How to check what AI says about your business in five minutes

Set a five-minute timer and open one AI assistant. You are not trying to “audit your AI presence” before the timer ends. You are collecting one clean piece of evidence you can compare later.

The finished note should answer four things: what you asked, what the system said, which businesses it named, and which sources it showed. That is a baseline, not a ranking.

Start with a buying question, not your business name

Asking “What is [my business name]?” only checks whether the system can describe a name you supplied. A customer who has not chosen you yet asks a different kind of question.

Use a category, a place, and one real constraint:

• “Which florist in Petaling Jaya can handle a small wedding?”
• “Recommend an air-conditioner repair company in Cheras that works weekends.”
• “Compare tuition centres in Subang Jaya for a Form 5 student struggling with chemistry.”
• “Where can I find a halal birthday cake in Shah Alam for 20 people?”

Pick the one closest to an enquiry you received recently. Natural wording is the point. Leave your brand name out unless a customer would genuinely include it.

Keep the answer, not just your reaction to it

Copy the full response into a note or spreadsheet. “We were missing” is too little information to help you six weeks from now, and a screenshot alone is awkward to search.

Capture these fields while the answer is still open:

Field	What to save
Prompt	Every word, including location and constraint
Engine	The named product you used
Conditions	Signed in or out; web search on or off where visible
Date	When the run happened
Your business	Named, absent, or described wrongly
Other businesses	Names and order shown
Claims	Services, location, price, hours, or reputation statements
Sources	Every citation or link exposed

Open the links before the timer ends if you can. A citation displayed beside an answer may support one claim without supporting the rest. Your note should say what the page actually confirms.

Disturb the test in one useful way

Change one buyer need, not just one adjective. A florist could test a small wedding, same-day delivery, sympathy flowers, and a fixed budget. Those are different jobs a customer may hire the same shop to do.

Three to five variations are enough for a first pass. Twenty near-identical prompts create a bigger spreadsheet without giving you twenty meaningful customer situations.

If your business appears several times, record that frequency within this small test. Do not convert it into a secure position. Generated answers vary, and nobody can honestly guarantee a permanent ChatGPT ranking.

Use a second engine as a second witness

Run the same core question on another named system. ChatGPT, Claude, and Perplexity can use different models, search tools, source sets, and product settings. Agreement is interesting; disagreement is also information.

Keep the results separate. Writing “AI recommends us” after one product named you hides which system was checked and makes the claim impossible to reproduce.

The same rule applies when Google Maps disagrees with the AI answer. Google Maps, Google Search, and each AI assistant are separate products. A strong Maps position does not compel another system to repeat the result.

End with one of three diagnoses

Do not finish with a vague “visibility problem.” Label what you actually observed:

1. Absence: your business did not appear in the tested answers.
2. Accuracy: it appeared, but a fact was wrong or stale.
3. Positioning: it appeared, but the description missed the service or customer type that matters.

An old opening time points to a factual correction. A missing service may point to a page that never states the service plainly. Absence needs wider investigation; one omission does not expose a single cause.

That distinction is the value of the five-minute check. It turns “ChatGPT does not like us” into a finding another person can inspect and act on.

Keep the original note when you check again

Repeat the same core prompts after a meaningful correction and then on a schedule you can maintain. Keep the old answers. Memory tends to preserve the business names and lose the wording, sources, and conditions that made the comparison useful.

A later mention does not prove your edit caused it. It tells you what changed between dated observations. The update timetable is not fixed, so evidence should be reported with that limit attached.

If you would prefer someone to prepare and explain this baseline, EasyFound can do the initial check manually. Request a free visibility check and the result will be written up by a person, with the prompts and limits included.