How to Test Whether AI Recommends Your Business: a 20-Minute Check

Owners ask me whether AI recommends their business, and almost nobody has checked. The ones who have usually asked once, got a bad answer, and concluded they are invisible. Or got a good answer and concluded they are fine.

Both conclusions are wrong for the same reason. One run tells you nothing, because these systems produce different answers to identical questions.

Here is a method that gives you something real, and it takes about twenty minutes.

Why one check is worthless

Ask ChatGPT for the best plumber in your town three times in three fresh chats and you will often get three different lists. Some names repeat, some appear once, the order moves.

That is not a malfunction. These systems are probabilistic. They assemble an answer fresh each time, and small differences in how the question is read produce different results.

Which means the only meaningful measurement is a rate, not a result. Not "does ChatGPT recommend me" but "out of thirty attempts, how many times was I named." That number can be compared to last quarter's. A single answer cannot be compared to anything.

This is also how you spot a bad AI visibility report. If somebody hands you a clean ranking position for AI search, ask how many runs it came from. If the answer is one, or if there is no answer, the number is decoration.

Step one: write the ten questions

The questions have to be the ones a real customer would type, not the ones you wish they would.

Three kinds are worth including.

The direct ask. "Best physiotherapist in Brantford." "Who should I call for emergency plumbing in Brant County." This is the highest-value question and the hardest to win.

The problem, not the service. This is where people forget to look. A customer with a problem rarely names the trade. "My furnace is running but blowing cold air, who do I call in Brantford." "I have knee pain from running, what kind of clinic should I see in Hamilton." These are long, specific and far more winnable, because your competitors are not writing for them.

The comparison or qualifier. "Physiotherapy clinic in Burlington that direct bills." "Electrician in Ancaster who does EV charger installs." Qualifiers are where a smaller business can beat a bigger one, because the qualifier narrows the field to whoever actually mentions it.

Write ten. Mix all three kinds. Include your town by name in most of them, because that is how people ask.

Step two: run them properly

Three rules, and they all matter.

Use a fresh chat every time. If you keep asking in the same conversation the system remembers what it just said and will tend to repeat itself. Every run starts clean.

Log out, or use a private window. If you are signed in, the assistant may know who you are and where you are, and it may have memory of your previous conversations. You will get a flattering answer that no customer would see.

Run each question three times, on each engine. Ten questions, three runs, three engines is ninety data points. That sounds like a lot and it goes quickly because you are only recording names.

Use ChatGPT, Gemini and Perplexity at minimum. They behave differently and the differences are informative.

Step three: record it in a way you can repeat

A spreadsheet with one row per run. Five columns: the date, the engine, the question, whether you were named, and which competitors were named.

That last column is the one owners skip and it is often the most useful. After ninety runs you will have a tally of which businesses in your town the machines trust. That list is worth studying. Go and look at what those businesses have that you do not, and it is usually unglamorous. More reviews. More pages. Listed in more places.

Then calculate two numbers.

Your mention rate. Times you were named, divided by total runs. If you appeared in 12 of 90, that is 13 percent. Write it down with the date.

Your mention rate per engine. You will often find you do well on one and poorly on another, and that tells you where to look.

Date the whole thing clearly. These results are a snapshot of a moving system, and a number without a date is useless in six months.

Step four: decide what it means

Named in most runs. You are in good shape. Keep reviews current and keep your details consistent, then re-check quarterly to catch a slide early.

Named occasionally. You are on the edge of being trusted. This is usually a corroboration problem rather than a website problem. You appear in some sources but not enough of them agree. Getting listed consistently on the third-party pages that already rank for your category is the usual fix.

Never named. Do not panic and do not buy an AI package yet. Go and check the ordinary foundations first, because this almost always traces to something basic. Search your own business name on Bing and see whether you appear at all. Check your Google Business Profile is complete. Check whether your website is even indexed.

Competitors named who are worse than you. This is the common and genuinely annoying result. It usually means they are easier to verify, not better. More reviews, more listings, more pages saying plainly what they do.

Then do it again

The number only becomes useful the second time you produce it.

Put a repeat in the calendar for three months out. Same ten questions, unchanged. That last part matters: if you rewrite the questions you have lost the comparison. Freeze the set, keep the spreadsheet, and the second run tells you whether the work moved anything.

That is the whole method. It is cruder than a search ranking report, it is the honest state of the art, and it is more than almost any of your competitors will ever do.

Common questions

How many times should I run each question?

Three per engine is a workable minimum for a small business. More runs give a steadier number, but three is enough to stop a single lucky or unlucky answer from misleading you.

Why do I need to log out to test this?

Because a signed-in assistant may know your location and may remember your earlier conversations, including ones where you talked about your own business. That produces a flattering answer a real customer would never see.

Which AI assistants should I check?

ChatGPT, Gemini and Perplexity cover most of the current usage and they behave differently enough to be worth checking separately. Add Google AI Overviews by running the same searches in Google and noting what the AI summary at the top says.

What is a good AI mention rate for a local business?

There is no published benchmark, and anybody quoting one is inventing it. The useful comparison is against yourself over time and against the competitors showing up in your own results.

Can I pay a tool to track this for me?

Tools exist and some are useful for saving time. Judge one on whether it shows you the variance between runs and stores the raw answers. If it reports a single clean position with no indication of how many runs produced it, it is hiding the thing you most need to see.