llms.txt

Overview

AEO (Answer Engine Optimization) covers how your site is understood and cited by AI assistants (ChatGPT, Perplexity, Claude, etc.). llms.txt is the primary convention: a Markdown file at /llms.txt that gives models a curated map of your most important pages.

It complements (does not replace) robots.txt and sitemap.xml. It is not a W3C/IETF standard; treat it as durable, machine-readable site documentation you control.

Reference: llmstxt.org · GitHub spec repo

When to use llms.txt

Not a substitute for: SEO (use sitemap + structured data), crawl policy (use robots.txt), or full-text search indexing.

File location and delivery

• Canonical path: https://example.com/llms.txt (site root).
• Optional: subpath (e.g. /docs/llms.txt) or /.well-known/llms.txt (root is preferred).
• Content-Type: text/plain; charset=utf-8 (Markdown body; plain text MIME is conventional).
• Multilingual: one file per language root (e.g. /en/llms.txt, /es/llms.txt).

Static file

Place public/llms.txt in Next.js/Vite/static hosts. Served as-is at /llms.txt.

Dynamic route (Next.js App Router)

When content comes from a CMS or database, use a route handler:

// app/llms.txt/route.ts
export async function GET() {
  const content = buildLlmsTxt(); // assemble Markdown string
  return new Response(content, {
    headers: {
      'Content-Type': 'text/plain; charset=utf-8',
      'Cache-Control': 'public, s-maxage=3600, stale-while-revalidate=86400',
    },
  });
}

Build the Markdown string in code; keep sections in spec order. Fetch dynamic slugs server-side (blog posts, docs) and append link lines.

Spec format (required order)

Per llmstxt.org, sections appear in this order:

1. H1 — project or site name (required; only mandatory section).
2. Blockquote — one short summary (> ...) with key context for interpreting the rest.
3. Body (optional) — paragraphs or lists with extra guidance; no headings in this block.
4. H2 sections — each section is a “file list” of markdown links.
5. ## Optional (special) — secondary links agents may skip when context is limited.

Link line syntax

Each entry is a markdown list item:

- [Link title](https://absolute-url): Optional one-line description.

• [name](url) is required.
• : description is optional but strongly recommended.
• Use absolute URLs.
• Prefer links to .md versions of pages when available (same path + .md, or index.html.md for extensionless URLs).

Minimal example

# Acme Docs

> Acme is a task API for teams. REST + webhooks; auth via API keys.

Use the Quick Start first, then the API reference for endpoints.

## Docs

- [Quick start](https://docs.acme.com/quickstart.md): Install SDK and send your first request
- [API reference](https://docs.acme.com/api.md): All endpoints and error codes

## Examples

- [Node webhook handler](https://github.com/acme/examples/blob/main/webhook.ts): Verify signatures and process events

## Optional

- [Changelog](https://docs.acme.com/changelog.md): Release history

Companion conventions

Markdown mirrors (.md URLs)

Pages useful to LLMs should expose a clean Markdown copy at the same URL + .md (or index.html.md). Example: https://docs.fastht.ml/docs/tutorials/quickstartforweb_devs.html.md.

llms-full.txt / llms-ctx

• llms-full.txt — expanded file with page content inlined (single fetch for agents).
• llms-ctx.txt / llms-ctx-full.txt — generated via llms_txt2ctx CLI from llms.txt (XML-ish bundle for tools like Claude).

Use when documentation is large; most marketing sites only need llms.txt.

vs robots.txt and sitemap.xml

llms.txt is aimed at inference (on-demand agent context), not training policy or full-site enumeration.

Authoring guidelines

• Concise blockquote and descriptions; avoid marketing fluff agents cannot act on.
• Curate — link to 10–30 high-value pages, not every URL in the sitemap.
• Stable URLs — prefer canonical paths; update llms.txt when slugs change.
• No ambiguous jargon without a one-line explanation.
• Test — feed the file (or llms-ctx output) to an agent and verify it answers common questions.
• Put secondary links (changelog, legal, old posts) under ## Optional.

What to include (typical sections)

Adapt section names to the site; only ## Optional has special semantics in the spec.

Framework integrations

Common generators (see llmstxt.org integrations):

• VitePress: vitepress-plugin-llms
• Docusaurus: docusaurus-plugin-llms
• Drupal: LLM Support recipe
• CLI: llms_txt2ctx (Python) — parse and expand to context files
• PHP: llms-txt-php — read/write spec-compliant files

For custom stacks (Next.js + Sanity, etc.), a route.ts handler that builds Markdown from queries is the usual pattern.

Common mistakes

• Missing H1 — only required element; file is invalid without it.
• Wrong section order — blockquote before H2s; body before first H2.
• Relative URLs — use absolute https:// links.
• Dumping sitemap — too many low-value links defeats the purpose.
• Using llms.txt for robots rules — use robots.txt instead.
• Broken link format — must be - [Title](URL): description; parsers expect this shape.
• Skipping descriptions — weak for agents choosing which link to fetch next.

Related skills

• sitemap — exhaustive URL list for search crawlers (different purpose than llms.txt).
• json-ld — schema.org structured data for Google rich results.

Additional resources

• [reference.md](reference.md) — Official spec, tools, directories, examples, comparisons.
• Spec: https://llmstxt.org/
• GitHub: https://github.com/AnswerDotAI/llms-txt