A repeatable method for checking whether ChatGPT names your brand: which prompts to run, what to record, and why a single manual check proves nothing.
ChatGPT has no ranking report. There is no position to look up, no dashboard inside it, and no notification when the answer about your category changes. The only way to find out what it says is to ask it and write down what comes back.
That part is easy, and most teams have already done it. The difficulty is that the same question asked twice does not reliably produce the same answer. Phrasing, conversation history, and model updates all move the result, so one check cannot tell you whether a brand appears reliably or appeared by luck.
A single ChatGPT check is an anecdote. What you need is a rate — how often your brand appears across many runs of the same question.
Do not start with your brand name. Asking “what is [brand]?” only confirms ChatGPT has heard of you, which is almost never the question that matters. Start with the prompts someone would use before they know you exist.
Ten to twenty prompts is enough to start. Write them down in a fixed list, because the method depends on asking the same questions every time.
ChatGPT uses the current conversation, and sometimes stored memory, as context. If you ask about your category in a thread where you have already mentioned your own company, the answer is contaminated and will flatter you.
“Were we mentioned?” throws away most of the signal in the answer. For each response, record:
The competitor and source columns are usually the most valuable. They tell you who you are actually losing to, and which pages ChatGPT is reading when it decides.
Run the full prompt list again at a fixed interval — weekly or monthly — and keep the results side by side. One run gives you a snapshot. Several runs give you a mention rate per prompt, and the direction it is moving.
This is also the step where manual tracking collapses. Fifteen prompts run weekly is sixty answers a month to generate, read, and code consistently, and the numbers are only comparable if whoever does it applies the same judgment every time.
Do the manual version at least once. It teaches you what the answers look like and what is worth recording, which makes any tool you adopt afterward far more useful.
The prompts where your brand never appears matter more than the ones where it always does. Those are the conversations happening without you, and nothing in your analytics will ever report them.
ChatGPT can recommend your brand without linking to you, and cite your documentation without naming you as an option. These are different problems with different fixes, so track them as separate columns.
Appearing in forty percent of answers sounds poor on its own. If the strongest competitor appears in thirty, it is a lead. Visibility in a generated answer is relative by nature.
Automate when you want the trend rather than the snapshot — when the question shifts from “does ChatGPT know us?” to “is our presence improving or eroding, and against whom?”
InfluenceLayer runs this method for you: prompt clusters executed against ChatGPT on a schedule, every response analyzed for brand mentions, competitor mentions, and cited sources, with the results accumulating into a trend instead of a spreadsheet.