You cannot tell whether ChatGPT recommends your product by asking it once. Responses vary between runs, between accounts, between model versions, and between plain mode and search mode, so a single chat is an anecdote, not an answer. The honest version of the question is: in what percentage of relevant buyer conversations does your product appear, and where in the list?
That question has a real answer, but it takes sampling to get it.
Why one conversation proves nothing
ChatGPT's outputs are probabilistic. The same prompt can produce a list with your product on Monday and a list without it on Tuesday, with no change in the underlying model. Add the variance between model versions, and between the base model answering from its weights versus search mode retrieving live pages, and the spread gets wide.
This cuts both ways. Founders who saw their product named once assume they are covered, and founders who got one bad answer assume they are invisible. Both conclusions are unsupported by a sample size of one.
Your own account is the worst place to check
ChatGPT's memory and chat history personalize answers. If you have discussed your own product with it for months, it will name you more often for you than it does for a stranger, which is precisely the number that does not matter. Check in a temporary chat or a logged-out session at minimum, and treat even that as a single draw from a wide distribution, not the verdict.
What a representative sample looks like
A defensible answer to "does ChatGPT recommend my product" comes from a structured sample:
- Prompt variety: recommendation, comparison, alternatives, and use-case prompts, phrased the way real buyers phrase them
- Multiple runs per prompt, because a single run hides the variance you are trying to measure
- Both modes: base model answers and search-enabled answers, which often name different brands
- Clean sessions with no memory, so personalization does not contaminate the read
- Competitor logging: record every brand named, not just yours, so you get position and share, not just presence
Turn the sample into a metric
Score the sample as a mention rate: the share of relevant conversations in which your product appears, broken out by prompt type and position. That number is comparable month over month and honest about uncertainty in a way no screenshot ever is. The framework for building it is in how to measure AI visibility, or the free scan runs the sampling for you and reports mention rate, position, and competitor share across five providers.
Get your real mention rate
Stop guessing from single chats. The free scan samples buyer prompts across ChatGPT and four other providers and reports how often you actually appear, and where.
Frequently asked questions
Does ChatGPT give the same product recommendations to everyone?
No. Recommendations vary between runs because outputs are probabilistic, and memory, chat history, custom instructions, and whether search mode triggers all shift the results further. Two users asking the identical question can get different brand lists, which is why measuring recommendation frequency requires sampling many clean sessions.
How many prompts do I need to know if ChatGPT recommends me?
Enough to cover the main buyer intents with repetition: as a floor, a panel spanning recommendation, comparison, alternatives, and use-case prompts, each run multiple times in clean sessions. What matters most is consistency, running the same panel the same way over time so changes in your mention rate are real signal.
Should I test ChatGPT in search mode or regular mode?
Both, and record them separately. Regular mode answers from model weights, which reflect training data and change only at retraining. Search mode retrieves live pages, so it can include newer brands and it cites sources. The two modes often produce different consideration sets, and they respond to different optimization work.
Why did ChatGPT recommend my product yesterday but not today?
Normal variance. ChatGPT samples from a probability distribution, so brands near the edge of the consideration set appear intermittently. If your mention rate across many runs is around 50 percent, you will see exactly this flip-flopping. The fix is raising the underlying probability with stronger sources, not chasing individual answers.
Related answers