote: what you do, for whom, and where. Write it so it still makes sense when it is the only thing read.
3. A few plain facts. A short list with no headings: founder, location, what you sell, how you charge, how to reach you. These are the details an assistant would otherwise guess.
4. Sections of links. Each section has a second-level heading and a list. Each item is a Markdown link, then a colon, then a note that says what the page answers.
5. Your key pages. Services or products, pricing, case studies, documentation, policies, and contact. Use full addresses that start with https.
6. An Optional section. A section named Optional, placed last, holds links an agent may skip when it is short of room. Blog posts fit well here.

The proposal also suggests a Markdown copy of each important page at the same address with .md added. This matters most for documentation. For a company site the file alone is a fair start.

An example you can copy

Here is a complete file for an invented furniture shop. Change the names, facts and links, and keep the order.

# Northfield Furniture

> Solid wood furniture made to order in Leeds, delivered across the UK. Dining tables, beds and shelving, with a 10 year guarantee.

- Founded in 2012. Workshop and showroom in Leeds.
- Every piece is made to order. Delivery takes 6 to 8 weeks.
- Prices include UK delivery. Returns are accepted within 30 days.
- Contact: hello@northfield.example, 0113 496 0000.

## Products

- [Dining tables](https://northfield.example/dining-tables): Oak and walnut tables in four sizes, with prices
- [Beds](https://northfield.example/beds): Bed frames in UK standard sizes
- [Shelving](https://northfield.example/shelving): Wall and free-standing shelves

## Buying

- [Delivery](https://northfield.example/delivery): Areas covered, lead times and costs
- [Returns and guarantee](https://northfield.example/returns): The 30 day return policy and what the guarantee covers
- [Contact](https://northfield.example/contact): Showroom hours, phone and email

## Optional

- [Wood care guide](https://northfield.example/blog/wood-care): How to clean and oil solid wood
- [Oak or walnut](https://northfield.example/blog/oak-or-walnut): The differences in colour, hardness and price

Mistakes to avoid

• Plain addresses in place of Markdown links. The Lighthouse audit fails a file that has no link written as name.
• Listing every page. A file with hundreds of links is a second sitemap, and it defeats the purpose.
• Links with no note. "Services" tells an assistant nothing. "Services: what we build and what each costs" does.
• Facts that differ from the site. If the file says delivery takes two weeks and the delivery page says six, an assistant may repeat either one.
• Writing it once and leaving it. Prices, products and addresses change. An old file is worse than none.
• Serving it as an HTML page. It should arrive as plain text, with no site menu or cookie banner around it.
• Linking to pages that robots.txt blocks, or that need a login.

How to add it to your site

• Static site or custom build: put a file named llms.txt in the public folder, so it is served at yourdomain.com/llms.txt.
• WordPress: several SEO plugins can now generate the file. Read what they produce and trim it, because the default is often every post.
• Shopify: you probably have one already. On 13 October 2026 we opened /llms.txt on three Shopify stores, and each returned a file that Shopify had generated, titled "Agent Instructions". It tells shopping agents how to browse and buy from the store. Open yours before you install an app for it.
• Next.js, Laravel and similar: build the file from the same data that builds your pages. Then it changes when the site changes.

The last option is what we use. Our own file at keviontechnologies.in/llms.txt is generated from the data behind our pages and our sitemap. It lists services, industries, case studies, extensions and company pages, and puts blog posts under Optional. Nobody has to remember to update it.

To check yours, open the address in a browser. You should see plain text that starts with a line beginning with one # sign. Then run Lighthouse and look at the Agentic Browsing category.

Do the basics first, because they are the things AI tools and search engines demonstrably read: clear page text, a sitemap, correct robots.txt rules, and a filled-in title and description on each page. Our article on a missing meta description on a Shopify home page covers one of those, and our guide to AI shopping agents covers what a store needs for agents that buy.

How Kevion can help

Kevion Technologies is a senior software engineering team in Ahmedabad, India, founded by Yash Patel. We build websites and web applications, and we add AI features to business software.

If you want an llms.txt that is generated from your site's own data and stays correct, we can set that up as part of web application work or an AI project. We will also tell you plainly when a file like this is not worth your money yet.

Send us your site address and we will reply with what we would put in the file and what we would fix first.