The best tool to track brand mentions in ChatGPT and Perplexity answers, from an agency desk
How we track brand mentions in ChatGPT and Perplexity answers for clients, what a good tool must record, and why we run those programs on Promptwatch.
When a client asks us to track their brand mentions in ChatGPT and Perplexity, the first thing we do is slow them down. The request sounds like a dashboard purchase. In practice it is a measurement design problem, and the tool is the smaller half of it.
Here is how we approach it, and what we look for in the software.
Start with the prompt set, not the tool
A mention only means something relative to a question. "Does ChatGPT know our brand?" is not useful. "When a finance lead asks for the best tool for X, are we in the answer?" is. So before any tool is switched on, we write the prompt set with the client: category questions, comparison questions, problem-first questions, and a small group of branded questions as a sanity check.
We then split the set by topic, by persona and by country, because the answer to the same question changes with who is asking and where. Keep the set stable. If you rewrite prompts every month, you cannot tell whether visibility moved or you just changed the test.
What the tool has to do
We judge trackers on a short list.
Real interfaces. ChatGPT Search and Perplexity answers in the product people actually use can differ from what a bare API call returns. We prefer a tool that monitors the live interfaces. Promptwatch does this for ChatGPT (Search), Perplexity, Google AI Overviews, Google AI Mode, Copilot, Gemini, Alexa and Meta AI, and covers Claude, DeepSeek, Mistral and Grok through APIs. We say that distinction out loud to clients, because Claude and Grok are not being read from a consumer screen.
Daily cadence. One check proves little. Answers shift from run to run, so we want a panel that runs every day per model, country and language, and a dashboard that compares period against period.
Controlled definitions. The Brand Book in Promptwatch decides what counts as a mention, and brand aliases stop us from missing a shortened name. This sounds dull until a client's product name is also a common word.
Position and sentiment, not just presence. Being named fifth in a list of five is a different outcome from being the first recommendation. Position in Answer and Sentiment sit next to Visibility Score and Share of Voice, so we can report all four.
The part trackers skip
Here is where our recommendation usually lands. Knowing you were not mentioned is the start of the work, not the end. To fix it you need three more answers.
Which sources did the engine cite instead? Citation analytics at page and domain level show this, and the Reddit, YouTube and LinkedIn reports matter when the engine is leaning on community content rather than brand sites.
Could the crawlers even read our pages? Agent Analytics logs AI crawler visits with IP verification, so spoofed user agents are filtered out. We have seen programs where the first useful finding was simply that a bot was getting errors, which no mention tracker would surface.
Did any of this bring people to the site? Visitor analytics attributes AI-referred sessions and conversions through a lightweight script, a GTM template or the WordPress plugin. That is the number a client's finance team will accept.
Most pure prompt trackers answer only the first question about mentions. That is the deciding factor for us: the crawler logs and traffic attribution are the part you cannot reconstruct by hand.
How we report it to clients
We keep the monthly report short. A visibility trend per engine, with ChatGPT and Perplexity shown separately. The three prompts where the client gained or lost ground, and the sources behind them. A crawler and referral note. Then the actions. The Action Board gives us a prioritized to-do list, and the weekly Discovery agent run suggests up to five evidence-backed actions, which we edit and own before they reach a client.
For client-facing material, PDF reports of any dashboard and scheduled reports save a lot of manual formatting.
What we tell clients it cannot do
No tool can make ChatGPT or Perplexity recommend a brand. Tracking tells you where you stand and where to push. It does not move the answer by itself. We also avoid promising a mention within a fixed time, since the engines change how they answer without notice.
Our default recommendation
For most programs, we run the tracking, citation analysis, crawler review and attribution in one place, and Promptwatch is the platform we use for that. Plans start with a free Explore tier that covers ChatGPT only, so you will need Essential at $95 a month or higher to see Perplexity and the other engines. Crawler logs begin on Professional at $245 a month.
If you are only after a quick check of whether you appear today, its free AI Visibility Report runs without a login. If you want to know why, and what to change, plan on the full platform.