An /llms.txt file is a plain-text markdown summary of your site, served at the root, designed for large language models to read. Think of it as robots.txt for meaning instead of permissions. Done right, it's the single highest-leverage SEO file you'll ship in 2026.
The format
llms.txt lives at https://yourdomain.com/llms.txt. It's markdown. It starts with an H1 (your site name) followed by a blockquote summary, then sections for About, Services, Pricing, Pages, and Contact. No HTML, no JavaScript, no images. Just structured text.
What to include
- A blockquote summary under the H1 — agents quote this verbatim
- Pricing in plain numbers — $2,400 not "starting at low four figures"
- Every public URL — agents won't infer them from a sitemap
- Contact + response time — turns citations into leads
- A content license line — explicit permission for AI systems to cite you
What to leave out
- Marketing fluff ("revolutionary," "synergy," "industry-leading")
- Internal jargon and product codenames
- Login URLs, admin paths, checkout success pages
- Anything that contradicts your actual website — agents notice
Common mistakes
1. Serving HTML instead of plain text — set Content-Type: text/plain or just ship a static file in /public 2. Linking to a sitemap.xml and calling it done — sitemap is for crawlers, llms.txt is for understanding 3. Forgetting to update it — stale pricing in llms.txt becomes wrong citations everywhere 4. Hiding it behind auth or a CDN that requires JS — agents fetch with simple HTTP; if curl can't get it, neither can they
How to test it
Ask ChatGPT: "Read https://yourdomain.com/llms.txt and summarize what this company does." If the summary matches your positioning, your file works. If it confabulates or pulls outdated info, fix the file.
The strategic point
llms.txt isn't a ranking trick. It's a contract between your brand and the agents that increasingly mediate how customers find you. Sites that ship a clean llms.txt today are training the corpus of how their category gets described tomorrow.