Over the past few months, there has been increasing discussion about how artificial intelligence systems discover and use information from websites. With the rise of AI search, the idea of llms.txt
has also gained attention - a text file intended to provide AI systems with structured information about a website and its most important content.
Instead of just reading about it, I decided to test it myself.
I added llms.txt
to four of my websites, all built with ClassicPress. In this article, I’ll explain how I implemented it, what I put in the files, and what I learned from the process.
llms.txt
is a plain text file placed in the root directory of a website:
https://example.com/llms.txt
The idea is to provide AI systems with a concise and structured description of a website - for example, what the site is about and which pages or resources are most important.
This is different from robots.txt
.
robots.txt
is primarily used to provide instructions to automated crawlers about which parts of a website they should or should not access.
llms.txt
has a different purpose: it is intended to make it easier for language models to understand the structure and important content of a website.
There is one important caveat, though: llms.txt
is not a magic way to make AI systems use your website, and it does not guarantee visibility in ChatGPT, Gemini, or other AI services.
That is why I approached it as an experiment.
I manage several websites and actively work on their technical and SEO optimization.
Rather than writing about llms.txt
purely from a theoretical perspective, I wanted to see what would happen if I implemented it on real websites.
This gave me the opportunity to experiment with the structure while keeping full control over what information I made available.
There is another reason behind my approach, and it is related less to the technical implementation and more to the future of search.
I don't want my websites to simply become sources of information that an AI system summarizes for users while the original website receives no visit.
This is the zero-click problem.
If an AI system has enough information to answer a user's question directly, the user may have no reason to visit the original page.
For me, this is an important trade-off.
On one hand, I want my content to be discoverable and understandable by new AI systems. On the other hand, I want users to reach the actual website when the information is useful to them.
That is why I did not put the full content of my articles into llms.txt
, and I did not try to turn the file into a replacement for the website itself.
Instead, I chose a more limited approach: provide information about the website and point to its most important resources without handing over all of the content in one place.
In other words, I prefer llms.txt
to act as a map to my content, not a copy of my content.
In its simplest form, the file can contain a title, a short description, and links to important resources.
For example:
> A short description of the website and its main topic.
## Main Content
- [Homepage](https://example.com/)
- [Category 1](https://example.com/category/category-1/)
- [Important Article](https://example.com/important-article/)
It is just a plain text file. No database or special software is required to create it.
Since my websites are built with ClassicPress, I could simply create an llms.txt
file in the root directory of each website.
It then becomes available at:
https://example.com/llms.txt
This is one of the things I like about the approach: the implementation is extremely simple.
I didn't need to install a large plugin just to create a single text file.
This is one of the parts I consider particularly important.
The goal is not necessarily to put every page of a website into llms.txt
.
If a website has hundreds or thousands of pages, an enormous list of URLs may not be the most useful way to represent the site.
I preferred to identify the main resources and keep the file clear and easy to understand.
That is also why I see llms.txt
more as a structured entry point to the content than as another sitemap system.
Probably the most convenient part of the experiment was that I didn't have to make my websites more complicated.
llms.txt
is a plain text file and can be created and edited very easily.
For smaller websites, this makes it particularly convenient.
This is probably the most important lesson.
I would not say:
"I added llms.txt and my website started appearing more often in AI."
That kind of claim would require actual data and controlled measurement.
For now, I consider llms.txt
an additional way to structure and describe website content, rather than a guaranteed SEO or AI ranking factor.
llms.txt
does not replace the fundamentals of a good website:
In other words, there is little point in adding llms.txt
to a poorly structured website and expecting it to solve the underlying problems.
Based on my experience so far - yes, if the implementation is easy and doesn't add unnecessary complexity to the website.
I wouldn't spend a huge amount of time on it, and I wouldn't present it as a mandatory SEO technique.
But if I can create a clear file in a short amount of time that describes the website and points to its most important content, I don't see much reason not to do it.
Especially when a website has a clearly defined topic and a manageable number of important resources.
I added llms.txt
to four of my own projects. One of them is Proverkite.com, a Bulgarian website with practical guides and informational content. It was a good example for this experiment because the site contains a substantial amount of structured, topic-focused content.
Working with real websites allowed me to apply the same idea to different types of content and see which elements made the most sense when structuring the file.
These are real projects rather than demo websites, which was important to me.
If I were starting from scratch, I would think about the structure of llms.txt
when building the website itself.
The reason is simple: once a website already contains a lot of content, you have to decide which pages are actually the most important.
With a new website, that structure can be planned from the beginning.
llms.txt
is an interesting idea, but I think it should be used with realistic expectations.
I don't see it as a replacement for SEO, sitemaps, or structured data.
Instead, I see it as another possible way of presenting website content in a clear and structured format.
At the same time, I don't want to make it easier for my websites to become sources for zero-click answers, where a user gets everything they need from an AI system without ever visiting the original website.
That is why I chose a moderate approach: help AI systems understand what is available on my websites without handing them the complete content on a plate.
The biggest benefit for me has not necessarily been the file itself, but the experiment.
Instead of waiting for someone to tell me whether llms.txt is useful or useless, I implemented it on real websites and started observing what happens.
And that is the approach I would recommend to other website owners as well:
Test. Measure. And don't assume that every new AI or SEO technique is proven just because it is popular.