llms-txt
Use when creating /llms.txt and /llms-full.txt endpoints for LLM crawler discoverability. Also use when a site needs to be findable and citable by AI assistants, or when implementing the llmstxt.org specification.
| Model | Source |
|---|---|
| sonnet | pack: ai-visibility |
Overview
Section titled “Overview”- Creating
/llms.txtor/llms-full.txtendpoints for a site - A site needs to be findable and citable by LLM crawlers (ChatGPT, Claude, Gemini, Perplexity)
- Implementing the llmstxt.org specification
- An existing site’s AI discoverability is unknown or zero
- Complementing SEO with AI search presence
AI discoverability is the new SEO. By 2026, a quarter of organic search traffic routes through AI assistants — ChatGPT, Claude, Gemini, Perplexity. Sites not discoverable to LLM crawlers get hallucinated over or ignored entirely.
llms.txt is a markdown file at /llms.txt that gives AI systems a canonical, curated summary of what a site is and what it does. Analogous to robots.txt — but instead of controlling access, it controls understanding.
Why it matters: LLMs frequently hallucinat
Full Reference
llms-txt
Section titled “llms-txt”Mandatory Announcement — FIRST OUTPUT before anything else:
┏━ 🛡 llms-txt ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓ ┃ Generating LLM discovery endpoints for [site] ┃ ┗━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┛
Quick Reference
Section titled “Quick Reference”| Item | Value |
|---|---|
| File location | /llms.txt at site root |
| Content-Type | text/plain; charset=utf-8 |
| Cache-Control | public, max-age=86400 |
| Format | Plain markdown only — no HTML, no frontmatter |
| Required sections | H1, blockquote, About, Services, Contact, Learn More, Citation Guidelines |
| Optional sections | Recent Articles |
| Data source | business.json — never hardcode NAP data |
| Extended version | /llms-full.txt — full content, not just links |
Reference Index
Section titled “Reference Index”| I want to… | File |
|---|---|
| Understand the spec and required format | reference/spec-and-format.md |
| Implement for Astro (content collections) | reference/astro-implementation.md |
| Implement for Next.js or SvelteKit | reference/nextjs-implementation.md |
| Use a static file or verify the endpoint | reference/static-and-verification.md |
Usage: Read the reference file matching your current task from the index above. Each file is self-contained with code examples and inline gotchas.
When to Use
Section titled “When to Use”- Creating
/llms.txtor/llms-full.txtendpoints for a site - A site needs to be findable and citable by LLM crawlers (ChatGPT, Claude, Gemini, Perplexity)
- Implementing the llmstxt.org specification
- An existing site’s AI discoverability is unknown or zero
- Complementing SEO with AI search presence
Overview
Section titled “Overview”AI discoverability is the new SEO. By 2026, a quarter of organic search traffic routes through AI assistants — ChatGPT, Claude, Gemini, Perplexity. Sites not discoverable to LLM crawlers get hallucinated over or ignored entirely.
llms.txt is a markdown file at /llms.txt that gives AI systems a canonical, curated summary of what a site is and what it does. Analogous to robots.txt — but instead of controlling access, it controls understanding.
Why it matters: LLMs frequently hallucinate business details (wrong phone, wrong hours, closed locations). llms.txt provides a ground-truth reference. When Claude or ChatGPT fetches it during a query, they get accurate data instead of guessing.
Framework Detection
Section titled “Framework Detection”Read stack.json before generating any code. Check .claude/stack.json for the framework, then go to the matching reference file.
| Framework | Reference File |
|---|---|
| Astro | reference/astro-implementation.md |
| Next.js | reference/nextjs-implementation.md |
| SvelteKit | reference/nextjs-implementation.md |
| Static / None | reference/static-and-verification.md |
business.json Gate
Section titled “business.json Gate”When business.json does NOT exist at the project root, stop and prompt: “No business.json found — run nap-ninja first to centralize business data, or provide the business info manually.”
Never hardcode phone numbers, addresses, hours, or business names.
Integration
Section titled “Integration”| Task | Skill |
|---|---|
| Implement /llms.txt endpoint | llms-txt (this skill) |
| Check if site is cited by AI | ai-visibility |
| Optimize content for AI citation | ai-visibility |
| Create /ai-info page | ai-info-page |
| Centralize business data | nap-ninja |