web dev//SEO//AEO//llms.txt

`llms.txt` is a proposed standard for a plain text file, served at `/llms.txt`, that gives LLMs a concise, structured summary of a website (who runs it, what it contains and where to find things), so that AI systems reading the site can understand it and cite it. A typical file holds the site name, a description of the author, the site structure (sections with URLs) and article listings with one-line summaries.

llms.txt is a proposed standard for a plain text file, served at /llms.txt, that gives LLMs a concise, structured summary of a website (who runs it, what it contains and where to find things), so that AI systems reading the site can understand it and cite it. A typical file holds the site name, a description of the author, the site structure (sections with URLs) and article listings with one-line summaries.

It sits beside robots.txt, which tells crawlers what to access, and the sitemap, which tells them what exists; llms.txt tells LLMs what the content means. Its message amounts to a version of the site without noise, offered to any AI that reads it. It is written for the web agent and the answer engine rather than the crawler, and it replaces neither the sitemap nor robots.txt, which still say what exists and what may be fetched.

It comes in two variants. llms.txt is a curated, hand-written summary, concise and editorial, like a human introducing their site to an AI; llms-full.txt is auto-generated and exhaustive, every article in full plain text, a machine dump for comprehensive indexing.

The file is kept ASCII-safe, because Unicode characters (em dashes, smart quotes) can corrupt under different encodings: -- stands in for the em dash and straight quotes for curly ones.

It was proposed by Jeremy Howard in September 2024 and is not an official standard (as of early 2026), but it is gaining adoption. The idea is that a site made trivially easy for an LLM to understand is more likely to be cited.

Text hidden from humans to steer a model is the opposite move (spam).