An llms.txt example, line by line, with three real files compared

An llms.txt example, line by line

Below is a complete llms.txt file for a made-up company, Example Bakery Supplies, which sells ovens to small bakeries. The company and every line of the file are invented to show the format.

Because this page is not a code editor, the file is shown as a table. Each row is one line of the file. The second column names the character the line begins with, and the third gives the text that follows it. Blank lines are left out of the table, but the file needs one between each part.

Read top to bottom, the file answers three questions in order. What is this site? What does it sell, to whom, and for how much? Where do I go for each topic? An AI tool that reads only the first five lines already has the facts most buyers ask about.

One link line, in three parts

Every link line has the same three parts on one line: a name in square brackets, an address in round brackets straight after it, and an optional note after a colon. The line begins with a dash and a space, like an ordinary Markdown list.

The addresses end in .md because the proposal suggests offering a clean Markdown copy of each page at the same address with .md added. That is optional. A link to the ordinary page works too, and many real files do just that.

The rules each line follows

The proposal at llmstxt.org sets out the parts in a fixed order, and only the title is required. Everything else is a recommendation.

1. The title comes first: one hash, a space, and the name of the site or project.
2. A summary may follow as a quoted paragraph, starting with a greater-than sign. It should hold the key facts needed to understand the rest.
3. Any number of plain paragraphs or lists may follow, with no headings. This is where details and caveats go.
4. Then any number of sections, each opened by a second-level heading with two hashes, each holding a list of links.
5. A section headed Optional may come last, for links a reader can skip.

The Optional section changed meaning in version 2 of the proposal, published in August 2026. The changes page says it is still allowed and still a useful convention for secondary links, but it no longer tells any tool what to leave out. Version 2 also lets a page point to its llms.txt file with a standard link tag, and lets a file in a folder cover only the pages in that folder.

The summary line does the most work. If a tool reads only one line after the title, make it the one that says what you sell, to whom and for how much.

Adapting the example to your kind of site

The shape stays the same for any site; what changes is which facts go in the summary and which pages earn a section. The table gives a starting point for four common kinds of site.

The proposal sets no length limit. A useful test is whether a person could read the whole file in two minutes and come away knowing what you do. If the file has grown into a full index of every page, a short llms.txt pointing to a few key sections will serve a reader better.

The details paragraph is the place for the facts people get wrong. If buyers often think you sell to consumers when you sell only to businesses, say so there in one plain sentence. It is the kind of fact a reader can easily miss when it is spread across several pages.

Three real llms.txt files, compared

Real files follow the title rule and then go their own way. The three below were fetched on 10 October 2026, and each is public.

The llmstxt.org file is the cleanest model. Its summary says, in one sentence, that it is a proposal to add a /llms.txt file to websites, and that the file is Markdown giving background and links to more detailed Markdown files. Its one section lists three links, each with a note.

Stripe's file shows a different use. Its opening paragraphs are instructions to coding agents, such as checking the package registry for the latest version of Stripe's library rather than relying on a remembered version number. It even has a section headed Instructions for Large Language Model Agents. For a developer tool, the file is a place to steer the agents that write code against it.

Anthropic's file is mostly an index. It lists 762 links to English documentation pages, and names eleven other languages with a pointer to the website instead. Where it adds a note, it puts it after a spaced hyphen rather than a colon, which is a small departure from the proposal. It is useful to an agent looking for one page and too long to read as a summary.

Mistakes to avoid

Most problems with llms.txt files are the ordinary problems of any document: it is wrong, it is out of date, or it cannot be fetched. Five are worth checking for.

• A summary that does not say what you sell. A line of brand language tells a tool nothing it can repeat.
• Facts that disagree with your pages. An old price in llms.txt is worse than no llms.txt, because a tool that reads it gets the wrong figure.
• Links to pages that have moved. Every link should open, and ideally lead to text rather than a page built only with scripts.
• A file that cannot be fetched. Chrome's Lighthouse audit flags a page when a server error occurs while fetching the file.
• Expecting it to replace robots.txt. An llms.txt file cannot block or admit a crawler; only robots.txt does that.

The first two mistakes have one cure: build the file from the same source as your pages, so a price changes in one place. The Lighthouse point comes from its llms.txt audit page, which also says a missing file is marked not applicable rather than failed. The last point matters more than the file itself. Our page on AI crawler access explains which crawlers to let in.

How to check your own llms.txt file

A five-minute check catches nearly every problem above. You need a browser and, if you like, the command line.

1. Open yoursite.com/llms.txt in a browser. It should show plain text, not a web page or an error.
2. Check that the first line is one hash, a space and your name, with nothing before it.
3. Read the summary and ask whether a stranger would know what you sell, to whom and for how much.
4. Open every link and confirm it loads the page the note describes.
5. Compare each price, plan name and product name with the live page that states it.
6. Run Lighthouse in Chrome on your home page and read the llms.txt result in the agentic browsing section, which Google marks as experimental.

The file can be served as plain text or as Markdown. On 10 October 2026, llmstxt.org served its own file as text/plain and Stripe served its file as text/markdown, so either works in practice.

What the file will and will not do

An llms.txt file helps a tool that chooses to read it, and there is no public evidence that the main assistants read it before deciding whom to name. Treat it as tidy housekeeping, not as a way to be recommended.

Google's guide to optimizing for generative AI says Google Search ignores such files and that they neither help nor harm a site. An Ahrefs study of 137,210 domains found that 97 per cent of the llms.txt files it saw got no requests in May 2026. Our companion post, what is llms.txt, goes through that evidence in full.