What tools help track brand mentions across ChatGPT, Perplexity, and Gemini?
To track brand mentions across ChatGPT, Perplexity and Gemini we use a fixed prompt set, a Brand Book and Promptwatch. This is the workflow and why it holds up.
Tracking brand mentions across ChatGPT, Perplexity and Gemini is a job we do every week for clients. The tooling matters less than a few habits, and most of the mistakes we see come from skipping those habits. Here is the workflow, with the tool we run it on.
Why a quick manual check misleads
Clients often show us a screenshot: "ChatGPT does not mention us." It may be true. It may also be one unlucky run, a personalized answer, or a question phrased in a way real buyers do not use. We have seen the same prompt give different brand lists on consecutive tries. So we never accept a single check as evidence in either direction.
The fix is volume and consistency. Run a fixed set of prompts, repeatedly, on a schedule, and read the pattern.
Our workflow
1. Build the prompt set
We write 30 to 50 prompts with the client. Category questions ("best X for Y"), comparison and alternatives questions, and a few including the brand name. We tag them by topic and intent. We then freeze the set for a month, because changing prompts mid-stream makes trends unreadable.
2. Define what counts as a mention
This is the step people skip. Does "Acme CRM" count when the answer says "Acme"? What about a misspelling, a parent brand, a product line? In Promptwatch we write these into the Brand Book, which holds aliases and analysis instructions that decide what counts. A tracker with loose definitions produces numbers that move for no reason.
3. Run it across engines
Promptwatch runs prompt panels daily per model, country and language. ChatGPT Search, Perplexity and Gemini are monitored through their real interfaces, together with Google AI Overviews, Google AI Mode, Copilot, Amazon Alexa and Meta AI. Models such as Claude, DeepSeek, Mistral and Grok are tracked via API, and we label them that way in client reports.
We report Visibility Score, Share of Voice, Position in Answer and Sentiment, as trends. The heatmap helps when a client asks "where are we weakest?" because it can compare brands against models or topics against countries.
4. Look at the sources
A mention tells you the result. Citations tell you the cause. Citation analytics show which pages and domains the engines used, including Reddit and YouTube reports. The offsite mentions list shows pages that name the brand without linking. When Perplexity cites a review site and Gemini cites something else, that difference usually points to where to work.
5. Check access
If a mention is missing, we rule out crawler access before blaming content. Agent Analytics shows IP-verified AI crawler logs, with errors and an explanation per failed request. Crawler logs start on Professional ($245 a month), so on an Essential ($95) client we check by hand first.
6. Report and act
The Action Board gives us a prioritized list from the data, and we edit it before anyone acts on it. For client reports we use PDF exports or Looker Studio.
Other tools clients bring up
We tell clients honestly what we know from vendor pages:
- HubSpot AEO costs $50 a month and tracks ChatGPT, Perplexity and Gemini, a neat match for this exact question, with 25 prompts daily. It is a sensible fit for a HubSpot shop on a small budget. It does not cover Google's AI features, and we found no published crawler logs.
- AthenaHQ has a free tier and Starter at $295 a month, billed in credits. Strong on tasks.
- Scrunch AI costs $250 a month on Core with four LLMs, ChatGPT, Perplexity, Google AI Overviews and Copilot, so Gemini is not in that list.
- Profound is an enterprise program on a custom plan.
A cheaper tool is fine for a first look at mentions. When the follow-up question is "why," or "what is it worth," the extra layers pay for themselves.
Common mistakes
Counting once. Single readings are noise.
Not defining mention rules. Aliases and product names swing results.
Mixing engines in one average. A brand can do well in one engine and badly in another. Look at each.
Confusing mention with recommendation. Sentiment and position tell you whether the mention helps.
Expecting a promise. No tool can guarantee that an engine will name a brand.
What we tell clients
Mentions are a signal, not a guarantee. Answers shift between runs and model updates. We commit to a clear baseline, honest sampling, and evidence for the next action. We do not commit to a number.
If you are starting alone
Spend an hour writing your prompts, a second hour defining mention rules, and then compare a couple of tools on that set for a month. You will quickly see which tool shows you more than a count.