AI Rank Tracker API

How often the answer actually recommends them

shortlist_rate is the share of answered runs whose prose put the business on its shortlist. This is the number a client is asking for when they ask whether AI recommends them. appearance_rate sits alongside it and counts every run that named them at all, recommendation or passing mention — the two say different things, and which one is low is the actionable part.

Where in the argument they land

mean_shortlist_rank and best_shortlist_rank average only the runs that listed the business, so being named third in 3 of 10 answers reads as rank 3 at a rate of 0.3 rather than being flattened into one misleading score. Every entry also carries the answer's own sentence in mention, so a report can quote what the AI actually said.

Which source the model reached for

cited_domains lists the domains the answers cited for each business, most-used first, with a run count. A model sourcing a client from their BBB listing rather than their own site is telling you which page it treats as authoritative for that query — and that is a fixable problem. This is the one output here that points at work rather than just measuring it.

Everyone named in the same answers

Every run returns every business it named, so one sample gives you the client's rate and the rate of everyone beating them, measured on identical prompts at the identical moment. No separate competitive call, no mismatched timing to explain away.

A refusal is not an absence

Rates are out of the runs that actually answered. A run the provider lost is refunded and excluded; a run where the model declined the prompt is counted in samples_refused and excluded too, because “I can't help with that” is not evidence a business is invisible. samples_unreadable reports answers no shortlist could be read from.

One business, every spelling it was given

Answers spell the same business several ways — with and without LLC, with & or the word and. Those fold into one row, and name_variants records every spelling that went into it, so a merge is something you can check rather than trust. reader_version stamps which version of the prose reader produced the shortlist, so you never compare two runs that were read by different code.

Direct MCP integration

Drop-in support in Claude Desktop, OpenClaw, Hermes Agent, and any MCP-aware client.

Add to your client's MCP config (e.g. claude_desktop_config.json):

{
  "mcpServers": {
    "localseodata": {
      "url": "https://mcp.localseodata.com",
      "headers": {
        "Authorization": "Bearer sk_live_..."
      }
    }
  }
}

REST API

For Perplexity Computer, ChatGPT Custom GPTs, custom agents, and any platform that calls REST endpoints directly.

Base URL:

See the docs for endpoint reference and auth.