How to create an llms.txt — the honest guide.

An llms.txt is a plain text file in the root of your website (yourdomain.com/llms.txt) that tells AI systems, in curated form, what your company does, for whom — and where the most important content lives. It takes about an hour to write, costs nothing, and is the fastest single step a website can take towards GEO.

This post is the complete guide: format, content, common mistakes — and the part most guides leave out: what the file does not do.

What an llms.txt is — and where it comes from

The proposal comes from Jeremy Howard (Answer.AI), published in September 2024 and documented as an open standard at llmstxt.org. The idea in one sentence: robots.txt tells crawlers where they may go — llms.txt tells AI systems what is worth reading.

The background is a technical problem. Language models that look things up live (Perplexity, ChatGPT with search, Copilot) have a limited reading budget per query. A modern website consists of navigation, cookie banners, scripts and markup — the actual content is tedious for a machine to extract. An llms.txt delivers the essence pre-filtered: pure Markdown, no menus, no pop-ups, no packaging.

Why now — and why with sober expectations

Two things are true at the same time, and honest advice names both:

1. The channel is growing. A growing share of buying research starts in AI assistants instead of on a results page — we covered this in detail in our post on AI access. Answer engines cite sources they can read quickly and reliably. Machine readability is the entry condition.
2. The file is not a Google ranking lever. Google has stated clearly that AI Overviews require no special files or markup — what counts there is the same quality work as for classic search. An llms.txt addresses the other engines: Perplexity, ChatGPT, Claude, Copilot. Anyone selling you the file as an “AI SEO guarantee” is selling a promise nobody can keep.

Our assessment: effort one hour, risk zero, benefit growing. That is the kind of bet you take.

The guide: step by step

Step 1 — Set up the structure

The format is deliberately simple: one Markdown file with a fixed basic structure.

# Company name

> One paragraph describing the company in 2–4 sentences:
> what you do, for whom, where. This is the text an AI will
> use to describe you — write it accordingly.

## Core services

- [Service 1](https://yourdomain.com/service-1): One sentence on what it delivers.
- [Service 2](https://yourdomain.com/service-2): One sentence on what it delivers.

## Who we work for

Audiences and industries — concrete, not generic.

## Contact

- Email: hello@yourdomain.com
- Contact page: https://yourdomain.com/contact
- Location: City, Country

Three format rules that actually matter: one H1 (the company name), directly below it a blockquote as the short description, then H2 sections with link lists. Every link gets a one-sentence description — the AI should know what to expect behind a link before following it.

Step 2 — Select content: curate, don’t copy

The file is a table of contents, not a sitemap. What belongs in it is what a stranger would need to describe and recommend you correctly:

• What you do — services with one precise sentence each, with numbers where you have them.
• For whom — industries, company sizes, region. “DACH Mittelstand, dentists and law firms” beats “clients of every size”.
• How to reach you — email, contact page, location.
• Your most important pages — services, pricing, about. If pricing is public: link it. AI agents comparing providers skip what they cannot read.

What does not belong: every single URL, marketing filler (“innovative”, “holistic”), and anything that goes stale faster than you will maintain the file.

Step 3 — Take the file live

Place llms.txt in the root so it is reachable at yourdomain.com/llms.txt with status 200 as plain text. On most systems that simply means: put the file in the public folder (public/, static/, webroot). No plugin required.

Step 4 — Open the door that goes with it

An llms.txt is useless if the robots.txt next to it locks the AI crawlers out. Check whether GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot and Google-Extended are allowed to read your website — many security plugins and CDN defaults block everything that isn’t called Googlebot. Our AI access check verifies both in ten seconds: the door and the table of contents, live, no signup.

Step 5 — Maintain it like your price list

A stale llms.txt is worse than none: it actively feeds AI systems wrong information about your company. Our rule: the file gets updated with every change to services, pricing or contact details — which is why ours lives in the code repository next to the website, not in a forgotten FTP folder.

The next level: Markdown versions of entire pages

The standard provides for a second tier: besides the table of contents, every important page can get its own Markdown version at the same address plus .md. That is exactly how we do it ourselves — seo216.com/en/services.md is the machine-readable version of our services page, and every post on this blog also exists as an .md twin, linked in the <head> of each page. For an answer engine, that is the difference between “photographing through a shop window” and “being handed the catalogue”.

This is part of our GEO & SXO pillar: access, structure and citability as one system — not as a collection of files.

The four most common mistakes

1. HTML instead of Markdown. The file must be plain text. Creating it as a CMS page often ships an HTML page under a false name — our check detects exactly this case.
2. Copying the sitemap. 400 uncommented URLs are not context, they are noise. Curating means leaving things out.
3. Write and forget. Outdated services, dead links, old prices — the file then describes you incorrectly, with a stamp on it.
4. Maintaining llms.txt, ignoring robots.txt. The table of contents is on display, but the door is locked. The two get checked together.

What the file can’t do — and what comes after

An llms.txt makes you readable. It does not automatically make you citable. Whether ChatGPT actually recommends you is decided by your content: clear answers in the right place, sourced numbers, named entities, clean schema markup — the work we call LLM citation and measure monthly per engine.

The honest roadmap in three steps: first check for free whether your website is open to AI. Then create your llms.txt with this guide — one hour, once. And if you want to know whether and where you are already cited today, and who gets recommended instead: the full audit answers exactly that, within 24 hours, free.