AnswersStart here, for free

How do you find out what ChatGPT is telling buyers about your company?

Ask it. Genuinely: open it, ask the questions your customers ask, and read what comes back. That takes ten minutes, costs nothing, and is the right first move. Then ask the same question again, because the second answer will be different from the first, and understanding why is what turns ten minutes of alarm into something you can act on.

What can you find out yourself, right now?

Most of it. Which brands an assistant names in your category, roughly where you sit, and whether anything it says about you is wrong. That is enough to know whether you have a problem worth spending money on.

Open ChatGPT, Gemini, Claude, Perplexity or whichever your buyers use, and ask the question a buyer asks. Best running shoes for flat feet. Payment platform for a small shop. Whatever the honest version of that is in your category.

Two things matter about how you ask. Do not use your brand name: asking about you means you get discussed whatever the answer, which tells you nothing about whether you would have come up on your own. And ask the question a buyer types, which is usually a situation rather than a product: for a small team, that will not irritate sensitive skin, under fifty pounds.

Write down which brands come back and in what order, whether you appear at all, and anything stated about you that is not true. Ten questions across two assistants is twenty answers and about ten minutes, and it will tell you more than most people know about their own brand.

Why does the answer change when you ask again?

Because an assistant generates its answer rather than looking it up, so two answers to the same question can name different brands in a different order. It is the single most important thing to know before drawing any conclusion.

Ask the same question five times and you will often get five overlapping but different shortlists. This is not a fault and it will not be fixed: it is how the technology works, and it is the reason a single check is a starting point rather than a finding.

What follows from it is worth taking seriously. If you check on Monday and see yourself, and check on Friday and do not, nothing necessarily changed. If you do work for a quarter and the reading improves, you cannot say from one check whether the work did it. Both directions of that mistake are easy to make and expensive.

So the useful unit is not an answer, it is a rate. How often, out of how many tries, across which assistants. That is also why it stops being a ten minute job at the point you want to know whether anything is moving.

What should you write down while you do it?

Four things per answer: whether you were recommended, whether you were named at all, which competitors were named, and anything said about you that is wrong. Those four separate problems that need completely different work.

Being recommended and being mentioned are different outcomes, and a tally that mixes them will tell you visibility is improving while nothing is being sold. Keep them apart from the first day.

Record the competitors too. Which brands keep appearing beside you is the most actionable thing in the exercise, because those are the ones the sources your assistants read are describing well. And note where the assistant says it got something, when it tells you, because that list of sources is short and it is where the work eventually goes.

A spreadsheet with one row per answer is entirely adequate for this. Question, assistant, were you recommended, were you named, who else was, anything wrong.

When is this worth paying somebody to measure?

When you need to know whether something changed. Checking by hand answers what is true today, and cannot answer whether last quarter's work moved anything, because the noise between two checks is larger than most real improvements.

Three situations where hand checking stops being enough. You want to know whether work you did had an effect, which needs enough repetition to tell a real change from the variation. You are answerable to somebody for it, which needs a figure that means the same thing month to month. Or the question set is large enough that doing it by hand takes a week rather than an afternoon.

Until one of those is true, the spreadsheet is genuinely fine, and anybody telling you otherwise is selling. What a measurement adds is the denominator, the repetition, and a margin of error, so a movement can be called or not called honestly.