How do I check if ChatGPT recommends my business?

• Write ten questions a buyer would actually type: 'best dentist in Parramatta', 'emergency plumber near Fitzroy', 'who should I use for conveyancing in Hobart'. Never include your business name.
• Open each engine logged out, or in a fresh chat with memory off. History quietly personalises the answer towards what you have already discussed.
• Ask one question per session and paste the answer somewhere you can count it. Also copy the sources the engine cites.
• Tally two numbers: how many answers named you, and which other businesses were named, how often.
• Repeat the identical set a few weeks later. The comparison is the point; a single run is a snapshot of a moving thing.

This is genuinely useful and costs nothing but time. The catch is the time: ten questions across four engines is forty sessions, and forty answers is a small enough sample that one lucky mention looks like visibility.

• Asking about yourself by name. 'What do you know about [business]?' invites the model to describe an entity it was handed. It says nothing about whether you get named unprompted, which is the only thing a buyer's question tests.
• Testing while logged in. Chat history is the single fastest way to convince yourself you are visible when you are not.
• Treating one answer as the result. Ask the same question twice and you may get two different lists. Without a denominator there is no finding, only an anecdote.

A Grascada report is the same exercise run properly and written down: 30 buyer questions spread across intent and funnel stage, put to all four engines from clean sessions — 120 recorded answers. Every answer is stored verbatim with its timestamp and cited sources.

• Your mention rate out of all 120 answers, per engine and overall.
• The ranked list of businesses named in your place, with trading names, legal names and branch variants merged so nobody is counted twice.
• Every source domain the engines cited, and which of those you can legitimately claim or influence.
• An engine-confidence score showing how strongly the four engines agree — including when they unanimously name nobody.
• An 11-point readiness check of your own site: robots.txt and whether it blocks assistant crawlers, llms.txt, structured data, opening hours, reviews, FAQ schema, contact details, Open Graph and more, each with the specific change to make.
• The raw answers and citations as CSV, so every figure can be checked against its evidence.

It is one payment, one report, and there is a complete released run free to read below — nothing withheld, no sign-in.

Nobody can see what an individual customer was told, because their answer depends on their account, device and moment. Nobody can guarantee an AI position, because nobody outside the model providers controls the output. A measurement tells you the pattern and the evidence behind it — which is enough to act on, and all anyone honestly has.