llms.txt: what it does and doesn't do

Where did llms.txt come from?

It was proposed in 2024 as a convention, documented at llmstxt.org. The idea is simple. Websites are built for people and browsers, with menus, scripts and layout. A language model trying to understand a site would do better with a short, clean index written in markdown.

So the proposal says: put a file at yourdomain.com/llms.txt that starts with the site name, a one-line summary, and then lists the pages that matter, each with a short note.

A minimal version for a trade business might look like this:

# Example Plumbing Co.

> Licensed residential plumber serving Long Beach, Lakewood and Signal Hill. Repairs, water heaters, drain cleaning, repiping.

## Key pages

- [Business facts](https://example.com/facts): name, license, hours, service area, contact
- [Water heater repair](https://example.com/water-heaters): tank and tankless, repair and replacement
- [Drain cleaning](https://example.com/drains): kitchen, bath and main line
- [Service area](https://example.com/areas): cities and neighborhoods we cover

That is the whole idea. No special code, no tags, no tracking.

What does llms.txt actually do?

On its own, it does nothing until something reads it. It is not like a sitemap, which search engines officially support. It is not like robots.txt, which crawlers widely respect.

Where it can help:

• Tools that choose to read it. Some developer tools, documentation sites and AI coding assistants use llms.txt files to find the right pages quickly.
• People pasting your site into an assistant. A clean summary is easy to hand to a model when you want it to understand your business.
• Your own discipline. Writing one forces you to decide which pages matter and to describe them in one line. That exercise alone often shows gaps.

What doesn't llms.txt do?

This is where a lot of advice online goes too far.

It does not get you recommended. OpenAI, Google, Anthropic and other major assistant makers have not publicly stated that they use llms.txt to choose which businesses to name. Anyone claiming it boosts your AI visibility is guessing.

It does not replace your website. Assistants that search the web read your actual pages, profiles and listings. If those are unclear or inconsistent, a tidy text file will not fix it.

It does not control access. If you want to limit which AI crawlers fetch your site, that is a robots.txt question. llms.txt neither blocks nor permits anything.

It is not schema. Structured data like LocalBusiness and FAQPage is an established way to state facts in a machine-readable format. llms.txt is a reading list, not data. For the data side, see our schema checklist for contractors.

Should a local business add one?

If it takes you twenty minutes, sure. It costs nothing, it does no harm when accurate, and if more tools adopt it later, you are ready. We have one on this site for the same reason.

But put it near the bottom of your list. For a plumber, electrician or HVAC company, the things that clearly affect what AI assistants say are:

1. Clear service pages in plain text.
2. Consistent name, address, phone and hours everywhere.
3. A complete Google Business Profile with real reviews.
4. A business facts page that states the basics in one place.
5. Valid schema markup.

If those are not done, llms.txt is a distraction. If they are done, llms.txt is a small, sensible extra.

How do you write a good llms.txt?

Keep it short and true.

• Start with one plain sentence that says who you are, what you do and where.
• List ten pages or fewer. Your facts page, main services, service area, pricing or how pricing works, and contact.
• Write one honest line per page. What a reader will find there, not a slogan.
• Use full URLs so the links work wherever the file is read.
• Update it when pages change. A file that points to deleted pages is worse than none.

Avoid stuffing it with keywords or city lists. It is a summary, not a sales page.

How do you know if it is being read?

Check your server or hosting logs for requests to /llms.txt. Some hosts show this in their analytics. You may see a few hits from tools and crawlers. That tells you something fetched it, not that it changed any answer.

The better test is the one that matters: ask the assistants your customers use the questions those customers ask, and see whether the answers are correct. Our free AI visibility check does that with live web search on.

So is llms.txt worth it?

llms.txt is a neat idea with limited proof behind it so far. Add one if it is easy. Then spend your real effort on the pages, profiles and facts that assistants clearly do read.

If you want help sorting the high-impact work from the low-impact work, that is what our AI search (GEO) service is for. We start by checking what assistants actually say about you, then fix it at the source.