llms.txt: a plain summary of your site for AI assistants
Each published Anticode site answers at /llms.txt with a short Markdown summary written for AI assistants: what the business is, what it offers and at what price, how to get in touch, and a link to every page and CMS entry. It's assembled from the published pages each time it's requested, so it can't fall out of step with the site.
Why a site might want one
llms.txt is a proposed convention, documented at llmstxt.org, for putting a Markdown file at the root of a site so that a language model, or a tool working for one, can understand the site in one request instead of crawling and parsing every page. It isn't a ranking signal and no assistant is obliged to read it. What it offers is a clean, factual description in the business's own words, at an address anyone can predict. An automatic llms.txt for every site means nobody has to remember to write it, or to update it after the menu changes.
What's in the file
The file is built in a fixed order. Sections only appear when the site has something for them:
- Title and summary
- The site's name as a heading, its meta description as a quote, and the main headline if it says something different.
- Details
- The at-a-glance facts the home page shows near the top, such as hours and where to find the business.
- What's offered
- Each item from the site's price list or product cards, on any page: its name, its description and its price as written on the site. Up to 40 items.
- Get in touch
- Each form on the site, described in plain words (book a time, send a message, join the newsletter, ask for an account) with a link to the page it's on.
- Pages
- A link to the home page, each section that has a heading, each separate page, and each CMS collection with how many entries it holds.
- One section per collection
- Every entry linked by its title, with its meta description where you've written one. Up to 200 per collection.
- Optional
- Links to the sitemap and to the site's live security report.
Text from the pages is stripped of markup, has brackets removed and is trimmed to a sensible length, so nothing a visitor or an owner types can break the Markdown or smuggle in a link.
How it's built
When something requests /llms.txt, Anticode reads the published home page and any separate pages, picks out the parts listed above from their structure, and reads the project's CMS collections from the database. The result is cached for a few minutes. Because nothing is stored, publishing a new price list or adding a collection entry changes the file without any action from you. It's served as Markdown with open cross-origin access, so any tool can fetch it.
This works because the design engine writes pages with a consistent structure: sections with ids and headings, price lists as lists, facts in their own block. The same structure is what makes the sitemap and the generated legal pages possible.
Making the most of it
- Write a real meta description. It becomes the summary line at the top, the first thing a model reads.
- Put prices on the page. If your services and prices are in a price list, they appear under What's offered. If they're only in a PDF or an image, they don't.
- Fill in entry descriptions. CMS entries with a meta description are listed with it; those without are listed by title only.
- Keep section headings descriptive. "Prices for regular cleaning" tells an assistant more than "Pricing".
For the wider picture of how assistants find and cite small sites, see the guides on llms.txt and on getting cited by AI search.
What it can't promise
- No guarantee of use. Whether an assistant fetches llms.txt is up to the assistant. Anticode can make the file accurate; it can't make anyone read it.
- Not editable. You can't add notes or reorder it. Change the site and the file follows.
- Only what's on the pages. It summarises what you published. It won't add facts you haven't put on the site, which also means it won't invent any.
- Best for design-engine pages. A model-written app keeps most of its content in its own code, so its llms.txt says much less.
Questions people ask
Do I need to write llms.txt myself?
No. Every published site has one at /llms.txt, built from its pages and CMS entries whenever it's requested.
Does it include my prices?
If they're in the site's price list or product cards, yes, as written on the page. Prices that only appear in images or files aren't included.
Will it help me rank in Google?
No. It's aimed at AI assistants and tools, not at search ranking. Your sitemap and the pages themselves matter for search.
Can I see which AI assistants visit?
Your analytics count AI crawlers separately from people, and visits referred by AI assistants as their own source.
Sources
- The /llms.txt file proposal, checked