llms.txt: what it is, and whether you need one

What it is

The idea comes from the llmstxt.org proposal: put a file at /llms.txt that an AI tool can read in a single pass, instead of crawling a whole site and guessing what matters. It is a convention, not a standard that any engine is required to follow.

The format is deliberately simple Markdown, in a fixed order:

• A single H1 with the name of the site or business.
• A short blockquote summarising what it is, in a sentence or two.
• Optional paragraphs of plain context.
• H2 sections, each a list of links with a one-line description of what is on the page.

A minimal example

# Harbour Lane Dental

> Private dental practice in West London offering implants, Invisalign and
> composite bonding. Established 2009.

## Treatments

- [Dental implants](https://example.co.uk/implants): what is involved, timescale, and prices from £2,400.
- [Invisalign](https://example.co.uk/invisalign): suitability, treatment length and cost.

## Practice

- [Opening hours and location](https://example.co.uk/contact): Monday to Friday, 8am to 6pm.
- [Our dentists](https://example.co.uk/team): qualifications and GDC registration numbers.

Write it the way you would brief a new colleague: what you are, what you do, and where the answers are. State facts a buyer asks about — prices, hours, locations, qualifications — because that is what the file is for.

How to publish it

• Serve it as plain text at the root: https://yourdomain/llms.txt.
• Keep it short. It is an index, not a copy of the site.
• Keep it true. Update it when a price, address or service changes; a stale llms.txt is a stale fact with a machine-readable label.
• Do not block it in robots.txt.

Ours is at /llms.txt, if you want a working example of a site describing itself.

What it will and will not do

As far as we can tell, no major engine has publicly committed to reading llms.txt when it decides what to name. So we treat it as good hygiene rather than a lever: our audit records whether you have one, and shows it, but it does not move the score.

• It will not fix a robots.txt that blocks the crawlers. Stage one still comes first.
• It will not make a thin page answer a question. The pages it points to have to be worth reading.
• It may help tools that do read it find the right pages quickly, and it forces you to write down, in one place, what the business is. That second benefit is real even if no engine ever reads the file.

Where it fits in the order of work

Do it after the things that decide whether you are read at all — crawler access, a working sitemap, a page for each real question — and before nothing in particular. It takes about half an hour, which is why it appears in most plans as a small, cheap job rather than a priority.