What is the llms.txt file?

llms.txt is a plain-text Markdown file placed at the root of a site (for example, raven3.com.ar/llms.txt) that acts as a curated map so language models — ChatGPT, Claude, Gemini, Perplexity — can quickly understand what your site is, what it offers, and where to find the most relevant information, without having to crawl and parse the entire HTML.

The proposal emerged in 2024 and takes inspiration from two files you already know: robots.txt (tells crawlers what they're allowed to crawl) and sitemap.xml (lists every URL). llms.txt does something different: instead of listing everything, it selects what matters and summarizes it in natural language.

Why it matters to an LLM (and to your business)

Language models work with a limited context window. When someone asks ChatGPT or Perplexity about your industry, the model can't read your entire site — it needs concise, trustworthy fragments. HTML loaded with menus, scripts, banners, and UI components is noise to an LLM. A clean Markdown file with the essence of your business is a direct signal.

On top of that, many corporate sites are applications that render content with JavaScript. If the bot doesn't execute JS — and several don't — it sees an almost empty page. llms.txt solves that by delivering the key content in plain text upfront, with no dependency on rendering.

The standard structure

The format is simple and intentionally minimal:

  • An H1 with the brand or site name
  • A one-line blockquote summarizing what the company does
  • An H2 for each topical section (for example 'Main pages,' 'Policies,' 'Documentation') with a list of links, each with a short description

For very large sites or ones with extensive documentation, there's also an llms-full.txt variant, which includes the complete content instead of just links and summaries.

How we implemented it at Raven3

At raven3.com.ar/llms.txt we follow that exact structure: an H1 with the agency name, a one-line blockquote summarizing our value proposition, and two sections — Main Pages and Policies — with direct links and a one-sentence description for each. No filler, no keyword stuffing: just the information a model needs to understand what the site is about and recommend it accurately.

Common mistakes when building yours

Treating it like a traditional SEO file and stuffing it with keywords: LLMs don't rank by keyword density, they prioritize clarity and precision. Copying the entire sitemap: if you have 200 URLs, the file loses its value as a curated map. Letting it go stale: if you add or remove pages and don't update it, the model can end up citing outdated information. Assuming it replaces technical SEO: llms.txt is a complement, not a substitute for semantic HTML, schema markup, and quality content.

Checklist for building your llms.txt

  • Create the file as plain text and upload it to your domain root (yourdomain.com/llms.txt)
  • Start with an H1 with your brand name and a one-line blockquote summarizing what you do
  • Group your most important pages into 2 or 3 sensible sections (services, about the company, contact)
  • Describe each link in one clear sentence, without internal jargon
  • Verify that your robots.txt isn't blocking the AI crawlers you need reading the file
  • Review it every time your service offering or page structure changes