llms.txt is a text file in markdown in the root of your site, to be opened at uwdomein.be/llms.txt, that sums up in plain language who you are, what you offer and which pages matter. It is not a standard and not an official format: it comes from a proposal by the community and not one large supplier of AI answer engines has publicly confirmed that it fetches and uses the file. The evidence that it yields citations is therefore thin. Against that stands the fact that it costs nothing, breaks nothing and that the most valuable block in it is your own figures and claims, work you had to do anyway. Make it in an hour and expect no miracle from it.
What exactly is in it
The file is markdown. No XML, no JSON, no settings. Anyone who opens it reads plain text.
The proposal on llmstxt.org describes a fixed build-up. At the top one H1 with the name of the organisation. Below that a blockquote of one paragraph that sums up in one go what you do. After that free paragraphs with background. At the bottom sections with a list of links, each with a short description next to it, so that a reader who does not know the site knows where to be. There is also a variant llms-full.txt with the full text of your most important pages in one file.
Do not confuse it with robots.txt. robots.txt says who may come in. llms.txt says what is to be found inside and how you would sum it up yourself. Chapter 7 is about the first and this chapter about the second. Anyone who writes an llms.txt and meanwhile shuts out the AI crawlers has done nothing.
No standard and no official support
This you ought to know before you expect anything of it. llms.txt is a proposal and not a norm from a standards organisation. There is no validator from a search engine, no report in Search Console and no public confirmation from OpenAI, Google, Anthropic or Perplexity that their systems read the file. We have not found such a confirmation. Anyone who tells you that llms.txt improves your position in ChatGPT cannot underpin that today.
Google is explicit about this by now. In the official guide for generative AI features, updated on 15 June 2026, it says that you do not have to make separate machine-readable files because Google Search does not use them. John Mueller of Google compared the proposal with the old meta keywords tag. An independent measurement by SE Ranking across 300,000 domains found an adoption of about 10 per cent, of which almost nothing was ever fetched. Fair remains: Perplexity, Claude and coding agents do read the file, so for that side it holds value.
What is true: it is plain text at a plain URL. A crawler that may read your site can fetch it like any other page. Whether it does anything with it we do not know and nobody publishes that.
An llms.txt to take over
Replace the example text. The headings you can leave standing.
# Your Company
> One paragraph that says word for word what you do, for whom and in which
> region. No slogan. Someone who does not know you must know after these
> sentences whether this is the right place.
Your Company is a [what you are] in [place]. We do [what we do] and we do not
do [what we do not]. Active in [regions]. Founded in [year].
## Offering
- [Service 1]: what it is, how long it takes, from what amount.
- [Service 2]: what it is, how long it takes, from what amount.
- [Service 3]: what it is, how long it takes, from what amount.
## Most important pages
- [Home](https://uwdomein.be/): positioning and overview of the offering.
- [Services](https://uwdomein.be/diensten.html): packages with deliverables.
- [About us](https://uwdomein.be/over.html): who works here and what we stand for.
- [Contact](https://uwdomein.be/contact.html): reachability and response time.
## Key figures and claims
- [A statement with a number in it that you can defend.]
- [A second one with the unit and the period added.]
- [A third, about a mistake you often see in your trade.]
## How we differ
- [Something demonstrable, not an adjective.]
- [Something a competitor cannot simply say about itself.]
## Who it is for
- Company size, sectors, roles of the decision maker, occasions.
## Languages and regions
Site available in [languages]. Markets: [countries].
## Contact
- E-mail: hallo@uwdomein.be
- Response time: within one working day
- First conversation: free, 30 minutes
Our own file at semanu.be/llms.txt is built along this pattern and counts 780 words and about 6 kB. That is a usable upper limit. Our own rule: stay under a thousand words. If it gets longer, it is a second website and not a summary.
Which block is the most valuable
Key figures and claims. Far above the rest.
The reason is in chapter 6. An answer engine takes over what is concrete and leaves lying what is marketing language. "Broad experience in the sector" is never cited. "Two to five times the quotation over five years" is. Your list of links is largely superfluous for a machine, because it can read your sitemap itself. Your figures stand nowhere else together in one block.
One rule with it: every figure you put here must also stand on an ordinary page, in running text. This file is no place to claim something you do not dare write elsewhere. If it is not right on the page, it is not right here either.
| Block | What it serves for | Our verdict |
|---|---|---|
| H1 with the name | Ties the file to your entity | Indispensable |
| Summary in one paragraph | The sentence taken over at "what is this company" | Indispensable |
| Offering with prices or orders of magnitude | Answers the question what it costs | High |
| Key figures and claims | The only truly quotable material | The most valuable |
| Most important pages with a description | Points the way to the source | High |
| How you differ | Sets you apart from four similar companies | Moderate |
| Who you work for | Helps with questions with a sector or company size in them | Moderate |
| Full list of links of the whole site | Doubles your sitemap | Superfluous |
Do you need a generator
There are generators and plugins that turn your sitemap into a list of links and make a file out of that. That yields exactly the bottom part of the table above.
The blocks that matter no generator can write for you. Your summary in one paragraph, your prices and your figures stand nowhere in your sitemap. Do it by hand in a text editor. For an SMB site that is an hour, thinking included.
The honest verdict
The costs are an hour of work, zero euros and no risk. The file cannot harm your findability, because no search engine looks at it.
The benefits are unproven. That is an uncomfortable answer to a search that thousands of people type in. It is however the answer.
What it certainly yields is the exercise. Anyone who has to sum up their offering, prices and claims in eight hundred words notices within half an hour which of those stand nowhere on the site or are worded differently elsewhere. That text ought afterwards to end up on your ordinary pages as well, because there it is read for certain.
If you have to choose between llms.txt and structured data, choose chapter 4. If you have to choose between llms.txt and access for the crawlers, choose chapter 7. Put this file on the list of things you do when the rest is ready and not before it.
What you can do yourself and what you cannot
To do yourself, without a developer:
- Copy the example above and fill it in. An hour.
-
Save the file as
llms.txtand put it in the root of your site, next torobots.txtandsitemap.xml. -
Check whether
uwdomein.be/llms.txtshows your text and not a 404 page. - Update it twice a year, together with your robots.txt from chapter 7. A file with prices from last year is worse than no file.
What you need help with:
- A CMS where you cannot reach the root. With WordPress you put the file in the webroot over FTP or you use a plugin. Ask your web builder, it is five minutes of work.
- The translation into EN, DE and FR if you work multilingually. One file per site is enough. The summary in the language of your largest market is however the safest choice.
- The figures themselves. If you have not one number that you dare defend publicly, that is no text problem.
Frequently asked questions
What is llms.txt?
A markdown file in the root of a website, to be opened at uwdomein.be/llms.txt, that sums up in plain language what the company does, what it offers and which pages are the most important. It is meant as help for language models that have to sum up a site without reading all the pages.
Is llms.txt an official standard?
No. It is a proposal published on llmstxt.org and not a norm from a standards organisation. There is no validator, no report in Search Console and no public confirmation from OpenAI, Google, Anthropic or Perplexity that their systems use the file.
Does llms.txt help to be mentioned in ChatGPT?
That has not been demonstrated. The evidence is thin and anyone who claims the opposite cannot underpin it. What does count for citations is access for the AI crawlers (chapter 7), structured data (chapter 4) and concrete figures in your ordinary text.
Do I need an llms.txt generator?
No. Generators mostly turn your sitemap into a list of links and that is exactly the least valuable part of the file. The summary, the prices and the key figures stand nowhere in your sitemap, so those you write yourself. An hour in a text editor is enough.
Where does the file have to be?
In the root of your domain, next to robots.txt and sitemap.xml, so that it is reachable at uwdomein.be/llms.txt. After placing it check whether that URL shows your text and does not return a 404 page.
Would you rather not do this yourself: SEMANU carries out the full SEO and GEO audit for 1,450 euro, report and working session included. What it covers is on the guide page.
Measurability and findability belong in your software strategy
Measurability and findability are part of strategic advice, not a separate discipline. If you would rather not do this yourself, we take it up in the software strategy: what you measure, which threshold justifies an action and who follows it up.