Ask ChatGPT the questions your buyers ask, run each one several times, and count how many answers name you. Everyone samples. OpenAI gives you no report of where you were named, and the utm_source=chatgpt.com tag on referral links only shows the clicks that arrived. Track the share of runs that name you, week over week.
Why one check is not enough
ChatGPT rarely gives the same answer twice. In one test, 1,000 identical requests at temperature 0 returned 80 different completions. Day to day, the brands named overlapped at a Jaccard similarity of 0.45 to 0.59. A single "it mentioned us" proves little, and a single "it didn't" proves less.
What you test also matters. Calling the API is not the same as asking in the app, which decides for itself when to search the web. In February 2026 it searched on 34.5% of queries.
How to pick the questions
Start from what a buyer types before they know your name: which tool does a job, how to solve a problem, what to use instead of a product they already have. Write each question at least three ways and run every version at least three times. Count every run, including answers that named no brand at all. Showing up in 20 of 100 answers is 20% visibility.
Ways to track it
- By hand: a spreadsheet of questions, checked on a schedule. Free and slow, and fine for a short list.
- Free one-off checks: HubSpot AI Search Grader is free, needs no account, and checks an OpenAI model, Perplexity and Gemini in under two minutes. The Ahrefs AI Visibility Checker queries ChatGPT, Gemini, Perplexity, Copilot and Google AI Overviews with no signup. Both give you a first look.
- Paid trackers that sample on a schedule: Rankscale starts from $20/mo. Otterly.ai's $29/mo Lite tier has 15 prompts and daily checks. Profound starts at $99/mo for ChatGPT in one region and one language. At the top end, Evertune costs $800/mo and samples each prompt 100 times per model across 11 models.
- Inside your coding agent: if you build with Claude Code or Codex, Shipfound asks each tracked question 4 times on ChatGPT, Perplexity, Google AI and Claude, and labels the result, for example "named in 1 of 4 runs". It also ships the pages and fixes as pull requests. If you do not use either agent, one of the tools above fits better.
Frequently asked questions
How do I track brand mentions in ChatGPT?
Run your buyers' questions repeatedly and count how often you are named, by hand or with a tracker. Then watch referral visits tagged utm_source=chatgpt.com. Those are the clicks that came from an answer.
How do I get ChatGPT to recommend my product?
Be on the pages it cites. In an analysis of about 35,000 ChatGPT citation URLs, 17.1% of cited domains were UGC platforms and 11.1% were review sites. Markup alone is a weak lever: Ahrefs measured minus 4.6% on AI Overview citations after 1,885 pages added JSON-LD.
Does blocking GPTBot hide me from ChatGPT?
Not on its own. OpenAI's crawlers are independent, so you can allow OAI-SearchBot for visibility while disallowing GPTBot for training. To let the right ones in, see how to get indexed by PerplexityBot and GPTBot. To see which of them reach your pages, read how to tell if AI crawlers are hitting your site.