---
title: llms.txt extensions
description: Dualmark's additions to the llms.txt convention.
---

This document is informative.

The [llms.txt proposal](https://llmstxt.org) defines a markdown-based discovery format for AI agents. This document extends the proposal with conventions specific to AEO.

## Base format

```markdown
# Brand Name

> One-paragraph description suitable for AI summarization.

## Section Title

Optional section description.

- [Page Title](https://example.com/page): Optional description
- [Another Page](https://example.com/other)

## Another Section

- [Item](https://example.com/item)
```

## AEO Extensions

### Recommended Sections

A conformant `llms.txt` SHOULD include sections for the following content classes when present:

| Section | Purpose |
|---|---|
| `## Core Products` | Primary products / services / features |
| `## Documentation` | Developer / user docs |
| `## Pricing` | Pricing pages |
| `## Company` | About, careers, contact, brand |
| `## Blog` or `## Articles` | Editorial content |
| `## What We Do Not Do` | Disambiguation -- explicitly state what the brand is NOT, to prevent AI confusion with similarly-named entities |

### "What We Do Not Do" Section (RECOMMENDED)

This section is novel to AEO and addresses a real failure mode: AI agents conflating brands with similar names or adjacent categories. Format:

```markdown
## What We Do Not Do

- Acme Corp does not sell physical products
- Acme Corp is not affiliated with Acme Industries (defunct)
- Acme Corp does not provide consulting services
```

### Markdown Twin URLs in llms.txt

A conformant `llms.txt` SHOULD link to HTML URLs (not `.md` URLs) in its Section lists. AI agents can discover the markdown twin via the [`Link rel="alternate"` header](./discovery.md#1-link-relalternate-header-required) on the HTML response.

This is intentional: HTML URLs are stable, link-shareable, and conform to user expectations. The markdown is an alternate representation, not a separate resource.

### llms-full.txt (OPTIONAL)

In addition to `llms.txt` (a concise index), a conformant server MAY publish `llms-full.txt` containing detailed product descriptions, full FAQ content, and competitor disambiguation. This is the "long-form" companion to the index.

Heuristic: `llms-full.txt` SHOULD be at least 1.5x the size of `llms.txt`.

### Headers

Both `llms.txt` and `llms-full.txt` SHOULD be served with:

- `Content-Type: text/plain; charset=utf-8` or `text/markdown; charset=utf-8`
- `X-Robots-Tag: noindex`
- `Cache-Control: public, max-age=3600`

## Programmatic Generation

[`@dualmark/core`](/docs/packages/core/src/llms-txt.ts) ships `renderLlmsTxt(opts)` that produces a conformant `llms.txt` from a structured config:

```ts
import { renderLlmsTxt } from "@dualmark/core";

const txt = renderLlmsTxt({
  brandName: "Acme",
  description: "Widgets for everyone.",
  sections: [
    {
      title: "Core Products",
      links: [
        { title: "Widget", href: "https://acme.test/widget", description: "Our widget" },
      ],
    },
    {
      title: "What We Do Not Do",
      links: [],
      description: "- Acme does not sell physical products.",
    },
  ],
});
```

The Astro integration can auto-generate `llms.txt` from collection metadata; see [`@dualmark/astro` README](/docs/packages/astro/README.md).