llms.txt is a short, plain-text guide to your website, written for AI tools. It says who you are, what
you do, and which pages matter most, in a form a language model can read in one go. It's a new standard and a small
job. Here's what goes in it, who reads it, and how to publish yours.
Where it came from
Jeremy Howard proposed llms.txt in September 2024. His reasoning was simple: an AI tool answering a
question can only read so much at once, and a whole website is usually too much. Menus, footers, scripts and
cookie banners crowd out the few sentences that actually describe you. One small file at a known address gives the
tool the summary first and the links to follow when it needs more.
The file is meant to be read when someone asks a question, while the AI is working, rather than collected to
train future models. It lives at the root of your site, at yoursite.com/llms.txt, the same way
robots.txt does.
What it looks like
The file is written in Markdown, the same plain-text format many notes apps use. A small business needs only three parts:
# Meadowbrook Bakery
> A neighborhood bakery in Willow Creek baking naturally-leavened sourdough, pastries, and custom cakes since 1984.
Contact: hello@meadowbrookbakery.com · 503-555-0100
## Pages
- [Home](https://meadowbrookbakery.com/)
- [Menu](https://meadowbrookbakery.com/menu): breads, pastries and prices
- [Custom cakes](https://meadowbrookbakery.com/cakes): how to order, with 48 hours' notice
- A title. Your business name after a single
#. This is the only part the standard requires. - A summary. One or two sentences after a
>, saying what you do and where. Any extra details, like how to get in touch, go in plain paragraphs below it. - Your key pages. A list under a
##heading. Each line is a link, then a colon and a few words on what the page is for.
You can add more lists under more headings. A list headed ## Optional is for pages an AI tool can
skip when it's short on room.
Who reads it
Thousands of sites publish one, including the developer docs of OpenAI, Anthropic and Google's Gemini. Many documentation platforms create one automatically, and AI coding tools fetch them to read up on a library before using it.
The big assistants are a different story. None of ChatGPT, Claude, Gemini or Perplexity has said it reads the
llms.txt files of the businesses it answers questions about. So treat yours as an inexpensive bet: it
takes about fifteen minutes, costs nothing, and puts an accurate summary where AI tools that do look will find it.
Google's Lighthouse, the audit Bloomlight and many other tools run on your pages, now checks the file in its new agentic browsing section. Google describes that section as still under development.
What Lighthouse checks
If you don't have an llms.txt, Lighthouse marks the check as not applicable and your score doesn't
change. If you do have one, it fails the check when the file:
- has no title line starting with
#, - has no links in
[name](address)form, - is shorter than 50 characters, or
- can't be loaded because your server returns an error.
A file with a title, a summary and one page link passes all of them.
Writing a good one
- Write the summary for a stranger. Say what you do, for whom, and where, the way you'd explain it to a new customer on the phone.
- List the pages people ask about. Your menu, services, prices, hours, booking page and contact page. Five to ten is plenty for most small businesses.
- Use full addresses. Start every link with
https://, so it works wherever the file is read. - Keep it true. Update it when your hours, services or pages change. An out-of-date summary is worse than none.
Publishing it
Save the file as llms.txt and upload it to the root of your website, so it opens at
yoursite.com/llms.txt. Visit that address afterwards to check it shows as plain text. If your website
builder doesn't let you upload files to the root, ask its support team or your developer how to add one.
llms.txt can't block AI crawlers or control what they collect. That's the job of
robots.txt.
How Bloomlight helps
The free llms.txt builder writes
the file for you from a short form, in a format that passes Lighthouse's checks. Bloomlight's audits
flag a missing or broken llms.txt as a finding, with the builder ready inside it. It already knows
your monitored pages, so most of the list fills itself in.