llms.txt: Your Website's Business Card for AI Assistants

llms.txt is a simple text file in Markdown format that sits in the root directory of a domain — reachable at yourdomain.com/llms.txt. It explains to AI systems, in compact form, who is behind the website, what the business offers, and where the important content can be found.

The easiest way to picture it: what robots.txt has been for search engine crawlers for decades, llms.txt wants to be for AI assistants. Just with the sign flipped — robots.txt mostly says what crawlers may not do; llms.txt says what an AI system should know about the website so that it understands and cites it correctly.

For small businesses this is interesting because search behavior is shifting: people who used to google "car repair shop near me" increasingly ask an AI assistant instead. Whether that assistant knows your business, describes it correctly, and can recommend it depends on how well it understands your website — and that is exactly where the file comes in.

Why the topic is becoming relevant right now

Two developments have turned llms.txt from a niche experiment into a topic to take seriously:

First: AI assistants are becoming a search channel. More and more purchase decisions start with a question to an AI assistant instead of a Google search. These systems read websites — but they read differently than humans do: they need clear structure, unambiguous facts, and machine-readable signposts. What that means for businesses overall is covered in detail in our article on getting found in AI search.

Second: measurement tools are catching up. Google's audit tool Lighthouse has, since version 13.3 (May 2026), been testing in the new category "Agentic Browsing" how usable a website is for AI agents — the checks include whether a valid llms.txt sits in the domain root, whether the accessibility tree is clean, and whether the layout stays stable. When the web's most important audit tool explicitly checks for a file, it is no longer an insider tip.

How an llms.txt is structured

The convention is deliberately simple — it is ordinary Markdown that any text editor can write:

1. One main heading with the name of the website or business.
2. A short description right below it: who, what, where — in two or three sentences, with the facts an AI system should reproduce correctly (services, location, special features).
3. Structured link lists under subheadings: the most important pages with a full URL and one short descriptive sentence each, grouped by topic — for example "Main pages", "Services", "Locations".

Nothing more. No programming, no special format, no registration. The art lies not in the technology but in the selection: the file should distill the essentials, not duplicate the entire sitemap.

A live example from practice

One of our client projects shows what this looks like in concrete terms: the independent car repair shop behind maschinen-niedermeier.com. Its file sits at maschinen-niedermeier.com/llms.txt and follows exactly the structure described:

• At the top, the workshop name as a heading, followed by a dense short description: location, brand-independent repairs, on-site TÜV appointments (the mandatory German vehicle inspection) with weekdays and times, breakdown service, Google rating.
• Below it, structured link blocks: main pages, all workshop services from inspection to accident repair, plus the location pages of the service area — every URL with one sentence of context.

An AI assistant asked where you can get a short-notice TÜV appointment in the district finds everything it needs in this one file: the answer, the supporting page, and the contact details. In the Lighthouse audit, the site reaches the full score of 3 out of 3 checked criteria in the "Agentic Browsing" category — valid llms.txt included.

An honest assessment: what llms.txt can and cannot do

For all its topicality, the truth also includes this:

• It is a young quasi-standard. The llms.txt convention is a community proposal, not an adopted web standard. Not every AI system reads the file today, and nobody can guarantee how adoption will develop.
• It is no ranking miracle. An llms.txt alone brings neither better Google positions nor automatic AI recommendations. It makes a good website easier for machines to navigate — it won't save a weak website.
• But: the precaution is practically free. The effort is one to two hours, there are no running costs and no risks. If AI assistants keep growing as a customer source, you are among the websites that were cleanly readable from the start. If not, you have lost a useful short overview of your own website — in other words, nothing.

There is a side effect, too: whoever writes the file has to boil their own website down to the essentials. Many owners notice for the first time that their most important information — services, service area, differentiation — is not clearly stated on the website itself. The llms.txt thus becomes an involuntary quality check.

How to proceed

1. Take stock: list the 10 to 30 most important pages of your website and write one descriptive sentence for each.
2. Write the short description: two or three sentences that describe your business the way an AI assistant should reproduce it — concrete, with location and services, without marketing fluff.
3. Create the file: save it as llms.txt in Markdown format in the root directory of your domain, so that it is reachable at yourdomain.com/llms.txt.
4. Check it: open the file in the browser, spot-check the links, and keep it current when the website changes. If you like, let Lighthouse cross-check the "Agentic Browsing" category.

On all websites we build, llms.txt is now part of the standard scope — just like the sitemap, structured data, and clean meta tags. You'll find examples in the Showroom.

Frequently Asked Questions

Does llms.txt replace robots.txt or the sitemap?

No, it complements both. robots.txt controls what crawlers may do; the sitemap lists all URLs for search engines; llms.txt explains to AI systems, compactly and in curated form, what the website is about. A well-set-up website has all three.

Do AI assistants even read the file yet?

Inconsistently — the standard is young, and providers handle it differently. What is certain: the file does no harm, costs almost nothing, and is already actively checked by audit tools such as Lighthouse. That makes the risk-reward ratio unusually good.

Do I need a programmer for this?

Not for the file itself — it is plain text in Markdown format. You only need access to your website's root directory in order to place the file there. With website builders, exactly that is sometimes the hurdle, because they don't allow access to the root directory.

How long should an llms.txt be?

As short as possible, as complete as necessary. For a typical business website: one heading, three to five sentences of description, and 10 to 40 annotated links, grouped by topic. A file that lists hundreds of URLs without comment misses the point — curation is the point.