What Is llms.txt?

llms.txt is a proposed standard for a Markdown file, placed at /llms.txt on a website, that gives AI tools a short summary of the site and links to its most useful pages. For a health practice, it would work like a one-page map written for machines. It was proposed by Jeremy Howard and published on September 3, 2024, and it remains a proposal and not an official web standard.

How llms.txt works

An AI tool that wants to understand a website has a problem. Web pages are full of menus, scripts and layout code, and an AI model has limited room to read. The llms.txt proposal tries to fix that with a small curated file written in Markdown, a simple text format with headings and lists.

According to the proposal, the file has a set structure.

• A title with the name of the site. It is the only required part.
• A short summary in a quote block.
• Optional notes in plain paragraphs or lists.
• Sections of links, each with a short note about what the page contains.
• An optional section for secondary links that a tool can skip when it is short on room.

An AI agent would read the file, then follow the links to the pages it needs. Think of it as the sign-in desk of a building, with a short list of where to go. The proposal says the file can sit at the root of a site or at a subpath, and that it is meant to complement sitemap.xml and robots.txt, not replace them. It also suggests publishing clean Markdown versions of pages, which means plain-text copies of your pages at a matching address. For most practice sites that is a developer task and not an owner task.

Now the part that matters for your decision, which is who uses it. The proposal's site says thousands of sites publish an llms.txt file, and that documentation platforms generate one automatically. It says OpenAI, Anthropic and Google Gemini publish llms.txt files for their developer documentation. It describes software documentation as the place the file is used most, because coding tools use it to find API references. It also says the structure fits businesses and personal sites.

What it does not say is whether the AI products that send patients your way read these files on other sites. Google is more direct about its own side. Its AI features page says there are no additional requirements to appear in AI Overviews or AI Mode, that you do not need to create new machine readable files, AI text files or markup, and that a page must be indexed and eligible to show with a snippet. I did not find a statement from OpenAI, Google or Anthropic that their search products rely on the file.

So the honest status is this. The idea is reasonable, adoption is real in documentation, and the benefit for a medical practice's website is unconfirmed. The things AI tools clearly need are access and clear text. For access, see AI crawler.

For text, see extractability, which covers whether a machine can pull a clean answer from your page.

Example: llms.txt at a health practice

Picture a functional medicine clinic with three providers. The owner hears at a conference that "you need an llms.txt file to show up in ChatGPT." This example is illustrative, not a transcript of a real site.

Before touching the file, the owner runs a short audit and finds three things.

• The robots.txt file blocks every bot with "User-agent: *" followed by "Disallow: /", left over from a staging site.
• The price of the signature program appears only inside an image, so no text version exists.
• The provider bios are in a slideshow that does not load without scripts.

The owner's first instinct was to write an llms.txt file. That would have been the least valuable of the four jobs. OpenAI's bots page says sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers. A site that blocks every bot has opted out of that and everything else, and a map pointing to pages that bots are told to skip is useless.

So the clinic fixes things in order. First it removes the blanket block. Second it publishes the program price as text. Third it turns the bios into plain pages.

Only then does it write the file. It takes about an hour.

# Example Functional Medicine

> Functional medicine clinic in Denver, Colorado, with three providers.
> We test and treat thyroid, gut and hormone conditions.

## Key pages
- [Thyroid program](https://example.com/thyroid): what is included, price and who it fits
- [Our providers](https://example.com/providers): training and focus of each provider
- [Contact and hours](https://example.com/contact): address, phone and hours

The practice name and the links in this sample are placeholders. Replace them with your own details, and keep the notes short and factual.

The clinic then logs the date it created the file and sets a reminder to update it when pages change. It does not promise itself a jump in ChatGPT mentions, because nothing in the sources says one is coming. If one appears, good. If not, the clinic's pages are cleaner and its robots.txt is correct, and those are the changes with real support behind them.

How to check yours in 60 seconds

1. Type yourpractice.com/llms.txt into a browser.
2. If you see a page with your name and a list of links, you have one.
3. If you see an error page, you do not.
4. Note the date you checked, and whether anyone on your team or a plugin created the file.
5. While you are there, open yourpractice.com/robots.txt and look for a "Disallow: /" under "User-agent: *". That is the more important problem.

What to do about it

1. Fix the basics first. Check that robots.txt does not block search bots, hours and addresses are correct, and key facts are text and not images. See AI bot blocking.
2. If you want the file anyway, start with a heading that is your practice name.
3. Add a one or two sentence summary of what you do and where.
4. List your ten most important pages with a short note on each.
5. Use plain language and facts a patient would ask about, such as services, providers, location and what is included in a program.
6. Keep it current, and update it whenever a listed page changes or moves.
7. Treat it as optional housekeeping and do not expect a measurable jump from it.
8. Put the file's last-updated date in your site notes and recheck it when you add or retire a service, so the map does not point to a page that is gone.