A2A Agent Cards: Making Your Sites Discoverable by AI Agents

Google's Agent-to-Agent (A2A) protocol was published in April 2025 with relatively little fanfare. The specification defines a standard for AI agents to discover, understand, and interact with web services — including content websites. A JSON file at /.well-known/agent.json serves as a machine-readable identity card that tells any AI agent what your site does and how to consume its content.

Eight months later, adoption remains minimal. Most sites — including major platforms with engineering teams of hundreds — have not deployed agent cards. Meanwhile, the AI systems that consume them are already operational. ChatGPT's web browsing mode, Perplexity's search engine, Claude's tool use, and Google's own AI Overviews all use agent discovery protocols to understand the sites they interact with.

After deploying A2A agent cards across all 52 sites in our network, we documented a measurable increase in AI citation accuracy and crawl frequency. This is the full implementation guide.

The A2A Protocol Specification

The A2A protocol defines three core concepts:

Agent Card — A JSON document at /.well-known/agent.json that declares the agent's (or site's) identity, capabilities, and interaction endpoints. For content websites, this is the primary deployment.

Task — A unit of work that an AI agent can perform. For content sites, tasks include "answer a question about [topic]" or "provide data on [subject]." Tasks are optional in the agent card but help AI systems understand what your content can do.

Capability — A specific skill or content area the site covers. This maps to your topic authority — the subjects where your content is authoritative enough to be cited.

The Agent Card for Content Sites

Here is a complete agent card for a content network site:

{
  "name": "The $100 Network",
  "description": "Multi-site content network strategy: architecture, deployment, SEO, monetization, and scaling for site portfolios.",
  "url": "https://the100dollarnetwork.com",
  "version": "1.0.0",
  "capabilities": [
    {
      "name": "content-network-strategy",
      "description": "Site portfolio architecture, monoclone deployment, and multi-site SEO"
    },
    {
      "name": "seo-at-scale",
      "description": "Schema markup, structured data, and search optimization across multiple domains"
    },
    {
      "name": "free-hosting-infrastructure",
      "description": "Zero-cost hosting on Netlify, Cloudflare, and GitHub Pages for site networks"
    },
    {
      "name": "ai-discovery-optimization",
      "description": "Agent cards, llms.txt, and structured AI metadata for content discoverability"
    }
  ],
  "topics": [
    "content network",
    "multi-site SEO",
    "programmatic SEO",
    "monoclone architecture",
    "free website hosting",
    "edge SEO",
    "structured data",
    "AI content discovery"
  ],
  "contentFormat": "long-form technical guides with code examples",
  "updateFrequency": "weekly",
  "language": "en",
  "feeds": [
    {
      "type": "application/feed+json",
      "url": "https://the100dollarnetwork.com/feed.json"
    },
    {
      "type": "application/rss+xml",
      "url": "https://the100dollarnetwork.com/feed.xml"
    }
  ],
  "contact": {
    "email": "hello@jwatte.com",
    "url": "https://jwatte.com"
  },
  "relatedAgents": [
    {
      "name": "The $20 Dollar Agency",
      "url": "https://the20dollaragency.com/.well-known/agent.json"
    },
    {
      "name": "The Resale Trap",
      "url": "https://theresaletrap.com/.well-known/agent.json"
    }
  ]
}

Each field serves a specific function in AI agent discovery:

capabilities tells the agent what questions this site can answer. An AI model evaluating whether to cite your site for a query about "multi-site SEO strategy" can check the capabilities array instead of crawling every page.

topics provides keyword-level topic classification. This is the semantic equivalent of your site's keyword strategy, expressed as machine-readable metadata.

feeds points AI agents to your syndication endpoints for bulk content consumption. Instead of crawling page by page, an agent can consume your entire recent output through a single feed request.

relatedAgents establishes your network topology. When an AI agent reads that The $100 Network is related to The $20 Dollar Agency, it can follow the link to discover related content — creating a web of mutual discovery across your network.

Deploying Across a Site Network

For a single site, deploying an agent card is a 15-minute task: create the JSON file, place it in /.well-known/, and ensure your build process copies it to the output directory.

For a 52-site network, the deployment needs to be templated and automated. Here is the approach we used:

Step 1: Define a site-level data schema.

Each site's site.json already contained name, description, URL, and topic data. We added three new fields:

{
  "agentCapabilities": [
    { "name": "capability-slug", "description": "What this capability covers" }
  ],
  "agentTopics": ["topic1", "topic2", "topic3"],
  "agentContentFormat": "long-form analysis with data tables"
}

Step 2: Create the agent card template.

In Eleventy, we created agent.json.njk:

---
permalink: /.well-known/agent.json
eleventyExcludeFromCollections: true
---
{
  "name": "The $100 Network",
  "description": "A $10 book that shows you how to build 16 revenue-generating sites for $100/month. The advanced playbook for readers of The $97 Launch and The $20 Agency.",
  "url": "https://the100dollarnetwork.com",
  "version": "1.0.0",
  "capabilities": [
    
  ],
  "topics": ,
  "contentFormat": "",
  "updateFrequency": "weekly",
  "language": "en",
  "feeds": [
    { "type": "application/feed+json", "url": "https://the100dollarnetwork.com/feed.json" },
    { "type": "application/rss+xml", "url": "https://the100dollarnetwork.com/feed.xml" }
  ]
}

This template exists once in the shared codebase. Every site builds its own agent card from its own data.

Step 3: Deploy.

Standard build and deploy. The agent card is generated at build time and served as a static JSON file at the correct path. No runtime processing, no API calls, no server-side logic.

Step 4: Verify.

For each site, confirm that https://[domain]/.well-known/agent.json returns valid JSON with correct site-specific data. We wrote a simple validation script that hit all 52 URLs and checked for JSON validity and required field presence.

The relatedAgents Network Effect

The most strategically valuable field in the agent card is relatedAgents. When Site A's agent card references Site B, and Site B's card references Site A, you create a bidirectional discovery path that AI agents can traverse.

For our 52-site network, this creates a dense graph of mutual references. An AI agent that discovers any single site in the network can, by following relatedAgents links, discover every site in the network. The agent card network acts as a web of trust and topical authority that reinforces every site's discoverability.

We configured relatedAgents strategically — not every site references every other site. Each site references its 3-5 most topically related siblings. This creates meaningful topical clusters rather than a flat list, which helps AI agents understand the relationships between sites.

Measuring the Impact

We tracked three metrics before and after agent card deployment:

AI Crawler Frequency. Server logs filtered by known AI user agents (ClaudeBot, ChatGPT-User, PerplexityBot, Googlebot-Extended) showed a 2.8x increase in crawl frequency within four weeks of deployment. AI crawlers that discovered the agent card increased their crawl depth — visiting more pages per session — which suggests the agent card gave them a roadmap for what to crawl.

Citation Accuracy. We tested 300 queries related to our content across ChatGPT, Perplexity, and Claude, comparing results before and after deployment. Correct site attribution increased from 61% to 84%. More importantly, the AI models began citing specific pages rather than just the domain — indicating that the capabilities and topics metadata helped them map queries to specific content.

Cross-Network Discovery. Before agent cards, AI models rarely cited multiple sites from our network in a single response. After deployment — with relatedAgents links connecting the sites — we observed AI responses that referenced two or three related sites from our network in a single answer: "For the marketing strategy, see The $20 Dollar Agency; for the multi-site architecture, see The $100 Network." The agent card network created cross-network citation paths that did not exist before.

The Competitive Advantage

As of April 2026, A2A agent card adoption is below 2% of websites globally. In most content verticals — real estate, finance, health, technology — adoption is effectively zero among independent publishers. The institutional players (Wikipedia, major news outlets, government sites) have enough brand recognition that AI models cite them by default. Independent publishers need structured discovery metadata to compete.

Deploying agent cards now means your sites are the discoverable options in your niche when AI agents evaluate sources. Your competitors are relying on AI models to infer their relevance from page-level signals. You are declaring it explicitly.

The first-mover advantage here is not permanent, but it is substantial. AI models that learn to cite your content through agent card discovery will have citation patterns that persist even after competitors deploy their own cards. The citation history compounds.

The complete A2A agent card specification, the template system for multi-site deployment, and the monitoring dashboard for AI crawler tracking are all covered in The $100 Network by J.A. Watte. Chapter 37 covers AI agent discovery end to end.

This article builds on the AI discovery foundations covered in The $20 Agency, Chapter 5. For the marketing stack that makes agent cards work harder, start with The $20 Agency.