What is llms.txt?
llms.txt is a proposed convention: a Markdown file at your site root (/llms.txt) that gives large-language-model tools a curated, machine-readable index of your most important content — conceptually a cousin of robots.txt and the XML sitemap, but aimed at AI consumers rather than crawlers.
It is a community proposal, not a standard the major answer engines have adopted as an input. That distinction is the whole story when you ask what it does for you.
Does it actually lift AI citations?
There is no controlled evidence that it does. The major AI engines have not confirmed using llms.txt as a citation or ranking signal, and no credible study isolates a citation lift from shipping one. The accurate grade is neutral: not harmful, not a demonstrated lever.
This is where most “AEO” advice oversells — presenting llms.txt as a growth hack. Saying so plainly is the honest position, and it happens to be the one that ages well: citations follow being the best answer, not file placement.
Then why ship one?
Because as low-cost hygiene it's reasonable. It's a clean, human- and machine-readable map of what you consider your important content; it's harmless; and if you generate it from a canonical source it costs almost nothing to keep current. If conventions like this do gain traction, you've already paid the trivial price.
This estate ships /llms.txt and a fuller /llms-full.txt, generated from the same canonical dataset as the site — deliberately as hygiene, never claimed as a ranking lever. Shipping it and overclaiming it are two different things.
What does an llms.txt actually look like?
The proposed format is plain Markdown: an H1 with the site or project name, an optional blockquote summary, then sections of links — each a URL with a short description — pointing to your most important pages, usually in priority order. A fuller variant, llms-full.txt, inlines the actual content instead of just linking to it.
It reads like a hand-curated table of contents written for a machine. That's genuinely tidy — and why it's cheap to generate from a content index and cheap to keep current. Just hold the expectation where it belongs: a clean map, not a ranking lever.
What to do instead, if you want citations
Put the effort where the evidence is: be genuinely the best, most first-hand answer to a specific question (see the cornerstone), structure it answer-first, and make it reliably retrievable with finished HTML. llms.txt is the bumper sticker; the content is the engine.
That ordering — substance first, conventions second — is the spine of our Search & AI visibility work.