Markdown per page
The same page available at its address plus index.md. Agents would rather fetch the source than take HTML apart.
llms.txt explained
A table of contents for machines.
The llms.txt is a curated short form of a website in Markdown. It is cheap to make, does no harm and is not a switch. This page says what it delivers.
Definition
What sits in the file.
The llms.txt is a file in a website's root directory listing its most important pages in Markdown, each with title, address and a short description. The proposal comes out of the llmstxt.org circle.
The structure is plain: a heading with the site name, a short paragraph placing it, then sections with lists of links. Ours sits at llms.txt and can be looked at right away.
The purpose is orientation. An agent seeing a website for the first time otherwise has to guess which of 122 pages are relevant. The file takes that work off it, much like a table of contents.
Honestly
What is evidenced and what is not.
What is not evidenced is any effect on visibility. No major provider has confirmed evaluating the file. Anyone selling llms.txt as a ranking lever is overstating it considerably.
What is evidenced is the effort: ours gets generated during the build from the page list, so it costs no ongoing maintenance. That is exactly why we build it: the effort is close to zero, the possible benefit above zero.
More important than the file is the discipline behind it. Anyone maintaining an llms.txt has to decide which pages genuinely matter. That decision pays off without the file too.
One rule holds hard though: what is in the file has to exist. A short form pointing at placeholder pages costs agents requests for nothing. Our checker therefore tests it on every build.
Related
What helps alongside it.
robots.txt per provider
An explicit statement per named service instead of one blanket entry. Costs nothing and removes ambiguity.
Structured data
The actual lever for attribution. Details under schema markup for AI.
Common questions
About llms.txt.
Where does the file have to sit?
In the root directory, so at yourdomain.at/llms.txt. A file in a subfolder does not get found.
How long should it be?
As short as possible. It is a selection rather than a directory of every page. Anyone writing everything into it has selected nothing again.
Is this the same as a sitemap?
No, and the two lists should be built separately. The sitemap lists everything indexable, the llms.txt the most important content in a selection and with descriptions.
Does it need to be submitted anywhere?
No, there is no submission. It simply sits there and gets fetched or not.
Is there an llms-full.txt as well?
The proposal exists and means a version with full text. We do not use it, because the same content is already available as Markdown per page.
Do search engines use the file?
There is no evidence for that. It addresses language models and agents rather than classic crawlers.
Can it do harm?
Only if it goes stale or points at pages that do not exist. That is why it should be generated automatically rather than maintained by hand.
Does a small website need this?
At ten pages the benefit is slim, because an agent gets through quickly anyway. From roughly thirty pages the orientation gets more interesting.
How do you generate it automatically?
From the same source as the navigation. Ours gets built from the page list, including the description from the front matter.
What does not belong in it?
Legal texts, thank-you pages, error pages and anything not yet written. A recommendation pointing at an empty page is worse than no recommendation.
Next step
Let us find out where you stand in AI answers.
One conversation, 30 minutes, no sales pressure. You leave with an honest read on whether this is worth the effort for you.
