llms.txt, explained: what it is, what goes in it, and a template you can ship today
llms.txt is a plain-text file at your site root that gives language-model agents a concise, curated guide to what your site is and what a visitor can do there. Think of it as the agent-era cousin of robots.txt: where robots.txt says where you may go, llms.txt says what's worth doing here.
What actually goes in it
The format is deliberately simple Markdown: an H1 with your site name, a one-line summary as a blockquote, then sections linking to your key pages and tasks. No spec lawyering required — the useful test is: could an agent that reads only this file route a user's intent to the right page?
The mistakes we see scanning real sites
Having scanned thousands of pages for agent directives, the failure modes repeat: the file exists but lists marketing pages instead of actionable ones (pricing, signup, checkout, docs); the paths are wrong or redirect; or — most common by far — the file simply doesn't exist while competitors' do. An llms.txt that lists paths that 404 is worse than none: agents follow those links.
A template to start from
# Your Company
> One sentence: what this site lets a visitor accomplish.
## Key pages
- Pricing: /pricing
- Start a free trial: /signup
- Product docs: /docs
- Contact sales: /contact
## Notes
- Verified agents: declare a User-Agent and follow /robots.txt.Two rules when adapting it: only list paths that exist (pull them from your real nav, not your aspirations), and keep it short — this is a routing hint, not a sitemap.
Does it matter if the standard is young?
Fair question. Standards for agent directives are still shaking out, and betting your roadmap on any one spec would be premature. But the cost of publishing a truthful llms.txt is an afternoon, agents that look for it reward it today, and the discipline of writing one — naming your site's core tasks in machine-legible form — is valuable regardless of which spec wins. Publish it, keep it honest, and let your scanner verify it stays that way.