llms.txt explained: what the file does for AI assistants and what it does not

What is an llms.txt?

An llms.txt is a plain text file in Markdown format that can be reached at your-domain.com/llms.txt. It is not written for people but for large language models (LLMs) and for the tools AI assistants use to read web pages. The idea: instead of picking apart a website full of navigation, cookie banners, scripts and ads, a language model finds a tidy summary in a fixed place.

The proposal dates from 2024. It is deliberately simple: Markdown is easy to read for both people and machines, needs no special software and can be maintained in any text editor. The file complements existing files such as robots.txt and sitemap.xml, but does not replace them. robots.txt controls who may crawl what, the sitemap lists every address, and llms.txt explains in a few lines what the website is about and which pages really matter.

For a local business, this is very practical. When someone asks an AI assistant for a dentist, a tax adviser or a café in a particular district, the answer is assembled from sources across the web. The more clearly your own website states who you are, what you offer and where you are located, the better the chance that these details are picked up correctly. llms.txt is one building block for that, but not the most important one.

How is an llms.txt structured?

The proposal describes a fixed order. Only the first line is required; everything else is recommended. What matters is that the file stays short and points to real, reachable pages. An llms.txt is not a second website but a table of contents with a summary.

• →A first-level heading (H1) with the name of the business or project. This is the only required part.
• →A blockquote (line starting with >) with a short summary in one to three sentences: what, for whom, where.
• →Optionally one or more paragraphs with further details, such as address, opening hours or special features.
• →Sections with second-level headings (H2), for example “Services” or “Practice”, each followed by a list of links in the format [Title](URL) with a short note after a colon.
• →A section named “Optional” for links a language model may skip when space is tight.

Example: llms.txt for Muster Zahnarztpraxis

This is what the file could look like for a fictional dental practice in Berlin. All details are sample data. Make sure the name, address and phone number are written exactly as they appear in your legal notice, on your contact page and in your Google Business Profile.

The links point to the pages that matter most to patients. Each note after the colon says in a few words what the page contains. Nothing more is needed.

# Muster Zahnarztpraxis

> Dental practice in Berlin focusing on prevention, implants and treatment for anxious patients. Appointments by phone on 030 0000000 or online.

Address: Musterstraße 1, 10000 Berlin. Opening hours: Monday to Friday, 8 am to 6 pm. Statutory and private patients welcome.

## Services

- [Professional teeth cleaning](https://www.muster-zahnarztpraxis.example/en/teeth-cleaning): procedure, duration and cost
- [Implants](https://www.muster-zahnarztpraxis.example/en/implants): consultation, treatment and aftercare
- [Anxious patients](https://www.muster-zahnarztpraxis.example/en/dental-anxiety): how the practice handles dental anxiety

## Practice

- [Team](https://www.muster-zahnarztpraxis.example/en/team): dentists and prevention team
- [Contact and directions](https://www.muster-zahnarztpraxis.example/en/contact): address, phone, public transport
- [Book online](https://www.muster-zahnarztpraxis.example/en/booking): book an appointment directly

## Optional

- [FAQ](https://www.muster-zahnarztpraxis.example/en/faq): answers on costs, emergency service and first visits

Is llms.txt an official standard?

No. llms.txt is an open proposal, not a standard from a standards body and not a binding requirement from the major AI providers. There is no public commitment that ChatGPT, Gemini, Claude or Perplexity take the file into account for every request. Some tools and developer documentation sites use it, others ignore it. Anyone who promises you that an llms.txt alone will get you recommended in AI answers is promising more than can be shown.

In our study “AI visibility of Berlin businesses 2026” we checked the websites of 579 Berlin businesses, selected via OpenStreetMap, on 6 October 2026. 12.6 % had an llms.txt. At least 24 of the 73 files we found had been generated automatically by site builders or plugins, so they were not created on purpose. The file is spreading, but often more or less by accident.

Our advice is therefore sober: a well-maintained llms.txt takes little time, does no harm and may help if a tool reads it. It should never replace the fundamentals that search engines and AI systems demonstrably rely on.

What matters more than llms.txt?

AI assistants build their answers from many sources: your website, search indexes, directories and review platforms. For your business to appear correctly there, the fundamentals have to be right. This is where our study shows the biggest gaps: 57.5 % of the websites checked had no structured data from schema.org at all. Only 31.1 % described the business itself as a schema.org type, and only 10.4 % included address and phone number in their structured data.

Structured data is machine-readable information in the source code, usually in JSON-LD format. It states unambiguously: this is a dental practice, this is its name, this is where it is, this is its number and these are its opening hours. Search engines have evaluated this information for years, which makes it a far more reliable foundation than a new, voluntary file.

• →Structured data: LocalBusiness or a fitting subtype with name, address, phone, opening hours and links to your profiles.
• →Consistent business details: name, address and phone number identical everywhere, on your website, in your Google Business Profile and in directories.
• →Access for AI crawlers: in our study only 1.4 % blocked AI crawlers in robots.txt. Still, check that your file does not do so by accident.
• →Clean page structure: exactly one H1 per page, clear subheadings and a meta description. Only 47.8 % of the websites checked had exactly one H1, and 66.1 % had a meta description.
• →Quotable copy: every service page opens with two or three sentences that answer a customer's question directly.

How to create your llms.txt in five steps

An llms.txt can be written in half an hour if your website has a clear structure. If you would rather not start from scratch, our free llms.txt generator helps: you enter the key details and get a file in the right format, which you then review and adjust.

• →Write a short description: in one to three sentences, say what you offer, for whom and where. These sentences go into the blockquote.
• →Pick the important pages: services, contact, booking, team, FAQ. Ten to twenty links are enough for most businesses.
• →Create the file: save it as plain text named llms.txt, UTF-8 encoded, and place it in the root directory of your domain.
• →Check that it is reachable: open your-domain.com/llms.txt in a browser. The file must load without a login and without redirecting to an error page.
• →Maintain it: when you add services, change opening hours or move, update the file together with your website. An outdated llms.txt is worse than none.

Common mistakes and our conclusion

Most problems arise when the file is generated automatically and then forgotten. Typical issues are links to old or deleted pages, test pages, shopping carts or login areas, a summary without a location, and details that do not match the legal notice. A very long file listing every single subpage also misses the point, because llms.txt is meant to make the selection easier.

The conclusion: create an llms.txt once your fundamentals are in place, and keep it short and current. Spend most of your time on structured data, consistent business details and clear copy. The free website check at Website check shows within a few minutes whether your website is already in good shape.