Last month I deployed AI agent cards across 52 websites in a single weekend. Not because it was trendy. Because AI agents are already visiting these sites, and without an agent card, they have no structured way to understand what each site does, who runs it, or how to interact with it.
If you are still thinking of AI as a chatbot you paste questions into, you are behind. AI agents — autonomous software that browses, evaluates, and takes action on the web without human intervention — are already the fastest-growing category of web traffic. And they need a way to discover what your site offers.
That is exactly what an AI agent card provides.
What Is an AI Agent Card?
An AI agent card is a JSON file placed at /.well-known/agent-card.json on your website. It follows an emerging standard for machine-readable site identity, giving AI agents a structured description of your site, your services, your contact information, and your interaction preferences.
Think of it as a business card for machines. When an AI agent lands on your domain, it checks /.well-known/agent-card.json the same way a browser checks robots.txt. Except instead of telling the agent what to avoid, the agent card tells it what to engage with.
A basic agent card includes:
- Name and description of the site or business
- Capabilities — what services or content the site provides
- Contact information — how to reach a human
- Interaction preferences — whether the site accepts automated queries, API calls, or tool integrations
- Authentication requirements — whether an API key or OAuth flow is needed
Why Discovery Matters Now
Google's AI Overviews, ChatGPT with browsing, Perplexity, and dozens of specialized AI agents are all crawling the web to gather information, compare products, and answer user questions. When these agents visit your site, they are making decisions about whether to cite you, recommend you, or skip you entirely.
Without an agent card, these decisions are based entirely on your HTML content and whatever the agent can infer. With an agent card, you are explicitly telling the agent: here is who I am, here is what I do, here is how you can work with me.
The difference matters. In testing across our 52-site network, sites with agent cards saw a measurable increase in AI citation frequency within the first 30 days. The agents are not ignoring these files. They are reading them.
What AI Agents Actually Exist in 2026?
This is not theoretical. Here are the agents actively crawling the web right now:
- ChatGPT Browse — OpenAI's browsing agent that retrieves live web data
- Google AI Overviews — Google's AI-generated search summaries
- Perplexity — the AI search engine that cites sources
- Claude with tool use — Anthropic's agent framework
- Microsoft Copilot — browsing and summarizing web content
- Specialized shopping agents — comparing products and prices across sites
- Research agents — academic and market research tools that aggregate information
Each of these agents benefits from a standardized way to understand your site. The agent card provides it.
How We Deployed Agent Cards Across 52 Sites
Our network runs on a monoclone architecture — one codebase, one configuration file, 52 branded websites. This made deployment straightforward, but the approach works for any site.
Step 1: Create the JSON file. Here is a minimal agent card:
{
"schema_version": "1.0",
"name": "Your Site Name",
"description": "What your site does in 1-2 sentences",
"url": "https://yoursite.com",
"capabilities": [
"content_discovery",
"product_information",
"pricing_data"
],
"contact": {
"email": "hello@yoursite.com",
"human_url": "https://yoursite.com/contact"
},
"interaction": {
"accepts_automated_queries": true,
"rate_limit": "60/minute",
"preferred_format": "application/json"
}
}
Step 2: Place it in your .well-known directory. The file must be accessible at https://yoursite.com/.well-known/agent-card.json. On most static site generators, create a .well-known directory in your source folder and add passthrough copy in your build config. On WordPress, use a plugin or add a rewrite rule. On Cloudflare Pages, add it to your public directory.
Step 3: Validate the file. Hit the URL in your browser. Confirm it returns valid JSON with the correct Content-Type header (application/json). Use a JSON validator to catch syntax errors.
Step 4: Reference it in your HTML. Add a <link> tag in your <head>:
<link rel="agent-card" href="/.well-known/agent-card.json" />
This is not strictly required, but it provides an additional discovery mechanism for agents that parse HTML before checking .well-known.
Common Mistakes to Avoid
Forgetting CORS headers. If an AI agent is making a cross-origin request to your agent card, it needs appropriate CORS headers. Add Access-Control-Allow-Origin: * for the .well-known directory.
Using the wrong Content-Type. The file must be served as application/json, not text/html. Most static hosts handle this correctly for .json files, but verify.
Making the description too vague. "We help people" tells an agent nothing. "Personal finance education covering W-2 income optimization, real estate analysis, and digital business startup guides" tells the agent exactly what topics you are an authority on.
Ignoring the capabilities array. This is where you tell agents what they can actually do with your site. If you have a calculator, say so. If you have an API, say so. If you publish original research, say so.
What This Means for Your Next Site Audit
If you run any website that you want AI agents to discover, recommend, or cite, an agent card should be on your audit checklist. The implementation takes less than 15 minutes per site. The payoff is that every AI agent visiting your domain gets a structured understanding of who you are and what you offer.
We implemented this across 52 sites as part of a broader technical SEO upgrade that also included SpeakableSpecification schema, ItemList carousel markup, and server-side analytics. Each of these upgrades targets a different discovery channel — agent cards handle the AI agent channel specifically.
The web is fragmenting into human traffic and machine traffic. The sites that provide machine-readable identity will get discovered. The sites that do not will get skipped.
This strategy is covered in more depth in The $100 Network — including the monoclone architecture that makes deploying changes across 50+ sites a single-config operation. Buy The $100 Network on Amazon.