How to Test What ChatGPT Says About Your Brand Right Now

Ask most marketing teams what ChatGPT says about their brand and you’ll get a guess. Ask them when they last actually checked, in a real chat window, with the prompts a buyer would use, and most haven’t. That gap is the starting point for any GEO work worth doing.

Testing this yourself takes about twenty minutes and costs nothing. Here’s how to do it properly, and what to do with what you find.

Why brand-name prompts don’t tell you much

The instinct is to type your company name into ChatGPT and see what it says. That’s worth doing once, but it’s not the test that matters. A brand-name prompt tells you what ChatGPT knows about you when directly asked. It doesn’t tell you whether you show up when someone hasn’t heard of you yet.

The prompt that actually matters is the category prompt: the question a prospect asks before they know your name. “Best project management software for a 20-person agency.” “How do I find a fractional CFO for a Series A startup.” “What’s a good alternative to [competitor].” These prompts reveal whether your brand gets surfaced in the moment that counts, when someone is deciding what to consider, not just confirming what they already found.

Build a real prompt list, not one or two questions

A single prompt tells you almost nothing. Build a list of 10 to 20 prompts that reflect how your actual buyers search, across a few categories:

1. Direct category queries. “Best [category] for [use case].” Run a few variations with different qualifiers (company size, budget, industry).
2. Comparison prompts. “[Competitor A] vs [Competitor B]” and “[Your brand] vs [main competitor].”
3. Problem-first prompts. The question someone asks before they know a category exists. “How do I track brand mentions across AI tools” rather than “best brand monitoring software.”
4. Alternative-seeking prompts. “Alternatives to [well-known competitor].”

Pull these from your own sales conversations and support tickets if you can. The prompts your customers actually typed into Google before they found you are usually close to the prompts they’d type into ChatGPT.

Run each prompt in a fresh session

This is the part people get wrong most often. Testing in a single ongoing chat, where you’ve already mentioned your brand or company earlier in the conversation, contaminates the result. ChatGPT has context from that conversation that a first-time user wouldn’t have.

Open a new chat for every prompt. Don’t sign in with an account that has memory turned on and a long history of asking about your own company, since that history can bias the answer. If you want the cleanest read, test in a logged-out or fresh-context session where available.

Run each prompt at least two or three times, since LLM outputs vary between runs even with identical input. A brand that shows up in two out of three runs is in a meaningfully different position than one that shows up in zero out of three, and you won’t see that difference from a single test.

What to actually record

For each prompt, capture more than whether you were mentioned. Note:

• Whether you appear at all, and where in the answer (first mentioned, buried in a longer list, absent).
• What ChatGPT says about you, specifically whether the description is accurate, outdated, or generic.
• Which competitors appear alongside you, since this tells you who you’re actually being compared against in AI answers, which isn’t always the same set you compete with in paid search.
• What sources the model references, when it cites or names them. Some answers will explicitly mention where information came from (“according to reviews on G2” or “based on Reddit discussions”). This is the most actionable data point, since it tells you which channel is actually driving the answer.

What you’re checking	Why it matters
Are you mentioned for category prompts	Shows discovery-stage visibility, not just brand recognition
Is the description accurate and current	Outdated info signals stale source material feeding the model
Who else is mentioned	Reveals your real AI-era competitive set
What sources get cited	Tells you where to focus GEO effort first

Testing across models, not just ChatGPT

ChatGPT is the obvious starting point because it has the largest user base, but Perplexity, Google’s AI Overviews, and Claude all pull from different source mixes and can produce different answers for the same prompt. A brand can be well represented in one and invisible in another. If you only test one model, you’re checking one slice of how AI-driven discovery actually works, which increasingly spans several tools depending on the audience.

Turning a test into an actual program

A one-time test gives you a snapshot. It tells you where you stand today, and for most teams doing this for the first time, that snapshot alone is useful; it’s often the first real evidence of a gap between how a brand sees itself and how AI tools describe it.

The next step is deciding what to do with that gap. If the sources being cited are wrong or missing, that points to specific fixes: getting reviews on the platforms that matter, correcting information indexed from your own site, or building presence in channels like Reddit that show up disproportionately often in these answers. If you want a structured way to think about where your brand sits before investing further, our GEO maturity model walks through the stages most companies pass through on the way to a real program.

Running this test once is worth doing today. Running it consistently, with the same prompt list, on a regular cadence, is what turns a one-off curiosity into something you can actually manage.