Setting up llms.txt: a step-by-step guide for companies

What an llms.txt is and what it is not

People navigate a website through menus, images and design. AI systems do not need that, it rather gets in their way. The llms.txt is the attempt to give them the essentials directly: a plain text file without design, without navigation, only core statements and links to the pages where the detailed answers live.

What matters is what it is not. It is no guarantee that an AI will recommend you, and no substitute for good content. It is a convenience: it helps the machine grasp your facts quickly and correctly. Whether a recommendation results from it is decided by the content behind it, not by the file itself.

Where the file belongs

The llms.txt lives in the root directory of your domain, that is at your-domain.com/llms.txt. Exactly where the robots.txt lives too. It must be reachable as a pure text file, without login, without redirect. The structure follows a loose convention based on Markdown: a heading with your name, a short description block, then thematically sorted sections with links.

Technically, creating it is trivial: create a text file, fill it in, upload it to the root directory. Anyone using a modern website system often simply places it in the public folder. The effort lies not in the technology but in the discipline of keeping the content short and true.

Step by step

Start with a heading and a description block of two or three sentences: who you are, where you are, whom you are there for. Then come the sections. A section for your offering with links to the service pages. A section for facts that are often asked about, such as opening hours, prices, directions. A section with the pages that answer your most common customer questions.

Phrase every entry so that it can stand alone. Instead of "Experience unique moments" you write "Boutique hotel with 24 rooms, its own sauna, a five-minute walk to the lake". The first sentence is advertising, the second is an answer. Only the second helps the machine.

• Set a heading with the company name
• Description block: who, where, for whom (2 to 3 sentences)
• Offering section with links to the service pages
• Facts section: opening hours, prices, directions, distinctive features
• Answers section: links to pages that answer concrete customer questions
• Upload the file as /llms.txt to the root directory

A template to adopt

As a starting point you can adapt the following pattern. Replace the placeholders with your real facts and cut whatever does not apply. Keep everything brief; an llms.txt is not a brochure.

Example: first the line with the name, then a description block "We are a family business in Rottach-Egern and rent out e-bikes for tours around Lake Tegernsee." After that an offering section with links to prices and availability, a facts section with opening hours and daily rate, and an answers section that points to a page "E-bike tours at Lake Tegernsee". The result is a file from which an AI pulls the essentials in seconds.

Common mistakes

The biggest mistake is advertising language. An llms.txt full of adjectives helps no one; it should deliver facts, not mood. The second mistake is outdated or contradictory information. If the file lists different opening hours than the website and the directories do, you confuse the machine more than you help it. Keep the file in sync with your other sources.

The third mistake is to create the file and then forget it. If something about the offering changes, the llms.txt belongs updated like any other important page. It is a living document, not a one-time checkmark.

What the llms.txt alone cannot do

The file is a building block, not a miracle cure. It makes reading easier, but the real work remains: building answers to real customer questions on the web, gathering mentions in trustworthy sources, clearing up contradictions. The llms.txt shows the AI the way to your answers; you have to deliver the answers themselves.

It therefore works best as part of a whole: good, quotable content, a clean technical structure and an llms.txt that ties both together. Then the machine reads quickly, understands correctly and has something it can recommend.

llms.txt and robots.txt: the difference

Both files live in the same directory and are pure text, but they have different jobs. The robots.txt governs access: it tells automated visitors which areas they may view and which not. It is a barrier, not content. The llms.txt, by contrast, delivers content: it summarizes what it is about and points the way to the most important answers.

The two complement each other but do not replace each other. A robots.txt that blocks everything makes you invisible, no matter how good the llms.txt is. Conversely, the best access rule is useless if the machine, once on the page, finds only advertising and no facts. So check both: does the robots.txt allow the relevant pages, and does the llms.txt lead to real answers?

A complete example

To make the whole thing tangible, here is a continuous example for a small business. At the very top is the line with the name, say "Radhaus Tegernsee". Below it a description block: "Family business in Rottach-Egern. We sell and repair bicycles and rent out e-bikes for tours around Lake Tegernsee. Workshop with same-day service."

Three sections follow. Under "Offering" are links to the pages for sales, repair and e-bike rental, each with a short half-sentence about what can be found there. Under "Facts" are opening hours, the daily rate for the rental and the address, every detail brief and verifiable. Under "Answers" a link points to a page "E-bike tours at Lake Tegernsee: routes, prices, equipment", exactly the content that answers frequent customer questions.

The result is a file that a machine can grasp in a few seconds: who the business is, what it does, where the details lie. No line is advertising, every one carries a piece of information. That is exactly what an llms.txt should look like.

How to check whether your llms.txt works

The first test is trivial: open your-domain.com/llms.txt in the browser. If the plain text appears without errors and without a redirect, the file is reachable. Then check every link in it: does it lead to a page that actually delivers the promised answer? Dead or misleading links do more harm than the file does good.

The second test is about content. Read every line through the eyes of a machine and ask: is this a verifiable statement or a mood? Delete everything that is just advertising. The third test is the comparison with your other sources: do the opening hours in the llms.txt match those on the website and in the directories? Contradictions are the only thing that can actively make the file worse.

In the end only measurement provides real proof of effect. After creating the file, ask your most important customer questions again to several AIs and observe over the following weeks whether anything moves. Because AIs pick up new information with a delay, patience is required: the llms.txt is a signpost that only takes effect once the machine finds it on its next visit and also reads the linked answers.

For whom the effort pays off especially

An llms.txt costs little, but its benefit is unevenly distributed. Those who benefit most are businesses whose strengths can be expressed in clear facts and that live on recommendations: hotels, practices, law firms, trades, restaurants, local service providers. For them the single AI answer often directly decides an inquiry, and the easier the machine finds the facts, the more likely the business appears in that answer.

The effect is less immediate for very large websites with thousands of pages, where a single overview file can only depict a fraction. There too it does no harm, but it does not replace a clean structure across the entire site. And for businesses whose offering is hard to capture in facts, the real work lies anyway in making the strengths verifiable in the first place.

The honest recommendation is therefore: for most small and medium-sized companies the llms.txt is one of the cheapest sensible measures there is, done in an hour and without running costs. It is no miracle cure, but a low-hanging apple you should pick while tackling the bigger tasks.