Does ChatGPT Recommend Your Business? How to Check in 20 Minutes

No tools, no logins. The right questions, three assistants, and a scorecard that keeps one lucky answer from misleading you.

7 September 2026 · 8 min read

Most owners find out what ChatGPT says about their business by accident, from a customer who mentions it in passing. The rest ask it once, get one answer, and draw a conclusion from it. Both are mistakes. Checking properly is free and takes about twenty minutes, but it has a method, and skipping the method is how one lucky answer convinces you everything is fine.

The method in one line. Ask the questions your customers ask, in their words, across three assistants, score every answer on a simple grid, and repeat it a few days later before you conclude anything.

Step 1: Ask like a customer, not like the owner

The first instinct is to type your own business name. That only tests whether the assistant has heard of you, and the customer who already knows your name was never the one you were losing. The customers at stake describe a need, so that is what you type.

Write five to eight questions in a stranger's words. For an HVAC company in Mesa, the list might be:

“Best HVAC company in Mesa for a broken AC”
“Who should I call for AC repair in Mesa?”
“Affordable air conditioning repair near Mesa AZ”
“HVAC company open weekends in the East Valley”
“Recommend someone to replace a heat pump in Mesa”

Vary the wording on purpose: “best”, “cheap”, “near me”, the city, the wider area, a specific job. Different phrasings pull different answers, and the spread is part of what you are measuring. Stuck for wording? Your Google reviews are customers describing the job in their own words.

Step 2: Use three assistants, not one

ChatGPT, Claude, and Perplexity build answers from different sources and routinely disagree. Being named by one and invisible to another is normal, and knowing which is which tells you where the work is. Ask each assistant the same questions, in a fresh chat every time, so an earlier answer cannot color a later one. If the assistant offers to search the web as it answers, let it: live answers reflect what the current web says about you, answers from memory reflect the older web, and the difference is itself a finding.

Step 3: Keep score in writing

Draw a grid: one row per question, one column per assistant. In each cell record three things. Were you named at all? Who was named instead, by actual business name, because the rivals that recur across questions and assistants are the ones worth studying? And if you were named, was the description right? Assistants sometimes carry an old address, a dropped service, or hours from two owners ago, and a wrong mention loses the customer as surely as no mention.

The grid is the method: the pattern across fifteen or twenty cells is a measurement, where any single answer is an anecdote.

Step 4: Run it again in a few days

Assistants vary their answers between runs, and their sources shift over weeks. Repeat the grid before you conclude anything. Showing up in one run out of three is a real finding in itself: you are on the edge of the answer, close enough that removing the right barrier could make you a regular.

Reading your results

Named consistently, described correctly. You are in good shape today. Put a repeat check on the calendar for next quarter, because this shifts.
Named by one assistant, missed by the others. They can see you but do not trust you enough to name you every time, which usually points to thin or inconsistent facts about your business across the web.
Named with wrong details. Track the wrong fact to its source, an old directory listing, a stale Google profile, or your own site, and fix it there.
Never named, same rivals every time. The most common result. Something is keeping assistants from reading or trusting your business, and the recurring names show you what winning looks like in your market.

What this test cannot tell you

The grid shows the outcome, never the cause. From the outside you cannot tell whether assistants skip you because they cannot enter your site, because your pages arrive blank to machines, because the facts in your code contradict your Google profile, or because nobody else on the web mentions you. Each cause has a different fix, and guessing wrong wastes months. That diagnosis is what an audit is for; we lay out the whole of it in what is an AI visibility audit.

Two fast next steps. The free check reads your homepage the way AI tools do and flags the access and readability barriers in minutes. And if you would like to see a finished answer before you run your own grid, our industry pages each quote a real one for trades from HVAC to physical therapy.