In one sentence: llms.txt is a proposed markdown file at your site root that gives language models a curated map of your content, and it is a convention rather than a standard that anyone has committed to honour.
Start with the caveat
No major AI vendor publicly guarantees that it reads llms.txt. Anyone who tells you it makes a brand visible is selling something. The reason to write one is different and still sound: it costs about an hour, it forces you to decide which pages are canonical, and if it is read it can only help.
Treat it the way you treat a sitemap. Useful infrastructure, not a growth lever.
What the file contains
The convention asks for markdown at /llms.txt with a specific shape: an H1 with the site or product name, a blockquote summarising what it is in a few sentences, then sections of links written as markdown links with a short description after each one. Plain URLs without link text defeat the purpose, because the description is the part that carries meaning.
The summary is the highest-value part of the file and the part most often written carelessly. It should say what the product is, who it is for, and what makes its scope specific, in language a stranger would use rather than language your marketing site uses.
A structure that holds up
Group links by the decision a reader is making, not by your navigation. Product pages, then solutions, then pricing, then documentation or guides, then company and legal. Put definitions of your own measurement terms in the file if you use any, because that is exactly the kind of ambiguity a model would otherwise resolve by guessing.
Generate the file from your routes rather than writing it by hand. A hand-written llms.txt is accurate on the day it is written and misleading three months later, and a file that points at pages you deleted is worse than none.
Checking it
Validate four things: the file exists at the root and returns markdown rather than an HTML error page, it begins with an H1, its links are grouped into sections, and it uses markdown links instead of bare URLs. Then check that every page it points at is itself crawlable. A file pointing at blocked or client-rendered pages helps nobody, and this is the failure that is easiest to miss.
Where the effort actually pays
If you have an hour for machine readability, spend it in this order: confirm the retrieval agents can fetch your content pages at all, make sure your key answers exist in raw HTML, publish accurate structured data for your organisation and products, and then write llms.txt. That order reflects how much each step can change an answer, from most to least.