The three mistakes that ruin the result
- Asking in your own logged-in account. Memory and chat history prime the model with your business, and it will name you because you told it to. Use a temporary chat, an incognito window, or a signed-out session.
- Asking "do you know my business" instead of asking what a customer asks. Assistants are agreeable. Ask the buying question, not the identity question.
- Asking once. Answers vary between runs and between assistants. One run is an anecdote; six prompts across two assistants is a measurement.
What you are looking for
Three things, in this order. First, are you named at all. Second, are you named for the jobs that actually make you money, or only for the generic category question. Third, is anything the assistant says about you wrong, because a wrong fact is both the most damaging outcome and the fastest one to fix.
Write the answers down verbatim with today's date. This sounds fussy and it is the entire value of the exercise: without a baseline you will never be able to tell whether the work you do next changed anything.
How often to repeat it
Monthly is enough to notice a trend by hand. Weekly is better if you are actively working on it, because you want the change to be attributable to something you did rather than to a model update. Doing it once and never again is the most common outcome and produces nothing.
What a bad result does not mean
Being unnamed is not a judgment on your business, and it is not permanent. It means the pages the assistant retrieved did not describe you. That is a coverage problem, and coverage is fixable in a way that most marketing problems are not.
The two-minute protocol
Open a temporary or signed-out chat. Paste each prompt, swapping in your trade, city and business name. Then repeat the whole set in a second assistant.
6 prompts, in order
best [trade] in [city]I need [specific job] in [city] this week. Who should I call?top rated [trade] near [neighborhood]what do you know about [your business name] in [city]is [your business name] reliable, and what do people say about themwho are the best alternatives to [your business name] in [city]
- Score prompts one to three out of three: how many named your business anywhere in the answer?
- Score the same three again for position: were you first, or fourth and later? Being named last still counts, but it converts differently.
- For prompt four, mark every fact that is wrong. Hours, services, service area, ownership, whether you are still open.
- For prompt five, note the tone and where it says it got that impression.
- For prompt six, note whether it can produce a confident list of your competitors. If it can name them fluently but struggles to describe you, the gap is coverage, not category.
- Repeat all six in a second assistant and compare. Disagreement between assistants is normal and tells you the underlying sources disagree.
Keep the wording of these six prompts fixed forever. The moment you reword them, this month stops being comparable to last month, which is the only thing that makes the exercise worth doing twice.
You are looking for the term
AI visibility scan
What you just ran by hand is a small version of what the industry calls an AI visibility scan, and the fraction of prompts where you were named is your mention rate. Running it across a hundred questions rather than six, on a schedule, is the only real difference between the manual version and the automated one.
- Why does the answer change every time I ask?
- Assistants sample from a distribution and retrieve slightly different pages each run. That is why a fixed set of prompts repeated on a schedule is the only sound way to read a trend.
- Does asking about my own business teach the model about it?
- No. Your chat does not update the model for anyone else, and a single conversation does not change what other people are told.
- 01Web search toolOpenAI
- 02Web search toolAnthropic
- 03AI features and your websiteGoogle Search Central
- 04GEO: Generative Engine OptimizationAggarwal et al., arXiv
Every link above was opened and confirmed reachable on September 1, 2026.
The automated version
See what ChatGPT, Claude, Gemini and Perplexity say about your business
The free scan runs the buyer questions for your trade and city across all four assistants, shows who gets named instead of you, and lists the sources those answers came from. No signup, about thirty seconds.
Run my free scan