The real meaning of the llms.txt file for LLM systems: Results of independent AI test No. 3

Over the past few months, the idea that having an llms.txt file almost guarantees that your content will be included in responses from Large Language Models (LLMs) has been actively promoted. It’s been featured in blogs, recommended by niche influencers, and even included in service offerings by companies providing AEO (AI Engine Optimization) services. But is this truly the case?

The llms.txt file is positioned as a standard that provides LLM systems with a map of a website’s important content. However, there’s currently no confirmation that AI models actually use it. We’ve seen a lot of information about it, but no real evidence that llms.txt works.

Here are a few facts about the file and why it has become a topic of discussion:

• llms.txt is a text file, typically placed in the root directory of a domain (e.g., https://example.com/llms.txt). Using Markdown or similar formatting, it contains a list of URLs or descriptions of pages that the site owner considers “key,” “high-quality,” and “priority content” for LLM systems.
• The llms.txt initiative is currently a proposed standard; it is not officially confirmed or widely adopted by major language models or companies. Nevertheless, it is actively discussed within the SEO community. According to interviews and public statements, none of the major language model providers (OpenAI / ChatGPT, Anthropic / Claude, Google / Gemini, etc.) have officially confirmed that they read the llms.txt file or that it is part of their indexing or processing pipeline. This is what prompted us to investigate the matter.
• One of the reasons for the increased attention is that traditional files like sitemap.xml or robots.txt don’t always perfectly address how LLM systems search for information when generating responses. Therefore, the hype around it was fueled by a desperate attempt to find a “silver bullet” that would quickly get content into AI responses. Since everyone wants to believe in the best and find the easiest paths to their goals, the buzz around llms.txt continues to grow in popularity.

John Mueller from Google explicitly stated that, to his knowledge, site server logs do not show requests for llms.txt from major AI systems. He compared the file’s functionality to a keywords meta tag (i.e., more of a supplementary signal than a guaranteed method).

And if you ask AI assistants directly whether they use llms.txt, they will all deny it.

Despite the lack of official support, the topic is not being ignored by other professional tools. The most notable examples are Yoast SEO and Rankmath.

Yoast SEO is one of the world’s most popular WordPress plugins, helping to optimize sites for search engines. It has millions of users worldwide, and in June 2025, it released a feature that automatically generates an llms.txt file, highlighting the most important and relevant site content with a single click and no complex technical setup.

Rankmath is another well-known WordPress plugin. In July 2025, it also added the ability to create an llms.txt file. Rankmath emphasizes customization, allowing users to manually specify which blocks or pages should be included in the file and to set content priority.

When large plugins with a massive user base begin to support such a feature, it, first, draws attention to it (in this case, llms.txt), and second, creates the groundwork for widespread testing, data collection, and potential better adaptation of the standard in the future.

This led us to form a hypothesis about llms.txt files and designate this experiment as a priority. Our goals were to determine:

1. Do LLM systems actually access llms.txt?
2. Do they follow the links provided in this file, even if these pages are “orphans” (with no external or internal links)?
3. Can pages optimized in this way appear in responses generated by ChatGPT and other AI systems, assuming the model has “read” these pages via llms.txt?

Experiment Hypothesis and Tasks

Our hypothesis is that LLM systems might use the llms.txt file to:

• Discover key pages on a site that the owner deems important.
• Access the URLs specified in the file even when those pages are not navigationally linked to other parts of the site.
• As a result, make the content from these pages available for responses (e.g., in a ChatGPT answer), even if traditional internal linking or standard indexing doesn’t provide a direct path.

To test this hypothesis, we formulated the following tasks:

1. Determine if LLM bots are accessing the llms.txt file. We would check for records in server logs showing attempts to retrieve the llms.txt file.
2. Verify if models and bot agents are following the links listed in the file. We would check if they visit the pages specified in llms.txt.
3. Confirm whether the optimized pages appear in ChatGPT’s responses (or other LLMs) if the model has access to them via llms.txt. For example, would a made-up word we used to optimize the pages appear in the model’s responses if the model “visited” these pages via llms.txt?

Experiment Methodology

To ensure the results were representative and had broader meaning, we selected three websites that differ significantly in several key metrics:

• Topic: The sites belong to different industries.
• Popularity and Traffic: One site has relatively high organic and overall traffic, another has average traffic, and the third is less known. This would help us see if behavior varies by scale.
• Frequency in ChatGPT Responses: We chose sites that had already been mentioned or used in model responses (i.e., the model “knows” of their existence) to reduce the likelihood of them being completely unknown. One of these sites was our own luxeo.team, and the other two were additional projects over which we had full control to monitor everything and miss nothing.

To perform these tasks, our team took the following steps:

1. Created llms.txt files for each of the three sites, using the format proposed on llmstxt.org.
2. Added links to “isolated” pages that were not part of the main site navigation to the files to check if an LLM could reach them via llms.txt.
3. Optimized these “isolated” pages with made-up words to easily track if this content appeared in ChatGPT or other LLM responses after the model read these pages.
4. Configured server log collection (specifically access.log) to monitor:
   • Access to the llms.txt file.
   • Access to the orphan pages.
   • The source of requests (User Agent, IP, time) to understand if they were automated or bot agents, or possibly LLM services.
5. Conducted several checks at different times and from different accounts to rule out the influence of caching, random delays, or other random factors.

Experiment Results

After several weeks of monitoring, we collected data from the three sites and analyzed the server logs. The main task was to check if LLMs or their associated bots were truly accessing the llms.txt file and if there was any subsequent interaction with the pages. The results were revealing, but they did not support our hypothesis.

First, the logs recorded no requests for llms.txt files from LLM system bots or traditional search engines.

LLM bots scanned the sites’ pages, but not a single one accessed llms.txt. This means that no system, such as ChatGPT, Claude, or Google Gemini, currently uses llms.txt as a source for content parsing.

Second, we found no visits to the specially created pages. Since these pages had no other way of being accessed, their only “way out into the world” was being included in llms.txt. The lack of visits proves that this file is not currently a signal for crawling content.

Third, the made-up words we used to optimize the orphan pages (e.g., “Vliglotula” on one of the sites) did not appear in ChatGPT’s responses during additional tests. This further confirmed that the model did not have access to these pages and did not get any information from them.

Separately, we conducted an additional check to see if LLM systems accessed sitemap.xml, as this file is a standard tool and llms.txt essentially duplicates it in some sense. As a result, we were able to record some activity. We observed requests from Claude to sitemap.xml, which may indicate minimal testing or use of this file as a source of information. No requests to sitemap.xml from other AI assistants were found in the logs.

Thus, all the main points of our check yielded negative results, which we suspected, but were still surprised to confirm.

Experiment Conclusions

The experiment results indicate that the llms.txt file is not yet being used by LLM systems to gather information about websites. This confirms the initial doubts we had during the planning stage: too many factors suggested that this initiative was more theoretical than a genuinely implemented practice.

The fact that not a single LLM system used the content from the optimized pages—even those based on a unique made-up word—is especially important. The implementation of new mechanisms is possible in the future, but today there is no confirmation that llms.txt works.

Therefore, the experiment proves that llms.txt at this stage exists more as a “proposal for discussion” than as a tool that genuinely influences LLM interaction with websites.

Conclusion and Next Steps

Not everything that is actively discussed on the internet corresponds to reality. The experiment showed that llms.txt currently has no practical use and does not influence the indexing or use of content in LLM responses. It can be mentioned to clients as an additional service with potential for the future, but it is not advisable to rely on it now or allocate client budgets for it by making empty promises. However, the concept itself has the potential to develop within a year due to active attention.

Our further hypotheses for testing, the results of which we are already preparing for publication, are:

• Do links from websites (or pages) cited by LLMs increase visibility in AI responses?
• Do LLMs take into account reviews from different resources about products or services when forming responses?

Author

Dmytro Kovshun

Dmytro Kovshun is the founder of Luxeo Team – an SEO Outsourcing Company. As a leading specialist in the industry, he is recognized as an expert in SEO promotion of websites. With years of experience and a deep understanding of the field, Dmytro continues to drive success and innovation in SEO strategies, helping businesses achieve their online goals.