AI visibility tracker: check if ChatGPT names your brand

An AI visibility tracker answers one question: when a buyer asks ChatGPT, Perplexity or Gemini what to use, does the answer name you? Rank trackers cannot tell you. Google positions and AI answers are different lists, and an AI answer can cite a page that ranks nowhere.

You can build a simple AI visibility tracker with an agent. Write the prompts your buyers ask, run each one through the engines that matter, and record three things per answer: whether your brand appears, which brands appear instead, and which pages the engine cited. This post shows the steps, a real run, what each engine costs through looot, and where the method stops.

Last checked 2026-10-16, against looot's live catalog and one real ChatGPT run.

What does an AI visibility tracker measure?

An AI visibility tracker measures whether AI answers mention a brand for a set of prompts, and which sources they cite. The useful fields per answer are the answer text, the brands named, your position among them, and the cited URLs. Cited URLs matter most, because they are the pages you can change or earn a link from.

A real AI visibility check, step by step

On 2026-10-16 we ran one prompt through ChatGPT from the United States with looot's `cloro-monitor-chatgpt` endpoint: "What is the best web search MCP server for Claude Code?". The run took about 36 seconds and cost less than one cent.

The answer named four products in its entities list and listed 20 sources, 8 of them shown as citation pills. looot was not named. Most cited pages were the vendors' own pages and repositories, plus a few comparison posts. That tells us where to work: the comparison posts ChatGPT reads, and pages of our own that answer the question.

{
  "entities": [
    { "name": "Exa MCP", "type": "software" },
    { "name": "Tavily MCP", "type": "software" },
    { "name": "Brave Search MCP", "type": "software" },
    { "name": "Firecrawl MCP", "type": "software" }
  ],
  "sources": [
    { "url": "https://exa.ai/mcp", "position": 1 },
    { "url": "https://github.com/exa-labs/exa-mcp-server", "position": 2 }
  ]
}

To repeat it for your own brand:

1. Write 10 to 20 prompts the way buyers ask them, as full questions. "best crm for a 5 person agency" gets a different answer from "crm".
2. Pick the engines your buyers use and one country.
3. Run each prompt through each engine and save the raw answer.
4. Mark each answer: mentioned or not, position, brands named, cited URLs.
5. Run the same prompts again on another day and compare. Answers change between runs, so one run is a sample, not a score. Running it on a timer needs scheduled runs, not available yet, so today you start the rerun yourself.

Which AI engines can you check, and what does each cost?

looot's catalog has endpoints for ChatGPT, Perplexity, Gemini, Copilot, Google AI Overview and Google AI Mode. Some drive the engine's live interface, others use a scraper, so the same prompt can come back with a different answer on each. Prices are read from the catalog when this page loads:

Twenty prompts through four engines is 80 runs. Check the price of one run with `inspect` before a batch, and ask for approval above 100 calls. The AI search visibility page lists every endpoint for this job with its inputs and outputs.

How do you get mentioned more often?

Look at the cited sources first. If an engine keeps citing the same three comparison pages, those pages decide the answer. Your options are to be on them, to publish a page that answers the prompt better, or to make your own docs easy for crawlers to read. The generative engine optimization guide covers the page side. A plain `llms.txt` file is cheap to add, and the free llms.txt generator writes one. For the classic SEO checks that sit next to this, see what an AI SEO agent can run.

What it cannot do

• It does not run on a schedule or send alerts. Repeat runs need scheduled runs, not available yet.
• It runs without a buyer's account history, so an answer may differ from what a signed-in buyer sees. Treat it as a baseline.
• One run is one sample. The same prompt can name different brands an hour later, so judge trends over several runs.
• It does not keep a dashboard. Save the results to a sheet or file yourself, or ask the agent to write them.
• Each engine refuses some countries. The catalog has a countries endpoint to check first.