llms.txt for a Dubai business: what it is, how to write it

llms.txt is a plain Markdown file at the root of your site (yoursite.com/llms.txt) that tells AI tools, in a few hundred words, what your business is and which pages matter most. You write it by hand: a title, a one-paragraph summary, and short lists of links to your key pages with one line of explanation each.

The file costs almost nothing to make. Expectations are where people go wrong, so this article covers who reads it today, the exact format, a worked example and the mistakes that make it useless.

Where llms.txt comes from

Jeremy Howard of Answer.AI proposed llms.txt in September 2024. The specification lives at llmstxt.org. The idea: language models work with a limited amount of text at a time, and a typical business website wastes that space on menus, cookie banners, scripts and repeated footers. A curated Markdown file gives a model the important facts and links without the noise.

llms.txt is a proposal, without the status of a web standard. No body enforces it, and AI companies decide for themselves whether to read it.

Who reads it today

Be honest with yourself about this part before you spend time on it.

AI tools that fetch pages live. When someone asks an assistant to look at a specific website, or when a coding or research agent browses, the file gives that tool a map. Many developer documentation sites publish llms.txt for this reason.
Search engines and AI answer products. Neither Google nor the major AI providers have documented llms.txt as a signal their search or answer products rely on. Treat it as a hint some tools may use.

So why bother? Two reasons. It takes an hour and carries no risk. And writing it forces you to state, in plain words, what you sell, where, for whom and on which page. That same clarity is what makes your main pages easier for AI assistants to quote. The wider mechanics sit in how AI assistants choose who to recommend.

The format, line by line

The specification defines a fixed order. Only the first element is required.

An H1 with the name of the business or project. One line: # Business Name.
A blockquote summary: one or two sentences starting with >. This is the part a tool is most likely to repeat, so put the core facts here: what you do, where, for whom.
Optional free text: short paragraphs or bullet points with no headings. Use this for facts a model should not get wrong: service area, languages, opening hours, what you do not offer.
H2 sections with link lists. Each line follows the pattern - [Page title](https://full-url): one-line note.
A section called ## Optional, if you need one. The specification says tools may skip links under this heading when space is short, so put secondary pages there.

The file is plain Markdown text, and your server should return it as raw text with no HTML template around it. The proposal also suggests offering Markdown copies of key pages at the same address with .md added. That step is optional, and most small businesses can skip it.

Some sites also publish llms-full.txt, a single file with the full text of their documentation. The convention grew out of documentation tools after the original proposal. For a clinic or a café it adds little.

A worked example for a Dubai business

Hypothetical example: a dental clinic in Dubai Marina with an English site and a Russian section. The domain and details are invented.

# Sample Dental Clinic

> Private dental clinic in Dubai Marina offering general dentistry, implants,
> aligners and children's dentistry. Licensed by the Dubai Health Authority.
> Consultations in English, Russian and Arabic.

- Location: Marina Walk, Dubai Marina, Dubai, UAE
- Hours: Saturday to Thursday 9:00–21:00, Friday 14:00–21:00
- Booking: online form or WhatsApp; walk-ins for emergencies
- We do not offer cosmetic surgery or general medicine

## Services
- [Dental implants](https://example.ae/implants/): procedure, stages, what affects the price
- [Clear aligners](https://example.ae/aligners/): who they suit, treatment length
- [Children's dentistry](https://example.ae/kids/): first visit, age from 3

## Practical information
- [Prices](https://example.ae/prices/): price ranges in AED, insurance accepted
- [Contact and directions](https://example.ae/contact/): map, parking, WhatsApp link
- [Our doctors](https://example.ae/doctors/): names, licences, languages spoken

## Russian version
- [Главная на русском](https://example.ae/ru/): услуги и запись на русском языке

## Optional
- [Blog](https://example.ae/blog/): articles on dental care
- [Careers](https://example.ae/careers/)

Notice what the example leaves out: slogans, "best clinic in Dubai", review scores and every blog post ever written. Each line is a fact a person could check on the linked page.

llms.txt, robots.txt and sitemap.xml compared

Owners often confuse these three files. They do different jobs and you need all of them.

File	Job	Format	Who reads it	Binding?
robots.txt	Says which bots may crawl which paths	Plain text rules	Search and AI crawlers that respect it	Crawlers that follow the convention obey it
sitemap.xml	Lists every page you want indexed, with dates	XML	Search engines	A hint for discovery
llms.txt	Summarises the business and points to key pages	Markdown	AI tools that choose to read it	A hint only

Check one interaction in particular: if robots.txt blocks AI crawlers, a link in llms.txt does not override that block. Decide your crawler policy first, then write llms.txt to match it.

Two sites I built with llms.txt

The Moon Mars site (moonmars.ae, case here) has llms.txt next to its sitemap. It is a bilingual English and Russian site with 128 pages, service pages for shisha outsourcing, catering, consulting and technology, city pages for Abu Dhabi and Doha, and a catalogue of 133 lounges. A site of that size needs a short reading list more than a five-page site does, because a model will not read 128 pages to find the four that describe the service.

The Lucent Rental site I built for a supercar rental has llms.txt, sitemap.xml and robots.txt. Treat the three files as one task. robots.txt sets access, the sitemap lists pages, llms.txt explains which of them matter. If you are planning a new site, put all three into the brief for your website so nobody has to remember them after launch.

Bilingual and trilingual sites

Many businesses in the UAE run English, Russian and Arabic versions. A few practical rules:

Keep one file at /llms.txt. The specification does not define per-language files, so tools will look at the root.
Write the summary in English, the language most tools default to.
Add a separate H2 section for each other language, named in that language or clearly labelled ("Russian version", "النسخة العربية").
Link to the real language URLs (/ru/, /ar/). An automatic translation widget gives a tool nothing to read.
Keep facts identical across languages. If the Russian page lists different hours from the English page, a model has to guess which is right.

If your language versions have structural problems, fix those first. Site structure for SEO covers how pages should map to services.

What not to do

Stuffing keywords. Lines like "best dentist Dubai cheap dentist Dubai Marina" read as spam to a person and add nothing for a model.
Claims you cannot back up. Awards you did not win, invented review counts, "number one in the UAE". If a tool repeats them, you own the consequences; if it cross-checks them, you lose credibility.
Copying the sitemap. Two hundred links defeats the purpose. Aim for 5 to 20 links in the main sections.
Linking to broken or blocked pages. Every URL should return a working page and should not be disallowed in robots.txt.
Putting the file in a subfolder. /files/llms.txt or /en/llms.txt will not be found. It belongs at the root.
Serving HTML. Some website builders wrap uploaded files in a page template. Open the URL in a browser and check you see raw text.
Letting it go stale. Old prices, a closed branch or a phone number nobody answers are worse than no file.
Treating it as your GEO strategy. The file points to content. If the pages behind the links are thin, the pointer does not help. Writing articles that AI cites is where most of the work sits.

Questions to ask your web developer

Use these when you brief a developer or review an existing site:

Do we have /llms.txt at the root, and does it return plain text with a 200 status?
Which crawlers does our robots.txt allow or block, and does that match what we want from AI tools?
Will the file update when we add or remove a service page, or does someone edit it by hand?
Do all links in the file point to the canonical URL of each page, without redirects?

If the answer to the third question is "by hand", agree who owns that edit and put it in the monthly routine.

How to check it works

You cannot see who reads the file, but you can verify that it is correct and reachable.

Open yoursite.com/llms.txt in a private browser window. You should see Markdown text with no site header or footer.
Click every link. Each should open the intended page directly.
Search your server logs or hosting analytics for requests to /llms.txt. Requests from known AI user agents show that some tools fetch it; no requests tells you nothing either way.
Ask an AI assistant with browsing to read your llms.txt and summarise your business. Compare the answer with your summary. For a fuller test of what assistants say about you, follow this check.

Checklist

Write a one-line H1 with your business name.
Write a two-sentence blockquote: what you do, where, for whom, in which languages.
List hard facts as bullets: location, hours, booking method, what you do not offer.
Pick 5 to 20 key pages and group them under H2 headings by topic.
Give each link a one-line note that a person could verify on the page.
Add a labelled section for each other language version.
Move secondary pages under ## Optional.
Upload to the root, confirm it returns plain text, click every link.
Confirm robots.txt does not block the pages you list.
Add "check llms.txt" to your monthly Search Console routine and update it the same day a service, price or contact detail changes.

Questions and answers

Will llms.txt make ChatGPT recommend my business?

No one can promise that. llms.txt gives AI tools a clean summary and a short reading list. Whether an assistant recommends you depends on your pages, your reviews and what other sites say about you.

Do I need a developer to add llms.txt?

You can write the text yourself in an hour. A developer uploads it to the root of your domain in a few minutes, or you can do it yourself if your website builder lets you add files at the root.

Should I write separate files for English, Russian and Arabic?

The proposal describes one file at /llms.txt. Write it in English and add a labelled section that links to your Russian or Arabic pages, so a tool working in that language finds them.

How often should I update llms.txt?

Update it whenever a page in the file moves, a service changes or your contact details change. Add it to the monthly check you already do in Google Search Console.