Does ChatGPT recommend your business?
Here is how to find out in ten minutes.

The ten minute check

You need a phone or a laptop, a private window, and ten minutes. Nothing else, and no tool.

Open a fresh private or incognito window and make sure you are signed out of the assistant. This matters more than anything else in the method. Signed in, the answer is shaped by everything you have ever asked it, including questions about your own business. You would be measuring your own history rather than what a stranger sees.

Then ask these five questions, one per window, replacing the trade and the city with your own.

1. What's the best [your trade] in [your city]?
2. I'm looking for a [your trade] in [your city] for [your main service]. Who should I go to?
3. Recommend a [your trade] near me in [your city]
4. Top rated [your trade] in [your city]
5. Who does good [your main service] in [your city]?

Ask each one in its own fresh window. No follow-ups, no refining, no "what about X". The moment you have a conversation with it you are no longer measuring the first impression a customer gets.

Do it on three assistants, not one. ChatGPT, Gemini and Perplexity. That is fifteen answers. The reason is in the findings below, and it is the single most important thing on this page.

Write down four things for each answer

For every one of the fifteen answers, record:

• Every business named, in the order they appear. Position one matters more than position six.
• Whether you appear at all. This is the only binary on the list.
• What it said about you, if anything. Ratings, review counts, services, hours, the name of your practitioner.
• Which sources it cited. Most assistants show these. They are the most useful thing in the whole exercise and almost nobody looks at them.

How to read your score

Out of fifteen answers:

• Named in ten or more: you are genuinely visible. Check what is being said about you rather than whether you appear.
• Named in four to nine: you exist in some sources and not others. Usually one assistant can see you and another cannot.
• Named in one to three: you are a long way down the sources these systems read.
• Named in none: for customers asking an assistant instead of searching, you do not exist. This is more common than business owners expect, including for businesses with good Google rankings.

What we found when we ran this properly

In October 2026 I ran exactly this method against Las Vegas med spas. Five questions, three assistants, fifteen answers, all in fresh private windows on one day. The full study and the complete run log are published, so anyone can repeat it.

Four things came out of it that change how you should read your own results.

There is no single answer to check. Fifteen answers named twenty four different businesses in their written recommendations. Only two were named by all three assistants. Six different businesses took first place depending on which one you asked.

A business can be prominent on one assistant and absent from another. One clinic appeared in four of five ChatGPT answers and none of Gemini's. Another was the exact reverse. Neither has any way of knowing that without running the check.

Review count is not the gate you assume it is. A clinic with four reviews was recommended in an answer to "top rated". Another with two reviews turned up in a treatment recommendation. Established practices with thousands of reviews also appeared. Both routes exist in the data.

The answers can be wrong about you. One assistant named the surgeon at a real cosmetic surgery practice incorrectly, in a session where it had given the correct name three questions earlier. I verified the real name against three independent sources before publishing. If an assistant can invent a doctor's name, it can get your pricing, your hours or your services wrong too, and the customer reading it has no reason to doubt it.

The part most people skip

Look at the sources.

In the study, the assistants were mostly not citing the places businesses spend their time on. Across the fifteen answers they cited seven niche directories: DiscoverMedSpa, ClinicCompass, ThreeBestRated, TrustAnalytica, MedicalSpaDirectories, MedSpa Scout and Medical Spa Locator. ChatGPT cited Yelp on two questions and then cited three industry directories and no Yelp at all on the next.

Your category will have its own equivalent. Run the check, write down every source cited across all fifteen answers, and you have a list of the places these systems are actually reading when someone asks about your trade in your city. That list is specific to you and it did not exist before you made it.

What this check does not tell you

Fifteen answers is a snapshot, not a trend. Ask the same questions next week and some of it will move.

It shows no cause. You will finish with a list of sources the assistants cited and a list of businesses they named, and nothing connecting the two. Being listed somewhere might be why a competitor appears. It might not be. Treat the source list as the most promising thing to test rather than a lever known to work, and test it the way you would test anything else, by changing one thing and running the check again.

And it says nothing about quality. Appearing often means a business is well represented in the sources these systems read. It is not a measure of whether they are good at the job.

What to do with your results

1. Run it before you change anything. Write down the score. Without a baseline you will not be able to tell whether anything you do afterwards worked.
2. Check the basics are consistent everywhere you appear. Same business name, same address, same phone, same service list. When an assistant is assembling an answer out of seven different directories, a disagreement between them is how you end up described incorrectly or left out. The six things worth fixing, in order, are here.
3. Look hard at the directories doing the citing. Several are free. This is the most promising thing on the list and also the least proven, which is exactly why it is worth testing cheaply rather than assuming.
4. Check what is being said about you, not just whether you appear. Wrong hours and a wrong price cost you more than absence does.
5. Run it again in sixty days. Same five questions, same three assistants, same conditions. The second score is the one that tells you something.

Running this once takes twenty minutes. Working out what the source list actually means for your business, and what to change because of it, takes longer. That part is the job I do as an AI marketing consultant, including an honest answer about the places where AI will not help you.