Resources · Diagnose
How to check if ChatGPT knows your brand
The short answer
You don’t need a tool to find out what ChatGPT thinks of your company. You need ten minutes and a list of the questions your buyers already ask. Type them in, read the answers like a stranger would, and note where the model is right, where it’s confidently wrong, and where it draws a blank.
First, stop your own history from answering for you
If you’ve been chatting with ChatGPT about your company for months, it may “know” you from your own account rather than from the open web. That’s not the answer a buyer gets. Before you test anything, turn off memory and personalization, log out, or open a fresh temporary chat. Now you’re looking at the same blank slate a prospect sees.
Ask the direct question
Start plain. “What is [your brand]? What do they do?” Read the reply and check it claim by claim against what’s actually true. This is the cleanest signal of whether the model has any real picture of you, before you get into who wins which comparison.
Say you sell a scheduling app for barbershops. Ask “What is ChairTime and who is it for?” If the answer describes a generic calendar app, or invents a founder, or puts you in the wrong industry, you’ve learned something already.
Then ask what buyers actually ask
Nobody types your name when they’re still shopping. They ask the category question, and you want to know whether your name comes up in the answer. Work through a few shapes of prompt:
- Category.“Best [category] for [use case].” For the barbershop app: “best booking software for a small barbershop.” See if you get named at all.
- Comparison.“[Your brand] vs [competitor].” This tells you how the model frames you against the name buyers weigh you against, and whether it even knows you both.
- The facts.Ask for pricing, who it’s for, and who founded it. Then look hard for details that are wrong or simply made up. A confident price you never charged is the model filling a gap, and worth writing down.
Getting named in that category answer is a brand mention, and it’s the thing that eventually turns into a click and a customer. No mention, no shot.
Run each prompt more than once
One run tells you almost nothing. Ask the same question five times, each in a fresh chat, and count how often you show up. You might appear four times out of five, or once, or never. That hit rate is the real answer, and a single screenshot hides it completely.
Check more than ChatGPT
ChatGPT is one engine reading one slice of the web. Gemini, Claude and Perplexity read different sources and can hold a different opinion of you entirely. Perplexity is the easiest place to start, because it shows its citations inline, so you can see the exact pages it pulled to build the answer.
Run the same short list of prompts on all four. If you’re strong on ChatGPT and missing on Gemini, that’s a source-coverage problem on the pages Gemini leans on, not a sign that you’re invisible everywhere.
Read the result
Once you’ve run your list, every engine sorts into one of three buckets:
- It knows you and gets you right. The description matches reality and you turn up on the category questions. Keep the sources it trusts fresh and accurate.
- It knows you but the details are wrong. The riskiest one. The model sounds certain while stating a bad price or the wrong audience, and buyers believe it.
- It has never heard of you.You’re absent for that query. There’s no fact to correct yet, only a gap to fill.
Common questions
- Why do the answers change every time I ask?
- ChatGPT doesn't hand back one fixed answer. It generates a fresh one each run, so the same prompt can name you once and skip you the next time. That's why you ask each question a few times instead of trusting a single reply.
- Do I need to test anything besides ChatGPT?
- Yes. Gemini, Claude and Perplexity pull from different sources, so they can hold very different pictures of you. If you only check ChatGPT, you've measured one engine and guessed at the rest.
- ChatGPT got a fact about us wrong. What does that mean?
- That's the middle outcome: it knows you but the details are off. The model is working from sources that are stale, thin, or contradict each other, and it fills the gaps with guesses. It won't correct itself. You fix it by fixing the sources it reads.