What it is
A text file, placed at the root of the domain, at /llms.txt. It introduces the site in natural language and points at the pages that carry authority, each with a sentence of description. The format is simple Markdown: a title, a summary as a blockquote, then sections of links.
The underlying idea is to save a model time. Rather than letting it guess a site's structure by crawling, you hand it a map: here is who we are, here are the six pages that count, here is what each one is for.
What is honestly known
The convention was proposed by the community, not by the search providers. As these lines are written, no major provider has publicly committed to reading the file, and no public study demonstrates a direct effect on citations.
This is a point a lot of commercial content glides over. Selling an llms.txt as a proven visibility lever would be dishonest. Presenting it as a low-cost bet is accurate.
What is measurable, on the other hand, is the side effect. Writing the file forces you to answer three questions on one page: what does this company do, on which subjects is it legitimate, which pages prove it. It is common for a site to have no reference page at all on its main subject — and the exercise reveals that in half an hour.
How it is written
The structure comes down to four elements, in this order.
# Company name > One to three sentences saying what the company does, > for whom, and in which territory. ## Services - [Service name](https://example.com/service): what it covers, in one sentence. - [Another service](https://example.com/other): likewise. ## Resources - [Article title](https://example.com/article): subject covered (updated 12 March 2026). ## Contact Email: contact@example.com Phone: +44 20 1234 5678
Three practical rules. The descriptions are written for a reader, not for a crawler: "what it covers", not a list of keywords. URLs are absolute. And the file stays short — twenty to forty links at most, beyond which it stops sorting anything.
Generate it rather than writing it by hand
A static file goes stale the day a page changes its title. Generated from the same source as the site — content database, CMS, Markdown files — it stays correct on its own. That is an hour of development, once.
Our own file works that way: skgenius.com/llms.txt is produced on the fly from the site's database. You can read it, there is nothing secret in it. The long version, llms-full.txt, adds the page bodies as plain text for a model that wants to ingest everything at once.
What the file does not do
It permits nothing. If your robots.txt blocks GPTBot or ClaudeBot, those crawlers will read neither your pages nor your llms.txt. Permission is settled first, and it is the point the audit checks before anything else.
Nor does it replace content. A file pointing at six hollow pages simply announces six hollow pages, more clearly than before.
So, should you publish one
Yes, provided you file it for what it is: an hour of work, a reasonable bet on a convention that may take hold, and a clarification exercise that has value even if the file is never read. What you should not do is write it before settling crawler access, entity markup and page structure — all three documented on the pillar page.