Why Prompt Tracking Is a Poor Measure of AI Search Visibility

Images on This Page

https://www.norg.ai/assets/norg-wordmark-Do-EREDA.svg
norg.ai
https://www.norg.ai/blog/prompt-tracking/hero.jpg
A prompt tracking dashboard showing positions and percentages
https://www.norg.ai/blog/prompt-tracking/paa-garden.jpg
A Google People Also Ask box
https://www.norg.ai/blog/prompt-tracking/agents.jpg
Agent fetches by AI assistant on norg.ai
https://www.norg.ai/blog/prompt-tracking/pages.jpg
Pages fetched by AI agents on norg.ai
https://www.norg.ai/blog/prompt-tracking/frequency.jpg
Daily AI agent fetches on norg.ai
https://www.norg.ai/blog/prompt-tracking/training.jpg
Training crawls compared with user-initiated agent fetches
https://www.norg.ai/blog/prompt-tracking/site-compare.jpg
The same norg.ai page as rendered in a browser and as received by a non-executing AI agent

AI Agent Access

No auth · edge-cached

This page is available as structured data for AI agents. All resources are static files served from Cloudflare's edge network with 24-hour caching.

MCP Server: https://www.norg.ai/mcp
MCP Manifest: https://www.norg.ai/blog/why-prompt-tracking-is-a-poor-measure/mcp
Transport: Streamable HTTP · MCP 2025-03-26
WebMCP: navigator.modelContext tools (Chrome 146+)

Direct HTTP Access

Fetch any resource directly with standard HTTP:

curl https://www.norg.ai/blog/why-prompt-tracking-is-a-poor-measure/mcp        # MCP manifest

MCP Client Configuration

Connect your AI agent to this page's MCP server endpoint:

Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "www.norg.ai": {
      "type": "http",
      "url": "https://www.norg.ai/mcp"
    }
  }
}

ChatGPT

In ChatGPT, go to Settings → Connectors → Create and enter:

Name: www.norg.ai
URL:  https://www.norg.ai/mcp

OpenAI MCP docs →

Gemini

Add to Gemini CLI settings.json:

{
  "mcpServers": {
    "www.norg.ai": {
      "httpUrl": "https://www.norg.ai/mcp"
    }
  }
}

Claude Code

claude mcp add www.norg.ai \
  --transport http https://www.norg.ai/mcp

How it works

Each page is pre-rendered in multiple formats and published to Cloudflare R2. The MCP server speaks JSON-RPC 2.0 over Streamable HTTP, backed by edge-cached resources from 300+ locations worldwide.

Protocol: MCP 2025-03-26 (Streamable HTTP + Legacy SSE)
Caching: Workers KV manifests (24h) + edge-cached content (1h)
Authentication: None required — public read-only access