How Do I Know If My Brand Is Showing Up When People Ask ChatGPT or Perplexity About My Industry?

You find out by asking a fixed set of your buyers' real questions, across every engine they use, from clean sessions, with a buyer persona and location attached, repeatedly over weeks, and then counting two things: what share of those answers name you (your mention rate) and where you land in the list when they do (your average position). One manual check in your own ChatGPT window is not a measurement. It is a mood.

Most founders discover this the painful way. They type "best [category] tool for [use case]," watch three competitors get named, and conclude they are invisible. Sometimes that is true. Often the test itself was broken. Here are the mistakes that produce a false read, and what to do instead.

Mistake 1: Treating one prompt as a result

Language models are non-deterministic. Ask the identical question twice in the same minute and you can get two different brand lists, because sampling introduces variance by design. A single run tells you what happened once, not what your buyers see.

The fix is volume and repetition. Ten to fifty questions that your buyers actually type, re-asked on a standing schedule so you are reading a trend line instead of a coin flip. Clark, our AI search specialist, runs monitoring every day and rotates through your question pool, so any one question comes back around roughly weekly. That cadence is what turns noise into a number you can take to a board meeting.

Mistake 2: Testing from your own logged-in account

This is the quiet killer. Your ChatGPT account has memory, chat history, and whatever personalization has accumulated from months of you asking about your own company. The model has been nudged, in a small way, by you. It will mention you more readily than it would for a stranger in another country.

Test from clean, unpersonalized sessions only. And never put your own brand name or domain in the prompt you are measuring. If the question contains the answer, the mention rate you get back is a measurement of your typing, not your visibility. Our monitoring persona is automatically checked to make sure it never names the business or its website, for exactly this reason.

Mistake 3: Asking the question naked

"What CRM do you recommend?" and "What CRM do you recommend for a 10-person logistics startup in Austin that just raised a seed round?" return different companies. Framing moves the answer more than almost anything else, because the model is trying to serve the person in front of it.

Real buyers do not ask bare category questions. They arrive with context about their size, their stack, their industry, and where they are. So the ask should carry that context too. We prefix each monitored question with a short auto-written buyer persona, and for local and nationwide businesses a location line, with three of the four engines also receiving their native searcher-location setting. The question text itself stays the same across all four platforms, so you are comparing like with like.

Mistake 4: Checking ChatGPT and assuming the rest agree

They do not. ChatGPT leans heavily on what it absorbed in training and cites comparatively few live sources. Perplexity retrieves from the live web on nearly every answer and shows its citations. Gemini and Claude sit in between with their own retrieval behavior. A brand can be the top recommendation in one and completely absent in another, which is why a single-engine check is a quarter of a diagnosis at best. If you want to understand how the four differ in practice, we broke that down in our comparison of ChatGPT, Perplexity, Gemini, and Claude.

Mistake 5: Counting a source link as a mention

Plenty of tools inflate here. If an engine links your site in its sources but never writes your company name in the body of the answer, the buyer reading that answer does not see you. That is a citation, not a mention, and the two should be counted separately.

We detect every company named in an answer without telling the extractor which one is yours, then check afterwards whether any of those names match the business. It is a slower way to build the number. It is also the only way the number means anything.

Mistake 6: Chasing a single "visibility score"

Composite scores are easy to sell and impossible to verify. What is a 72 out of 100? Nobody can tell you, including the vendor who generated it.

Track figures that trace back to a specific answer you can read: mention rate, average position when mentioned, how many answers were analyzed, how many companies appeared, and your share of voice against named competitors. We deliberately do not publish a composite AI visibility score, because we are not going to invent a number and then ask you to trust it. The same logic applies to competitor data, where the useful part is the stored answers and their source links rather than a leaderboard rank. We covered that in tracking what AI tools say about competitors.

Mistake 7: Measuring visibility and stopping there

Mention rate is a leading indicator, not a result. The question your CEO asks next is whether any of it produced pipeline, and this is where most AI visibility programs go quiet.

Watch how the money actually moves. A VP of ops asks Perplexity which vendors handle multi-warehouse inventory, reads your name in the answer, clicks the post that got cited, leaves, comes back nine days later, and books a demo. In most analytics setups that second session lands in the direct bucket with no referrer attached, because many assistants strip the referrer or never pass one. The post that did the work gets no credit. The line item gets cut at the next budget review. We wrote up that gap in how AI assistants pass referral data.

So build the attribution path at the same time you build the monitoring, not a quarter later. Mention rate tells you the engines know you exist. Closed revenue tells you it mattered. Visibility you cannot connect to a dollar is a vanity chart with a better story.

If you want a baseline before building any of this yourself, the free AI visibility check will show you which questions your buyers ask and which companies get named instead of you. One snapshot is enough to tell you whether the gap is real. After that, the only thing that matters is whether somebody is watching it every day.