File structure: heading, summary, sections and link correctness.
What the validator checks
The format is simple, which is exactly why mistakes in it slip by.
Heading and summaryExactly one first-level heading at the top, with a summary blockquote right beneath it.
Section structureLinks should sit inside sections. Those before the first heading parse less reliably.
Absolute URLsThe file is read from outside, so relative links have nothing to resolve against.
Duplicates and foreign domainsThe same URL twice, or a link to another site. Not a format error, but a hint to revisit the list.
SizeIt should stay an overview. Hundreds of links turn it into a second sitemap and the point is lost.
Mistakes that show up
Not in the rootIt is looked for at /llms.txt. A file in a subfolder or with another name is simply not found.
Served as HTMLThe server returns an error page instead of text. Technically a response, practically no file.
A copied sitemapA thousand links with no notes. Not what the format was for, and of little use.
Never updatedWritten a year ago, half the links now point at deleted pages. A shorter but current file is better.
FAQ
Where should the file live?
In the site root at /llms.txt. A file in a subfolder or with another name will not be found — it is looked up at a fixed path, like robots.txt. Large sites sometimes add llms-full.txt with a longer description.
How many links should it contain?
A few dozen. It should stay an overview: its job is to show structure, not list everything. Hundreds of links turn it into a second sitemap, which is a different tool for a different purpose.
Are sections required?
The proposal says yes: links are grouped under second-level headings. Technically a file without them still parses, but the point is lost — a flat list is no better than a sitemap.
