How to Test Your Business Across Five AI Assistants in One Afternoon

You can find out where you stand with every assistant your buyers use in a single afternoon, for nothing, using a notebook and a browser. Most people who try this get an unusable result, and always for the same four reasons.

The work itself is not hard. The discipline is: the same ten questions, the same wording, a session that has never met you, and a record of who was named instead of you.

This is the whole procedure, timed, with the mistakes marked. By the end of it you will have a dated sheet you can compare against next month, which is considerably more than most businesses in your market have.

How do you test your business across five AI assistants?

Write ten buyer questions in fixed wording. Open a logged-out session on ChatGPT, Perplexity, Google AI Overviews, Gemini, and Claude. Ask all ten on each one, unchanged. Record every business name that comes back, including the ones that are not yours.

That is fifty answers. It takes about two and a half hours the first time and closer to ninety minutes once your question list exists, because writing the questions is most of the work you only do once.

The output is not a score. It is a sheet with a date on it, showing which questions name you, which engines never do, and which competitor turns up most often. Everything useful comes from comparing that sheet to the next one.

What to settle before you start

Three decisions, and all three are easy to get wrong in a way that quietly ruins the afternoon.

First, the questions. Ten, phrased the way a buyer would type them, with your business name absent from every one. "Who can help me get back into running after a knee injury" is a buyer question. "Best running coaches" is a survey question, and "is [your name] any good" is a vanity question. The twenty prompts worth trying is a reasonable place to draw from.

Second, the session. Logged out, or a private window. OpenAI's own explanation of what ChatGPT is describes a system that works from a conversation, and a conversation it has had with you for six months is not the conversation your buyer is having.

Third, the sheet. Ten rows for the questions, five columns for the engines, one date at the top. Each cell gets one of three marks and a note of who else was named. Nothing more elaborate than that survives contact with the second run.

The afternoon, block by block

Five blocks, about two and a half hours. The bars below are planned time rather than measured time, and the run itself is the block that stretches if an engine is slow.

ONE AFTERNOON, FIVE BLOCKSPLANNED TIME, NOT MEASURED

Write the ten questions 20 min In your buyers wording, saved once

Set up the sheet 10 min Ten rows, five columns, one date

Run all five engines 75 min Fifteen minutes each, wording unchanged

Mark every answer 20 min Named, not named, or named a rival

Read the pattern 25 min Which questions and which engines

AnswerHalo answerhalo.com

CLICK TO ENLARGE

In text, because nothing may live only inside a figure. Write the ten questions, 20 minutes, in your buyer's wording and saved once. Set up the sheet, 10 minutes, ten rows by five columns with one date. Run all five engines, 75 minutes, roughly fifteen minutes each with the wording unchanged. Mark every answer, 20 minutes, as named, not named, or named a rival. Read the pattern, 25 minutes, looking at which questions and which engines. That is 150 minutes, and the durations are a plan rather than a measurement.

The five engines, one at a time

They behave differently enough that treating them as one thing is the second most common way to waste the afternoon. What to expect from each, and what is worth writing down.

Google AI Overviews deserves the extra note. It is not the same product as Gemini, and covering one does not cover the other. Google's documentation on AI features in search treats it as a search surface, which is why it reaches people who would never describe themselves as using AI at all.

Perplexity is the one that teaches you the most per minute, because the citations tell you which pages produced the answer, and the way it chooses those sources is visible in a way the others are not.

Four mistakes that waste the run

One: asking about yourself by name. It feels like the obvious test and it answers a question no buyer is asking. Somebody who types your name has already found you. The run exists to find out what happens when they have not.

Two: improvising the wording. If question four comes out slightly differently on Gemini than on ChatGPT, you have not compared five engines, you have run five separate tests. Copy and paste, every time.

Three: asking once and believing it. The same question asked twice can produce different names. That is normal behavior, not a fault, and it means a single answer is an anecdote. Ask the ones that matter twice and record both, which is the same discipline behind the fifteen-minute check.

Four: recording only yourself. A sheet full of "not named" tells you that you have a problem. A sheet that also lists who was named instead tells you what those businesses have that you do not, which is the only part of the exercise that leads anywhere.

Then date the sheet, save it somewhere you will find it, and put a note in your calendar for a month's time. The second run is where the value is. The first one is just the photograph you will compare it against.

Our free check runs twenty-five buyer questions from your niche at $0 and emails a plain-English readout within the hour, which is the same idea done for you at four times the sample size. It runs as soon as you ask for it.