llms.txt: what the file does not do, and what the work behind it actually does

No engine reads the llms.txt file yet, and Google writes that no machine-readable file is needed to appear in its generative features. That finding is about the file. It says nothing about the exercise used to produce it, and that is where the value sits.

llms.txt is a format proposed in 2024: a file at the root of your site declaring to language models which pages matter and what they contain. The idea is sound. It answers a real problem, that of a site whose hierarchy is obvious to a person and unreadable to a machine.

Adoption has not followed. John Mueller, of Google, wrote that as far as he knows no AI service uses it, and he gave the way to check: it shows in your server logs. Google's documentation points the same way, stating that no new machine-readable file is required for its generative features.

So much for the file. Now the exercise, which is another matter entirely. Producing an llms.txt that holds up means deciding which pages truly represent your business, summarizing each in one accurate sentence, removing what is out of date and keeping the whole thing current. Done seriously, that is an inventory of your content and a reordering of your editorial priorities.

That inventory pays immediately, elsewhere than in the file. It surfaces duplicate pages, pages nobody stands behind, pages that should receive internal links and receive none. It becomes the basis for writing briefs and navigation decisions. Its value does not depend on who reads the file.

How to recognize an llms.txt that took work

Open yours and look. A file produced in one click restates your sitemap: the full list of URLs, in the order the system found them, with no description or with the meta description copied over. It takes five minutes to generate and it ages badly.

A considered file does the opposite. It selects, it ranks, it describes each page in words that sound like you, and it leaves three quarters of the site out. You can see at a glance that a person decided something. That version keeps its value the day an engine starts reading the format, and it already serves you today even if nobody reads it.