Do you know what ChatGPT says when someone asks about you?

Try it right now: open ChatGPT and ask about your own business by name. Then ask the question a stranger would: "who does emergency roof repair in Denver?" or "is there a good implant dentist near Cherry Creek?" Read both answers slowly. For most owners, this two-minute exercise is somewhere between enlightening and alarming.

Three things can happen. The assistant knows you and represents you well. The assistant knows you and gets things wrong: old address, discontinued service, a specialty you don't have. Or the assistant simply doesn't surface you, and answers with someone else. All three outcomes are quietly shaping revenue, and none of them appear in any analytics dashboard you own.

The answer isn't one answer

Here's what makes this hard to check by hand: there is no single "what AI says about you." ChatGPT, Google's Gemini and Claude each have different training data, different search integrations, different retrieval habits. One may know your business cold while another has never heard of it. And the phrasing matters: "who is Summit Roofing?" and "does Summit Roofing do emergency repairs?" exercise completely different knowledge.

From spot checks to instrument readings

A manual spot check is a start, but it's one engine, one phrasing, one day, and you'll phrase the question kindly, because you know the answer you're hoping for. Monitoring this properly means treating it like instrumentation:

• The same questions every time: general recognition ("who is…?") plus a "do they offer this?" check for each service that matters to you, phrased the way customers phrase things.
• Every engine, every run: ChatGPT, Gemini and Claude each answer the full prompt set, so you see where you're strong and where you're invisible.
• Verdicts, not transcripts: each answer parsed into yes / no / related with confidence, plus the sources the engine actually cited.
• A weekly cadence: because answers shift over weeks, and a trend is worth a hundred screenshots.

That's precisely what AnvixaAI's AI Visibility does: you pick the services worth asking about, and the platform runs the full battery weekly, turning the transcripts into metrics: how much the engines know, whether they cite your site as the source, and how confidently they answer.

What to do with a bad answer

Reading the answer is diagnosis. The treatment is making your business legible enough that engines stop guessing: published structured data on every page, a consistent entity graph, machine-readable facts for every service you want confirmed. That's the rest of the platform's job, and the weekly runs are how you watch the treatment land, engine by engine, week by week.