It describes you from guesswork
With nothing authoritative to read, a model assembles your business from scraped pages, stale directories and whatever a competitor wrote about you.
An open, freely implementable standard
AI is already describing your business.
These files decide what it says.
AI Discovery Files are small text files you publish at your domain root —
next to robots.txt. They give ChatGPT, Claude, Perplexity and
every crawler behind them a first-party account of who you are, what you
offer, and how you may be represented.
Why this exists
Nobody asked you before answering for you
Every day, an AI assistant tells someone what your company does. If you have not published anything it can read directly, it fills the gap itself — and it does not tell the user it was guessing.
It gets your name wrong
Abbreviations you never use, a merged identity with a similarly named company, a founder quoted saying something they never said.
It answers on your behalf
Services you dropped, markets you left, prices from two years ago — delivered confidently, with no link back for the user to check.
The standard
Eleven files, in three layers
Layer 1 establishes who you are. Layer 2 sets the rules. Layer 3 adds the detail. Start with the Required files and add the rest as they earn their place.
Core Identity
Establishes who you are and makes it readable by both machines and people. Start here — everything else refers back to it.
- llms.txt Required Your identity, in the format AI systems read first.
- llm.txt Required A compatibility copy for systems that ask for the singular name.
- llms-full.txt Conditional The expanded companion, when one file is not enough.
- llms.html Recommended The same information, readable by people.
- identity.json Required Your identity as structured, parseable data.
- ai.json Recommended Structured identity plus recommendation signals.
Guidance & Control
Sets the rules: when to recommend you, how to name you, and how AI crawlers should treat your content.
Enhancement
Adds verified answers and technical context, so AI systems have something specific to quote instead of inferring.
llms.txt and identity.json are Required — those two
carry your identity. Everything else refines how that identity is used.
Implementation
An afternoon, not a project
There is nothing to install and nothing to sign up for. The files are static text, and you own every one of them.
- 01
Write the two Required files
llms.txt says who you are in prose. identity.json says the same in structured data. Both start from a template here, and both are plain text.
Open the templates - 02
Upload them to your domain root
They sit alongside robots.txt, at example.com/llms.txt and example.com/identity.json. No build step, no plugin, no JavaScript.
Setup for your platform - 03
Add the rest as they earn their place
Layer 2 sets the rules AI should follow. Layer 3 adds verified answers and technical context. Publish what applies to you and leave the rest.
Read the specifications
Who publishes these
Anyone who would rather be quoted than paraphrased
Business owners
You want AI assistants to describe your company the way you would describe it yourself — correct name, correct services, correct boundaries.
Marketing and brand teams
You already control how the brand appears in print and on screen. brand.txt and faq-ai.txt extend that control to the answers people are given.
Developers and agencies
Ten static files, a published JSON Schema, and validation you can run in CI. Roll it out across every client site from one template set.
In Practice
The part the specifications leave to you
A valid file served with the wrong Content-Type is an unreadable file. Guides for the things that decide whether any of this actually works.
Publishing on your platform
Where the files go, the Content-Type that trips people up, and the llm.txt redirect on nine platforms.
The AI crawler table
Which bot trains, which bot searches, and what you actually lose by blocking each one.
Beyond the files
Preferred sources, Organization markup, IndexNow — the other levers on how AI sees you.
Open
Specifications, templates and examples — all public
Every file definition, a fill-in template for each, worked examples and a validation script live in one MIT-licensed repository. Nothing here is gated, and nothing you generate from it is encumbered.
llms.txt follows the format published by Jeremy Howard at
llmstxt.org. This project follows that format rather than redefining it.
Start with two files
llms.txt and identity.json. Copy the template,
replace the bracketed placeholders, upload to your domain root. That is the
Required set, done.