ScoreGeo

Blog · 9 min read

How to write a /llms.txt that actually works

Published on 18. Mai 2026 · Updated on 25. Mai 2026

A /llms.txt is a Markdown file placed at the root of your site that guides LLMs to your priority content. Proposed by Jeremy Howard in September 2024, it has become the de-facto standard for AI-friendly websites. This guide explains exactly what to put in it, what to avoid, and how to validate it.

What is /llms.txt, exactly?

/llms.txt lives at the root of your domain (https://yoursite.com/llms.txt). It's plain Markdown, served with Content-Type: text/plain or text/markdown. Its job: tell LLMs (ChatGPT, Claude, Perplexity, Gemini) which pages of your site they should read first to understand who you are and what you do.

Think of it as a robots.txt for context, not for crawl rules. robots.txt tells bots what they may crawl ; llms.txt tells them what to prioritize for understanding.

Why it matters in 2026

Three reasons make it non-negotiable for GEO-conscious sites:

1. **Major LLMs now read it.** Anthropic, Perplexity, You.com and others have confirmed they fetch /llms.txt during answer generation. ChatGPT-User retrieves it in browsing mode.

2. **It compresses your site for the model context window.** Instead of crawling 200 pages, the LLM reads a 1-page summary that points to your 5-10 most important URLs.

3. **It's a positive ranking signal** in the ScoreGeo grid (5 points). Many sites still don't have one — early adopters get free differentiation.

The minimum viable /llms.txt

Here's the structure that scores 5/5 in our analyzer:

`` # Your Brand Name > Your Brand is [what you do] for [who]. [Key value prop in one sentence]. ## Main pages - [Homepage](https://yoursite.com/): main landing - [Product/Service](https://yoursite.com/product): detailed description - [Pricing](https://yoursite.com/pricing): plans and prices ## Resources - [Documentation](https://yoursite.com/docs) - [Blog](https://yoursite.com/blog): in-depth articles - [Case studies](https://yoursite.com/customers) ## Optional - [Legal](https://yoursite.com/legal) - [Privacy](https://yoursite.com/privacy) ``

Three rules :

- **One H1**, your brand name, exactly as in your Organization JSON-LD.

- **One blockquote** (>) right under, your tagline in 1-2 sentences. This is the literal text the LLM may quote.

- **Sections with ##** then list items - [Title](URL): short description. Always absolute URLs.

What to put in (and not)

DO include

Your home, your core product/service pages, your pricing page, your documentation, your top 5-10 cornerstone blog articles, your case studies if you have them. Pages that answer real customer questions, not internal ops pages.

DON'T include

Login pages, account dashboards, anything behind auth, internal tools, old promotional pages, thin content. Also avoid listing every single article — keep it curated. A llms.txt with 200 links is noise.

Common mistakes

**1. Relative URLs.** [Pricing](/pricing) fails because the LLM may not know your domain. Always absolute: [Pricing](https://acme.com/pricing).

**2. No blockquote intro.** The > line is the literal sentence the LLM quotes back. If absent, it falls back to your meta description (less controlled).

**3. Listing only top-level URLs.** A llms.txt with just home + about is too thin. Aim for 8-15 links covering the breadth of your offer.

**4. Forgetting to update it.** When you ship a new cornerstone article, add it to /llms.txt. Treat it as living documentation, not set-and-forget.

**5. Wrong content-type.** Some servers serve .txt as application/octet-stream which makes browsers download. Configure your host to serve Content-Type: text/markdown; charset=utf-8 or text/plain; charset=utf-8.

Validate it

Three quick checks once published:

- curl -I https://yoursite.com/llms.txt → should return 200 OK with text/markdown or text/plain.

- Visit the URL in your browser → should display the Markdown source.

- Run a ScoreGeo audit on your home → the llms.txt criterion should pass at 5/5.

If you want bonus credibility, also publish /llms-full.txt which concatenates the actual content (not just links) of your cornerstone pages. Some LLMs prefer the full text version when the link version is too sparse.

Real example

ScoreGeo's own /llms.txt is public at https://scoregeo.ai/llms.txt — it follows exactly the structure above and scores 5/5 in our own audit. Copy the pattern, adapt the URLs, ship in 10 minutes.

Frequently asked questions

Is llms.txt an official standard?

It's a community proposal by Jeremy Howard (fast.ai, llmstxt.org) from September 2024, not an IETF RFC. But it has been adopted by Anthropic, Perplexity, You.com, Cloudflare and many SaaS — adoption velocity makes it the de-facto standard.

Should I put marketing copy in the blockquote?

No. Use factual descriptive language. The blockquote is what the LLM may literally quote. "Acme is a Paris-based invoice management SaaS for French SMEs with Cegid integration" works much better than "Acme is the leading revolutionary platform for next-gen finance teams."

How is /llms.txt different from robots.txt or sitemap.xml?

robots.txt controls crawl access (what bots can/can't fetch). sitemap.xml is an exhaustive list of all crawlable URLs with priority/changefreq metadata. llms.txt is curated and semantic — a few dozen carefully chosen links with descriptions, optimized for LLM context windows.

Will my llms.txt help Google AI Overviews?

Indirectly. Google has not officially committed to reading /llms.txt, but Bard/Gemini consume llms.txt-style hints when scraping Bing search results. Better safe than sorry — for a 10-minute investment, the asymmetric upside is worth it.