llms.txt: the complete guide

By UpGeo · Updated July 2026 · 7 min read

llms.txt is a plain-markdown file served at your site root (https://yourdomain.com/llms.txt) that gives AI language models a curated, token-efficient summary of your site: who you are, what you do, and where the pages that matter live. Think of it as a sitemap written for models instead of crawlers.

Why it exists

HTML is a terrible format for language models: navigation, cookie banners and scripts bury the substance. When an AI engine has a limited token budget to understand your site, a clean markdown brief beats making it parse your homepage. The llms.txt proposal (Jeremy Howard, 2024) standardizes that brief, and adoption has spread across documentation sites, SaaS companies and AI-tooling vendors.

Honest caveat: no engine officially guarantees it reads llms.txt. But it costs 10 minutes, several AI crawlers demonstrably fetch it, and it doubles as the single source of truth you'll reuse for structured data and brand messaging. Low cost, real upside — exactly the kind of bet GEO is made of.

The format

llms.txt is markdown with a small amount of structure:

1. An H1 with your site or project name — required.
2. A blockquote one-liner summarizing what you do.
3. Optional free paragraphs of context.
4. H2 sections containing link lists: - [Title](url): note.
5. An optional ## Optional section for secondary links models can skip when tokens are tight.

A complete example

# Acme

> Acme makes self-sealing stem bolts that cut aerospace assembly time by 40%.

Founded in 2020, Acme serves 500+ manufacturers across 30 countries.
Products are certified to AS9100 and ship from three regional warehouses.

## Key pages

- [Product overview](https://acme.com/products): Full catalog with specs
- [Pricing](https://acme.com/pricing): Plans start at $29/mo, volume discounts
- [Documentation](https://acme.com/docs): Integration and installation guides
- [Case studies](https://acme.com/customers): Measured results from real customers

## Optional

- [Careers](https://acme.com/careers)
- [Press kit](https://acme.com/press)

llms.txt vs llms-full.txt

llms.txt is the index; llms-full.txt is the whole book. The full variant concatenates your actual documentation/content into one big markdown file so a model can ingest everything in a single fetch. Publish both if you have real docs: the index for quick orientation, the full file for deep answers about your product. Keep facts in them consistent with your site — contradictions are how hallucinations start.

Deployment checklist

1. Generate the file — by hand or with our free llms.txt generator.
2. Upload to the web root so it's served at /llms.txt with Content-Type: text/plain or text/markdown.
3. Verify it's reachable while logged out, over HTTPS, without redirects.
4. Make sure robots.txt doesn't block AI crawlers from fetching it — see the AI crawler list.
5. Re-check quarterly: stale pricing or dead links in llms.txt are worse than no file at all.

Common mistakes

• Marketing fluff. Models extract facts. "Award-winning synergy platform" gives them nothing; "cuts assembly time 40%, from $29/mo" gives them a quotable claim.
• Dumping every URL. This is curation, not a sitemap. 5–15 links beats 200.
• Letting it rot. An llms.txt that contradicts your site teaches models wrong facts — the one outcome worse than invisibility.