How to Check if ChatGPT Recommends Your Brand

Somebody on your team has already done it: opened ChatGPT, typed “best tools for [your category]”, and either celebrated or quietly closed the tab. That instinct is right. The method is wrong. One answer from one assistant on one day tells you almost nothing, because assistants give different answers every time you ask.

Here is how to check properly, first by hand, then automated. The manual version costs an afternoon and a spreadsheet, and it is worth doing at least once even if you buy a tool later, because it teaches you what the numbers mean.

Step 1: write prompts like a buyer, not like a marketer

You are not testing whether ChatGPT knows your brand exists. You are testing whether it recommends you when a buyer asks a buying question. So write the questions buyers actually ask:

• “What’s the best [category] for [audience]?” (“best help desk for a two-person startup”)
• “Alternatives to [the market leader]”
• “[Competitor] vs [competitor], which should I pick?”
• “How do I solve [the problem your product solves]?”

Write 15 to 25 of these. Avoid prompts that contain your brand name; those measure recognition, not recommendation. The exception is one or two “is [your brand] any good?” prompts to check what the assistant says when asked about you directly.

Step 2: sample, don’t spot-check

Ask each prompt several times, in fresh chats, with no custom instructions or memory switched on. Log every answer. This is the step everyone skips and it is the one that matters most: assistants are probabilistic, and your brand might appear in four answers out of ten. A single ask cannot tell you whether that number is four or zero.

Ten runs per prompt is a reasonable floor. Fewer than that and any percentage you compute is mostly noise.

Step 3: score the answers

For each answer, record four things in your spreadsheet:

1. Mentioned? Did your brand appear at all (including obvious aliases)?
2. Position. If mentioned, was it the first brand named, the third, the fifth?
3. Competitors. Which other brands appeared, and in what order?
4. Citations. If the assistant cited sources (ChatGPT does when it searches the web), which sites did it cite?

From those columns you can compute the numbers that matter: your visibility rate (share of answers that mention you), your share of voice against competitors, your average position, and which sources are shaping the answers. Our plain-English definition of AI visibility explains each metric in more depth.

Step 4: read the citations, not just the mentions

The citation column is the actionable one. If ChatGPT keeps citing a review site where you have three reviews and your competitor has eighty, that is a concrete task. If it cites a community thread from two years ago that misdescribes your product, that is another. Mentions tell you the score; citations tell you why.

Step 5: repeat on a schedule, or automate it

The manual method’s real cost is not the first afternoon, it is every following week. The numbers only become useful as trend lines: did the model update change anything, did your new comparison page move share of voice, is a competitor climbing. Run the loop weekly by hand and it will quietly stop happening within a month. That is the honest pitch for automating it.

This is what CitePrism does: it runs your prompt set against ChatGPT, Claude, Gemini, Perplexity, Google AI Overviews, Google AI Mode and Microsoft Copilot on a schedule, stores every raw answer, and turns the results into visibility rate, share of voice, rank and citation metrics over time. The free plan runs on your own keys, so the sampling costs you API price, roughly five cents per prompt across the four API engines, and Pro runs on ours from $29 a month. And because it stores raw answers, you can always click through from a chart to read exactly what the assistant said.

Whether you use a spreadsheet or a tool, be clear about what you are measuring: structured sampling of the assistants, not real user conversations, which are private and invisible to everyone. We wrote about that distinction in what AI visibility tools actually measure.

FAQ

Why do I get different answers than my colleague for the same prompt?

Assistants are probabilistic, and answers vary run to run even in identical conditions. Personalization adds more variance: custom instructions and chat memory both shape answers. For measurement, use fresh chats with memory and custom instructions off, and rely on many samples rather than any single answer.

Does asking ChatGPT about my brand repeatedly change what it says later?

No. Your prompts do not update the model, and with memory off nothing carries between chats. Sampling is read-only: you can measure as often as your API budget allows without influencing future answers.

How often should I re-check?

Weekly trend points are enough for most brands, which means running the full prompt set at least daily if you want error bars tight enough to trust a weekly movement. Model updates ship without notice, so continuous scheduled sampling catches shifts that a monthly manual check would misattribute or miss.

Can I just check ChatGPT and skip the other assistants?

You can start there, but recommendations happen across ChatGPT, Perplexity, Gemini and Claude, and they disagree more than you would expect. Perplexity in particular is worth including early: it cites sources on almost every answer, which shows you exactly which sites are shaping your category.