llms.txt: what it is and how to implement it

llms.txt is a text file that lives at the root of your site and offers language models a condensed, navigable version of what you do. The idea is simple: if robots.txt tells crawlers what they may visit and sitemap.xml tells them what exists, llms.txt tells an LLM what matters and where to read it in a clean format.

We have one live at landaverdelabs.com/llms.txt. In this article we cover what it is exactly, how it's structured, how to set it up on your site in under an hour, and the part almost nobody tells you: how much you can actually expect from it today.

Where it comes from

Jeremy Howard published the proposal in September 2024, and the full specification lives at llmstxt.org. The problem it attacks is real: LLM context windows are limited, and modern web HTML is full of navigation, scripts, and noise that gets in the way when a model tries to understand a site. A concise markdown file, curated by the site owner, solves that.

The specification proposes two things:

1. A /llms.txt file at the domain root, in markdown, with the site's essential information and links to expanded content.
2. Markdown versions of your important pages, available at the same URL with .md appended.

The format, piece by piece

According to the spec, the file follows a fixed markdown structure:

• An H1 with the site or project name. The only required section.
• A blockquote with the summary. Two or three lines condensing what the site is and what it offers.
• Optional context paragraphs. Whatever a model needs to interpret the rest correctly.
• H2 sections with link lists. Each link formatted as [name](url) with an optional note explaining what's there.
• An "Optional" section at the end. Secondary links a model can skip when context runs short.

In ours, for example, we describe what we do, who we work with, what we don't do, and how to reach us. That last part, the "what we do NOT do", is among the most valuable: it keeps a model from recommending you for something you don't offer.

How to implement it

No plugins, no tools. It's a plain text file:

1. Write it by hand. Open an editor and follow the structure above. Think about what an AI assistant should know when it describes your business to a potential customer. Be concrete: services, location, who you serve, how to contact you.
2. Upload it to the root. In Next.js or Astro it goes in public/llms.txt. In WordPress, upload it via FTP to the root folder or use a static-files plugin. It must respond at yourdomain.com/llms.txt.
3. Verify it serves as plain text. Open it in the browser: you should see raw markdown, no HTML template around it.
4. Update it when your offer changes. A stale llms.txt misinforms with authority. Review it whenever your services or pricing change.

The whole exercise takes under an hour and has no side effects: search engines ignore it without penalty.

What llms.txt does not do (and this needs saying)

Here's the honest part missing from most articles on the subject: llms.txt is a proposed standard, not an adopted one. As far as we know at the time of writing, no major AI provider has officially confirmed that its models consume it systematically. Implementing it does not guarantee you'll appear in ChatGPT or Perplexity.

So why do it? Three reasons:

• Near-zero cost, zero risk. One hour of work for the chance that the agents which do read it understand you better.
• AI agents are growing. More and more traffic comes from assistants browsing on someone's behalf. A curated, readable summary is the best face you can show them.
• The exercise itself is worth it. Writing your llms.txt forces you to condense what you do, for whom, and why. If you can't write it in 40 lines, your website probably isn't communicating it either.

What moves results today is what sits underneath: citable content, correct schema, authority. llms.txt is the bow, not the gift. If your site doesn't show up in AI engines, the cause is almost never this missing file: we walk through the full diagnosis in why your site doesn't appear in ChatGPT.

Our recommendation

Implement it, spend the hour, then invest the rest of your energy in the fundamentals. In our GEO service, the llms.txt is a standard deliverable, but it's one of twenty: it ships alongside citable structure, complete schema, and monthly measurement of your AI visibility. That's the right order of priorities.