The llms.txt template for hotels

What is llms.txt

It is a plain Markdown file at the root of your domain, next to robots.txt and sitemap.xml, that tells an AI assistant who you are, what you sell and which page answers which question. For a hotel it is the one place where your rooms, policies, direct booking benefits and contact details are written so a machine reads them without guessing.

So why do it now? Because it costs half an hour, it is the only surface written for agents rather than for readers, and because almost nobody in hospitality has one yet.

llms.txt is not robots.txt, and mixing them is the most common mistake

They sit in the same place and their names look alike, so people write one and expect the other to happen. They do opposite jobs.

The short version: robots.txt is the doorman and llms.txt is the brochure you leave on the table inside. You need both, they are written in different languages, and one never does the other's job. If your file has a Disallow line in it, what you published is a robots.txt with the wrong name.

What the audit found

In August 2026 we requested /llms.txt from 155 hotel chains and 177 independent Spanish hotels, and read every file that answered.

Adoption is 17.3% among Spanish chains, 20.5% among international ones and 6.3% among independent Spanish hotels, counting only files that describe the hotel. Spanish groups try harder than anyone and also break it more often: 42 attempted it and 17 ended up with something that works.

Sixteen serve a file that is literally empty. Somebody created it, published it and never filled it in. Fifteen more return their own homepage in HTML with a 200 status, which looks perfect in a browser and is nothing to a machine. And nine Spanish groups published a permissions notice with robots.txt syntax inside the llms.txt, which never names a single hotel.

Twenty domains could not even be checked, because their firewall answered 403 to an ordinary browser request. That is the bigger problem: a wall that blocks unknown agents blocks the assistants that would have recommended you. Fix access before you worry about the file.

What is inside the template

Twelve blocks, in this order. The structure comes from the best of the seven working files, above all Wyndham's, plus the official spec at llmstxt.org.

1. Name and one-line summary. The H1 and a blockquote saying what you are, where, in which category and what makes you different. This is the part an assistant quotes.
2. Key facts. Address, GPS, phone, email, check-in and check-out times, number of rooms, languages.
3. What this hotel is, and what it is not. casi nadie lo tiene A list of what you are not: not beachfront, not all-inclusive, not for large groups. It stops an assistant recommending you to a guest who will complain.
4. Book direct, with a discount code. casi nadie lo tiene The perks of booking direct instead of through an OTA, and a live discount code. Naming a code is more useful to a guest than the words "book direct", and a code that exists only in this file is the only way to know a booking came from an AI recommendation.
5. Rooms. One line per room type with size, beds, view and capacity.
6. Seasonality and price behaviour. casi nadie lo tiene High and low season, whether prices rise with demand, typical booking lead time, peak dates, closures. A price goes out of date; how the price behaves does not.
7. Dining, facilities, location. Restaurants and hours, spa, pool, parking, distances to the airport and the landmarks.
8. Policies. Cancellation, pets, children, accessibility, payment. What people ask an assistant before booking.
9. FAQ with related terms. casi nadie lo tiene Each answer carries the other words guests use for it, so the assistant matches the question however it is phrased.
10. Languages and URLs. casi nadie lo tiene Your URL pattern plus an explicit rule against building a URL by translating a slug. Every Spanish hotel has two languages and this is where assistants invent dead links.
11. Preferred descriptions by language. casi nadie lo tiene The exact sentence you want quoted, in each language. You write your own headline instead of letting the model draft it.
12. Guidance for AI systems. Never invent availability or pricing, always send guests to the booking engine, use the official name, do not construct URLs.
13. Agent access. casi nadie lo tiene Your machine entrances in one place: the full booking URL as the only supported path to complete a booking, the phone as human handoff, this guide in each language. An agent that has to guess where the transaction happens books through an intermediary.
14. Sitemaps. The most repeated section in the whole corpus, and the cheapest to add.
15. Technical footer. Canonical domain, file version, last updated, next review, and the command to verify it is really published.

How to publish it, in four steps

1. Fill in every placeholder. Replace each [COMPLETE: ...] with verified data from your own systems. If you cannot verify a figure, delete the line instead of guessing: an unfilled placeholder is obvious, a wrong price is not.
2. Save it as llms.txt. Plain text, UTF-8, no .md and no .doc. The filename is part of the convention.
3. Upload it to the root of your domain. It has to open at yourhotel.com/llms.txt. In WordPress that is the folder holding wp-config.php; in most hosting panels it is public_html, htdocs or www. If your site is built with a page builder, ask whoever runs it to add a static file.
4. Verify it with curl, not with your browser. A browser makes a soft 200 look fine. Run the check below: you need a 200, a text content type, and a first line starting with #.

The verification command

curl -sI https://yourhotel.com/llms.txt | head -3

You want HTTP/2 200 and a content-type of text/plain or text/markdown. If it says text/html, your site is serving its homepage at that URL and the file is not really there. That is the single most common failure, and it looks perfectly fine in a browser.

Frequently asked questions