How to check if ChatGPT recommends your business

You can find out a lot about how AI engines treat your business in an afternoon, with nothing but a browser and a spreadsheet. This guide is the method: which questions to ask, how to ask them so the result means something, what to write down, and where a hand check stops being reliable.

Step one: stop asking about yourself

The instinct is to type your own business name and see what comes back. Do it once if you are curious, but understand what it measures. A question that contains your name will almost always produce an answer about you, and usually a kind one. It tells you whether the engine has heard of you and whether its description is right. It does not tell you whether a stranger looking for what you sell would be sent your way.

For that, you need non-branded questions: the ones a buyer asks before they know you exist. The share of those answers that name you is your organic mention rate. Keep it separate from anything you learn by asking about yourself by name. The two measure different things, and adding them together produces a number that flatters you by construction.

Step two: write the questions your buyers ask

Aim for ten to fifteen questions. Mix a few kinds, because buyers do not all arrive the same way:

• Direct discovery — “Best plumbers in Grand Rapids, MI”, “Who are the most reliable roofing contractors in Boise”.
• Problem-first — the way people ask mid-crisis: “My basement is flooding, who should I call in Grand Rapids”, “My furnace stopped working, who repairs them near Boise”.
• Selection — “How do I find a trustworthy accountant in Tucson”, “What should I look for when hiring a kitchen remodeler”.
• Specific service — “Who installs tankless water heaters in Grand Rapids”, “Which family law attorneys in Tucson handle custody cases”.
• Reputation without a name — “Which landscapers in Boise have the best reviews”.

Write them the way a customer would, not the way a marketer would. If people call you saying “my AC is blowing warm air”, that is the question, not “residential cooling system diagnostics”. And put the city in the question. It matters more than it seems, for reasons covered below.

Step three: ask them cleanly

Assistants remember things, and personalization skews results. A few habits keep the test honest:

• Use a fresh conversation for every question. Earlier questions in the same chat shape later answers.
• Turn off memory or use a temporary chat where the engine offers one, or use a logged-out browser window. An assistant that knows you own a bakery will be more inclined to mention bakeries you care about.
• Ask more than one engine. ChatGPT, Perplexity, Gemini and Claude do not draw on identical sources, and a business named by one is often absent from another.
• Ask each question more than once. Run each at least two or three times, ideally on different days.

Step four: write down more than yes or no

A spreadsheet with one row per question, per engine, per run is enough. For each answer, record:

1. Whether you were named at all.
2. Where in the answer you appeared — first, or a passing mention at the end.
3. Whether what it said about you was accurate: services, area, phone, hours.
4. Who was named instead of you, or alongside you.
5. Which sources the answer cited, where the engine shows them.
6. Whether the engine failed to answer — an error, a refusal, a timeout.

Keep that last column separate. An answer you did not get is not tested, not not mentioned. If you count a failed run as a miss, your number is lower than the truth; if you quietly drop it, you lose track of how much you actually measured.

The sources column is the one people skip, and it is the most useful. Engines that search while they answer, like Perplexity and ChatGPT with search, often list where their information came from. If the same directory, roundup article or forum thread keeps appearing, that is a place your business needs to be.

Why a single check misleads

Here is the uncomfortable part. Run the same question twice and you may get two different lists. These systems generate answers rather than retrieve a fixed ranking, and the sources they pull in can change between runs. One check that names you is not proof you are recommended; one that does not is not proof you are invisible.

Location adds a second layer of noise. Engines may take your approximate location into account, so “best electrician near me” asked from your office is not the question a customer asks from the suburb across the river. That is why the city belongs in the question itself: it makes your test closer to what the customer sees, wherever you happen to be sitting.

Then there is time. Engines update their models and their search behavior, the web they draw on changes, competitors get reviewed and written about. A result from March says little about September.

When to stop doing it by hand

A hand check is worth doing once, and it is the right starting point. It becomes the wrong tool when you start changing things and want to know whether the changes worked. At that point you need the same question set re-run on a regular cycle, results recorded the same way each time, and enough answers that a single odd run does not swing the number. Fifteen questions across three engines, three runs each, is already 135 answers to read and classify — every month.

That is the job Responsii exists to do. The Visibility plan asks up to 40 buyer questions built from your category across every AI engine we support, once a month. It reports your organic mention rate separately from the questions that use your name, checks accuracy when you are named, shows who gets named instead and which sources the answers drew on, and flags answers it could not score confidently rather than counting them either way.

Whether you measure by hand or with a tool, the principle is the same: fix the questions, keep the method constant, and judge the trend, not the latest answer.

Try the free version first

If you want a real result before you build a spreadsheet, the free audit runs eleven buyer questions for your business — nine that never use your name and two that do — across two AI engines, and emails you what came back. It takes about two minutes to set up, with no account and no card.