Skip to content
Numinam, home page
AI visibility

llms.txt: should your business website have one?

A short file can guide the tools that read it. Your useful pages and your SEO still carry the weight, and Google Search does not use it as a ranking signal.

llms.txt: should your business website have one?
Published
Updated
Reading time
7 min
Author
SBSébastien BalieuFounder of Numinam

An llms.txt file can give a summary of your site and a set of links to the tools that choose to read it. Publishing one guarantees neither a citation in an assistant nor better rankings. For a small business, it makes most sense when the pages are already clear and keeping the file up to date costs little.

Google states that llms.txt has no influence on its visibility or rankings. So start with the pages that answer your customers’ questions, before you ask your developer for this file.

What is it for?

The llms.txt proposal defines a Markdown document that gives context and points to useful resources. Version 2 allows a file at the root or in a subdirectory, and favours links to content that agents can read easily.

Access control sits elsewhere. robots.txt sets crawl rules, and a private area still needs its own access protection.

Who actually reads this file?

Jeremy Howard published the proposal on 3 September 2024. It remains an open proposal, with no standards body behind it. Its author designed it mainly for tools that read a site at the moment they answer a request, for example a coding assistant to which a developer passes the address of some documentation.

The proposal’s site lists the tools that generate the file: documentation platforms such as Mintlify or GitBook, WordPress plugins such as Yoast SEO and AIOSEO, Wix or Drupal. OpenAI, Anthropic and Google publish an llms.txt for their own developer documentation. Publishing it for their own docs commits their crawlers to nothing when it comes to reading other sites’ files.

On the reading side, the signals are thin. The pages where OpenAI, Anthropic and Perplexity describe their crawlers make no mention of llms.txt. In June 2026, Google Search said these files have no effect, positive or negative, on visibility. Chrome added a check for the file in the experimental Agentic Browsing category of Lighthouse, which assesses how easily agents can read a site.

The file costs you very little if it is generated along with the site. I put it among the finishing touches: useful for the tools that ask for it, with no proven effect on citations.

What format should you follow?

The proposal sets a precise order. Only the title is mandatory:

  1. An H1 heading with the name of the site or the company.
  2. A blockquote (a line starting with >) that sums up the business in one or two sentences.
  3. Free-form paragraphs or lists, with no headings, for useful context.
  4. H2 sections, each containing a list of links.

Each link takes the form - [Name](URL): optional note. An H2 section called “Optional” groups the secondary links, which a tool can skip when it needs to save context.

The proposal says nothing about relative addresses. I use full URLs, so a tool reading the file outside your site knows exactly where to go.

What should go in your file?

Choose the pages that explain your offer, its terms and your proof. Use the company’s exact name and a factual description. Leave out private contact details, internal documents and outdated pages.

An example adapted to Numinam:

# Numinam

> Numinam, a CRO consultant, improves the conversion rate of websites, online shops and apps, and measures every change.

## Services

- [Website CRO](https://www.numinam.com/en/website-conversion-rate-optimisation): more enquiries from the same traffic.
- [Ecommerce CRO](https://www.numinam.com/en/ecommerce-conversion-rate-optimisation): purchase journey and checkout.
- [Free audit](https://www.numinam.com/en/free-cro-audit): conversion blockers, ranked by impact.

## Optional

- [Case studies](https://www.numinam.com/en/case-studies): completed projects.

If your site provides a Markdown version of its pages, you can reference it. Point only to .md addresses that exist. The file must stay readable, and each link must lead to the information it promises.

Version 2 also describes how to signal these Markdown versions: a <link rel="alternate" type="text/markdown"> tag on each page, or an equivalent HTTP Link header. That is a project in its own right, best kept for sites whose pages are generated from content that is already structured.

Write the descriptions like a spec sheet. “Innovative digital solutions” says nothing about the service, the audience or the area. “Conversion audit for Shopify stores in Belgium” covers all three.

How do you get it published?

Your developer can publish a static file or generate a route from the content. The right choice depends on the site. Automatic generation keeps price changes and removed pages from slipping through.

By technology:

  • On an Astro, Next.js or Hugo site, a file in the public folder is enough, or a route that builds it at build time.
  • On WordPress, Yoast SEO and AIOSEO can generate it. Check the result, because the plugin picks the pages for you.
  • On a custom server, a route that returns the file as text/plain or text/markdown.

Have the final address checked, for example /llms.txt, along with its HTTP 200 status and its text content. A file that returns an error page with a 200 status fails delivery.

This happens on single-page applications, which return their HTML page for any unknown address. The browser then shows your home page at /llms.txt, with a 200 status. Open the raw file with curl to be sure.

Also ask who owns the file after a change to the site. An updated services page next to an outdated summary can give two different descriptions of your offer.

Should you publish llms-full.txt?

Not by default. Bundling the whole site adds volume and maintenance work. A tool that needs a specific resource can follow a link to it.

If a user or a known integration asks for a full export, set its scope and how often it is updated. Don’t add thousands of words for the sake of filling a file.

The current page of the proposal does not mention this file. Some documentation generators produce it anyway.

How do you check the delivery?

  • Open the file from the public site.
  • Check that each link goes straight to a live page.
  • Compare the services, contact details and prices with the site.
  • Ask an assistant to answer a question, giving it the file and the linked pages.
  • Check the update after a real change to the content.

The assistant test checks how clear the document is. Whether that assistant would consult it unprompted is a separate question. Server logs can show that the file was actually read; on their own, they prove nothing about its effect on a citation.

Lighthouse checks that the file exists at the root of the domain, in its experimental Agentic Browsing category. That check covers the file itself. Whether the description matches your offer is yours to verify.

The GEO and SEO guide helps you set priorities before you put budget into this.

Frequently asked questions

Does Google use llms.txt?

No. Google Search says these files are unnecessary and have no effect on visibility or ranking. AI Overviews and AI Mode require no special file or markup.

Do you need an llms.txt to be cited by ChatGPT or Perplexity?

No. OpenAI and Perplexity document their search crawlers, OAI-SearchBot and PerplexityBot, without mentioning llms.txt. For search, those crawlers’ access to your pages counts for more.

How is it different from robots.txt?

The robots.txt file gives crawlers access rules, page by page or section by section. llms.txt summarises the site and offers links. The major crawlers read the first; whether the second gets read depends on each tool.

Is the file public?

Yes. Anyone who types the address can read it, your competitors included. Only put in information already published on the site.

Where should the file go?

At the root of the domain, at /llms.txt. Version 2 also allows a file in a subdirectory, for example /docs/llms.txt, which then covers the pages in that directory.