How to check what AI says about your business

Before you start: turn off the things that flatter you

The single most common mistake is auditing from an account that already knows who you are. Your search history, your saved places and your chat history all push the answer towards the one you were hoping for, and the result is a clean bill of health for a listing that is badly wrong for everyone else.

• Open a private or incognito window for anything in Google.
• In ChatGPT, turn memory off in settings, or use a temporary chat. Otherwise it may be answering from things you told it months ago rather than from what it actually knows.
• If you can, check from a phone on mobile data rather than your shop wifi. Location is a strong signal and your own address skews it.

The four questions

These are the four worth asking, in this order. They map onto the four things that actually cost you customers: not existing, being wrong, being unreachable, and losing to someone else.

1. Does it know you exist, and what does it think you are?

“What do you know about [business name] at [full street address]? Include opening hours, services and contact details.”

Include the street address. Without it you cannot tell an answer about you from an answer about a business with a similar name three towns over, and you will spend an afternoon fixing a problem that was never yours.

A reply of “I do not have specific information about this business” is a finding, not a blank. It means you are invisible at the exact moment someone is asking about you by name.

2. Would it recommend you?

“Which are the best [your category] in [your town], and why?”

Ask it plainly and do not mention yourself. If you are not named, write down who is — that list is your real competitive set as the engine understands it, which is often not the one you assume.

3. Are you open?

“Is [business name] in [town] still open? What are its hours today?”

This is the expensive one. A wrong “permanently closed” ends the customer’s search entirely — nobody calls to double-check a business they believe has shut down. If you get this, stop the audit and go straight to fixing the closed flag, because nothing else matters until it is gone.

4. Where did that come from?

“What sources did you use for that answer?”

Ask this after any answer that was wrong. Many engines will name a page. That page is almost always the thing you need to fix — and fixing the source is the only durable repair, because correcting the model itself is not something anyone can do.

Where to ask

Different engines pull from different places, so one clean answer proves very little. Check at least two:

• Google AI Overviews. Just search your business name and category in an incognito window and read the box above the results. This is the highest-traffic surface by a wide margin — see what AI Overviews changed for local business.
• ChatGPT. The one your customers name out loud.
• Gemini. Closest to Google’s own local data, so it often reflects your Business Profile fastest.
• Perplexity or Claude, if you have a minute more. Both cite sources readily, which makes step four easier.

What counts as a real problem

Not everything wrong is worth acting on. Sort what you find into three piles and only work on the first two.

• Fix today. Closed when you are open. Wrong address. Wrong phone number. A service you do not offer being promised to customers. These send people to the wrong place or lose them outright.
• Fix this month. Stale hours, an old business name, a missing service you do offer, absence from the “best in town” list.
• Ignore. Hedging (“hours may vary, call ahead”), a slightly clumsy description, or an answer that is vague but not wrong. Chasing these burns time you should spend on the first pile.

Write down what you got

Paste each answer into a document with the date. This sounds fussy and it is the entire value of the exercise: a single audit tells you about today, and what you actually need to know is whether things are getting better or worse after you have changed something. Without last month’s answer, this month’s is just a vibe.

Corrections also propagate at wildly different speeds — a Business Profile edit can show up in Gemini in days and in ChatGPT weeks later. Only a dated record tells you whether a fix landed or you are still waiting.

Doing it by hand has a shelf life

The audit above is genuinely worth running, and for a single location that rarely changes, once a month by hand is fine. It stops being practical when you have several locations, or when you need to know the week something breaks rather than the month.

That is what AI Business Monitor does: the same questions, every week, with the answers stored so you can see what changed and the source attached to every finding.