llms.txt is a proposed Markdown file (often at /llms.txt) that hands large language models a short, curated map of your site: who you are, and links to the pages that actually matter. Jeremy Howard published the idea in 2024 (llmstxt.org); a 2026 revision added clearer ways for agents to find the file and Markdown versions of pages.
It is not robots.txt. Robots.txt asks crawlers what they may fetch. llms.txt is a table of contents for when an agent needs context — closer to a guided sitemap than a lock. Nobody has to read it. It does not block an AI crawler, change Google rankings, or replace structured data. For AEO, it is optional hygiene: a clean list beats making a model chew navigation HTML.
A real-life example of llms.txt
A Midtown law firm puts /llms.txt at the root: name, a one-line summary, then links to practice pages and a fee guide. An assistant answering “typical retainer for a Decatur closing” can follow those links instead of scraping the mega-menu. GPTBot can still ignore the file. If the firm wants a bot out, that rule lives in robots.txt — and enforcement still lives at the CDN or WAF.
Publish a short, honest list. Then decide crawlers bot by bot. For being the answer, see AEO and GEO: the new SEO; for the sign on the door, On-Site SEO Guide.