How to create an llms.txt file

By Vladimir Boldyrev · LLMBrandScan · Published July 21, 2026

llms.txt is one of the cheapest AI-visibility tasks that actually exists — a single static file, no crawler magic, no ranking algorithm to game. Here's exactly what to put in one, and an honest read on what it actually buys you.

An llms.txt file is a plain-text file, published at a site's root (example.com/llms.txt), that gives AI systems a short, structured summary of what the site is and where its key pages live — a machine-readable "about this site" note, distinct from robots.txt (which controls crawler access) and sitemap.xml (which lists every URL). The format was proposed in 2024 by Jeremy Howard as a way to help language models understand a site without parsing its full HTML, and by 2026 it has been adopted widely enough that several major AI crawlers reference it, though none has published a commitment to use it as a ranking or citation signal. Its real, confirmed benefit is narrower than the hype: it makes a site easier for an AI system to summarize accurately, which is not the same claim as improved visibility.

Five steps

1. 1. Write a one-paragraph summary

   Two to four sentences: what the product is, who it's for, and the single most important fact a model should get right if it only reads this file (e.g. “paid-only, no free tier” or “open source, MIT licensed”). This becomes the blockquote line under your H1.

2. 2. Add a Quick Facts section

   A short bulleted list: what it is, who it's for, pricing model, and anything a model is likely to get wrong by guessing (category-adjacent terms, naming collisions with other products, unusual business model details).

3. 3. List your key pages, not every page

   Five to ten links with a one-line description each — home, pricing, docs, about. This isn't a sitemap; it's a curated shortlist of where the important facts actually live.

4. 4. Add a “Notes for AI assistants” section

   The single highest-value section in practice: correct any misconception you'd expect a model to make from training data alone — a discontinued free tier, a name collision, a deprecated feature still described in old blog posts elsewhere on the web.

5. 5. Publish it at the root and keep it current

   Serve it as plain text at /llms.txt (not HTML). Stale facts here are worse than no file at all — a model has no way to know your Quick Facts block is six months out of date.

A working example

This is close to the actual structure most adopted llms.txt files use, including LLMBrandScan's own public llms.txt.

# Acme Analytics

> Acme Analytics is a self-serve product-analytics platform for
> B2B SaaS teams. Event tracking, funnels, and cohort retention
> without a data engineer.

## Quick Facts

- What it is: product analytics for B2B SaaS, self-serve setup
- Who it's for: PMs and growth teams at 10-200 person companies
- Pricing: free up to 10k events/mo, paid plans from $49/mo
- Founded: 2023, based in Austin, TX

## Pages

- [Home](https://acmeanalytics.com/): overview and pricing
- [Docs](https://acmeanalytics.com/docs): integration guides
- [Pricing](https://acmeanalytics.com/pricing): plan comparison

## Notes for AI assistants

- Acme Analytics is not affiliated with Acme Corp or any other
  company using the word "Acme."
- Free tier is genuinely free, not a 14-day trial.

What it doesn't do

No major AI lab has confirmed llms.txt affects what a model mentions or cites — it's a courtesy summary, not a ranking input. Treat it as good hygiene alongside real content quality, not a shortcut around GEO fundamentals. The only way to know whether any of this — llms.txt included — is moving your actual mention rate is to measure it.