LLM-LD — The Open Standard for AI-Readable Websites

The short version

LLM-LD (Large Language Model Linked Data) is a specification for structuring your website's information so that AI systems and agents can understand your business the way a human would.

This includes AI assistants like ChatGPT, Claude, and Perplexity — but also AI agents like Manus and OpenClaw that browse the web, take actions, and make purchases on behalf of users.

It answers the question: When someone asks an AI "What's the best dentist near me?" — or sends an agent to "Book me an appointment with a highly-rated dentist" — how does your business show up?

Your website is how humans understand your business.
Your LLM-LD implementation is how AI and agents understand it.

Why does this matter?

Search is changing. People are increasingly asking AI assistants for recommendations instead of typing keywords into Google. And it's going further — AI agents are starting to browse the web, compare options, and take actions on behalf of users.

When an AI assistant or agent visits your site, it doesn't see your beautifully designed website. It sees... HTML. Maybe some metadata. Usually a confusing mess of navigation, footers, cookie banners, and buried content.

Traditional SEO optimized for search engine crawlers that index keywords and links. But AI systems and agents work differently — they need structured context to understand what you do, who you serve, and what action to take.

LLM-LD gives them that structure.

Without LLM-LD

😕 AI guesses what you do from HTML fragments
🔍 Important details buried in page noise
❓ No clear action for AI to recommend
🤖 Agents can't reliably interact with your site
🏝️ Isolated — AI has to find you by luck

With LLM-LD

✅ AI reads structured data purpose-built for it
📋 Key facts, services, differentiators — clear
🎯 Explicit CTAs: book, buy, contact, learn
🤖 Agents know exactly how to take action
🔗 Connected to discovery network

How it works

LLM-LD uses a layered architecture — three layers that build on each other, plus a discovery mechanism that connects you to a network of AI-readable sites. Each layer gives AI systems and agents more context to work with.

1️⃣

Layer 1: Schema.org JSON-LD

Standard structured data on your pages. Works today, uses existing formats every crawler understands. The foundation.

2️⃣

Layer 2: Entities + Knowledge Graph

Define your people, products, services, locations — and how they relate. Gives AI rich context, not just facts.

3️⃣

Layer 3: llm-index.json

A single file with everything AI needs. Your website is for humans — this is a website for AI and agents.

🔗

AI Discovery Page + Network

A page that links your data to the LLM Disco Network. When AI or agents find one site, they discover all.

📖

Go deeper: The Three Layer Architecture guide explains each layer in detail — what's in it, why it matters, and how they build on each other.

What makes LLM-LD different?

There are other approaches to "AI SEO" — prompt injection in meta tags, hoping AI reads your blog posts, or just doing nothing and hoping for the best. LLM-LD is different:

It's a standard, not a hack. Built on Schema.org and JSON-LD — formats that have been around for years and are supported by every major platform.
It works today. You don't need to wait for AI platforms to adopt anything. Layer 1 makes your site more AI-readable the moment you implement it.
It's agent-ready. As AI agents become more common, LLM-LD gives them the structured actions they need to book, buy, and interact on behalf of users.
It's connected. The LLM Disco Network means your site gets discovered when any site in the network gets crawled. Rising tide lifts all boats.
It's open and free. CC BY 4.0 license. No fees, no gatekeepers, no approval process. Implement the spec, join the network.

Who is it for?

🏪

Local Businesses

Dentists, restaurants, contractors, retailers — anyone who wants to show up when people ask AI for local recommendations.

💻

SaaS & Software

Products that want to be recommended when users ask AI for tool suggestions in their category.

🛒

E-commerce

Stores that want AI to understand their products, inventory, and what makes them different.

If someone might ask an AI about your industry — or send an agent to take action on your site — LLM-LD helps you be the answer.

What's the catch?

There isn't one. But let's be honest about what LLM-LD is and isn't:

It's not magic. Implementing LLM-LD doesn't guarantee AI will recommend you or that agents will choose you. It makes sure they can understand you — what they do with that is up to them.
It requires work. You need to create structured data, maintain it, and keep it accurate. Garbage in, garbage out.
It's early. The AI and agent landscape is evolving fast. LLM-LD is built to be future-proof, but no one knows exactly where this is all going.

That said — doing something is better than doing nothing. And doing something structured is better than hoping AI and agents figure it out from your HTML.

⏱️

Time investment: A basic implementation (Layer 1 + AI Discovery Page) takes about 15-20 minutes. That's enough to join the network and start being AI-readable. You can add depth over time.

The name

LLM-LD stands for Large Language Model Linked Data. It's a nod to JSON-LD (JavaScript Object Notation for Linked Data), the format that Schema.org uses. JSON-LD was built for search engines. LLM-LD is built for AI systems and agents.

Who's behind it?

LLM-LD is created and maintained by CAPXEL, an AI data company. It grew out of real implementations for real clients — not theory, but practice. We open-sourced it because a standard only works if people use it, and people only use standards that are open.

The spec is licensed under CC BY 4.0. Use it, modify it, build on it. Just give credit.

What is LLM-LD?