AI answer engines recommend a handful of businesses per question. Finding out whether you are one of them is a measurement, and like any measurement it is worth doing properly rather than typing your own name into ChatGPT once and drawing a conclusion.
Here is a method.
Step 1: write the questions a customer would actually ask
Not your keywords. The sentence a person says out loud.
"Best emergency plumber in Tacoma" rather than "plumber tacoma". "Who can fix a cracked iPad screen same day near me" rather than "ipad repair". Answer engines take conversational questions, and the phrasing changes the answer.
Write five to ten. Cover the three shapes of intent:
- Urgent — "open now", "same day", "emergency"
- Comparative — "best", "most reliable", "cheapest"
- Specific — the service, the brand, the niche you actually want
Step 2: understand why one check proves nothing
Two things make a single check misleading.
Location. For anything local, the answer changes across a town. A business that is the obvious recommendation from one neighbourhood is absent three miles away. Checking from your own office — the one place you already know you are known — is the least informative place to stand.
Variation. Ask the same question twice and you may get two slightly different lists. Generative answers are not deterministic. A single sample cannot distinguish a real absence from a coin flip.
The fix for both is the same: sample a grid of points around your service area, repeat on a schedule, and look at the pattern rather than any one answer.
Step 3: read three things, not one
When the results are in front of you, there are three separate findings and they lead to different work.
Are you named? The binary, per point on the map. The pattern matters: a clean boundary usually means proximity, a scatter usually means the sources are thin.
Who is named instead? This is the most useful output, and the most ignored. The competitors an AI names are often not the ones you think about — they are the ones the web describes most clearly. Share of voice, measured over a grid, tells you who owns the answer today.
What does it say about you? When you are named, read the sentence. It is usually assembled from your own pages and profiles, and it is the fastest way to find out that the internet believes something outdated about your hours, services or location.
Step 4: fix sources, then re-measure
There is no ranking dial to turn. What answers are built from is what the web says about you — your own pages, your business profiles, directories, reviews, and anywhere your services are described in text a model can read.
So the work is ordinary: make the facts consistent, make the service descriptions specific, and make sure the pages that answer the question exist. Then measure again. The number worth watching is the change, not the level.
Step 5: set a cadence
Monthly is enough for most businesses; weekly if you are actively working on it. What matters is that it is the same questions from the same points each time, otherwise you are comparing two different experiments.
Marketing OS tracks this on a schedule across five surfaces — ChatGPT, Gemini, Google AI Overviews, AI Mode and Grok — and keeps the history so the trend and the share of voice are there without anybody re-running anything.
Start free
The AI visibility checker runs a real grid scan for one keyword on one surface with no account and no email. It is a genuine scan rather than a preview, which is why it is bounded — but it is enough to answer the first question honestly: when somebody near you asks, whose name comes back?