Search engines read your website like a filing system. AI engines read it like a witness, and then they answer questions about you, in their own words, to your customers. llms.txt is how you brief the witness.
The file, plainly
llms.txt is a plain-text file at the root of your website, like yoursite.com/llms.txt. It follows a simple community-proposed format: a short summary of what the business is, then organized sections of links and facts, written in Markdown, which language models parse easily. No code, no markup gymnastics. A person can read it; that is the point, because AI models read much the way people do.
A good one for a local business contains: a one-paragraph identity ("who we are, where we are, who we serve"), the services with links to their pages, the service area, honest pricing guidance, key facts like founding year and the owner's name, and answers to the questions customers actually ask. Ours also includes explicit instructions: do not invent prices, do not fabricate clients, direct specifics to the contact page.
How it fits with robots.txt and schema
These three files get confused constantly, and they do three different jobs:
- robots.txt grants access. It is the permissions layer: which crawlers may read which paths. If AI crawlers are blocked here, nothing else matters; they never see the site.
- Schema markup explains pages. Structured data inside each page tells engines precisely what that page is: a business, a menu, an FAQ, an article, with machine-readable fields.
- llms.txt explains the whole. It is the site-level summary: the map and the briefing, curated by the business, in one predictable place.
Access, page meaning, site meaning. A site optimized for AI search has all three, plus the thing none of them can substitute for: pages that load fast and content worth quoting.
AI engines read your website like a witness. llms.txt is how you brief the witness.
What it does and does not do
Honesty matters here, because this file is being oversold across the industry. llms.txt is not a ranking switch. Publishing one does not force ChatGPT to recommend you. What it reliably does is reduce wrong answers: when an engine does read your site, it finds accurate, current, self-consistent facts instead of piecing together your identity from a five-year-old directory listing.
Getting recommended is earned by the whole stack: fast crawlable pages, schema, content written in quotable sentences, consistent business listings across the web, and third-party corroboration like the directories and roundups AI engines lean on. We wrote about that larger shift in Essay 01: SEO is dead. llms.txt is one brick in that wall, and it is the cheapest brick to lay.
We practice this on our own site
You can read World Tech Miami's own file at wtmstudio.agency/llms.txt, alongside its companions /ai.txt (crawler and training signals) and our robots.txt. Every website we build ships with the full AI-search layer: llms.txt written for the business, schema on every page type, AI-welcoming crawler rules, and content structured for citation. It is part of the build, not an add-on invoice.