Knowledge
How to find out what AI says about your business
Open ChatGPT, Claude, Gemini and Perplexity, ask each of them the same questions about your category – without mentioning the company name – and make a note of which brands they mention. It takes about thirty minutes and doesn’t require any tools.
Last updated: 7 September 2026 · reading time: approximately 6 minutes
Rules
Before you start
The most common mistake is to type your own company’s name into ChatGPT and draw conclusions from the response. This merely checks whether the model recognises the name. The customer isn’t asking about your company, because they don’t know about it yet. They’re asking about the category.
That is why the test only makes sense if the brand name is not mentioned in the question. In that case, the answer reflects the real-life situation: who the model recommends to someone who is still looking.
Four principles without which a test is meaningless
Do not give the company’s nameA question about a name checks whether the model recognises it, not whether they recommend it. Those are two different things.
Ask questions the way a customer wouldIn the customer’s language, not your own jargon. „Who carries out the repairs”, not „maintenance services in the field of”.
Log out or use incognito modeThe model remembers previous conversations and can suggest a brand you’ve already talked about.
Write down your questions before you startYou’ll repeat the same set in a month’s time. Otherwise, you won’t have anything to compare it with.
Set
Five questions to get you started
Enter your service and town in place of the brackets. Ask each question using the four templates separately, without continuing the conversation.
01
Who provides [service] in [town]?
A basic question about the category. It checks whether you’re even being considered.
What to look out for: do you fall, and where?
02
How do you choose a [service] and what should you look out for?
Advisory question. The model sets out the criteria and often indicates the source from which they were taken.
What to look for: whose website is being cited
03
[Competitor] or [another competitor]? Which is better?
A comparison question without your name. It shows who forms the default pair in the category.
What to look out for: do you come across as the third choice?
04
How much does [the service] cost, and what determines the price?
Transactional enquiry. It very often ends with a quote from someone’s website showing their price list.
What to look for: where the model gets its range from
05
What do you know about [Your brand]?
The only question with a name. Ask it at the end, on its own, and don’t mix up the result with the previous ones.
What to look out for: does the model recognise the brand, and do they get the facts right?
The first four questions measure what the customer sees. The fifth examines something else, namely brand recognition itself, which is why it is kept separate. A brand may be very well known to the respondent and yet not feature in any of the first four questions.
Transcript
How to record results
All you need is a spreadsheet with questions in the rows and models in the columns. It’s important to note down not only whether you’re eliminated, but also who the model names instead of you and where it gets its data from.
A simple form to fill in
Test date: 7 September 2026
| Question | ChatGPT | Claude | Gemini | Perplexity |
|---|---|---|---|---|
| 01 Who does it? | I’m not falling over | 3rd place | I’m not falling over | I’m not falling over |
| 02 How to choose | quotes a competitor | quotes the ranking | no sources | quotes the ranking |
| 03 A or B | I’m not falling over | I’m not falling over | I’m not falling over | I’m not falling over |
| 04 How much does it cost? | a fork from the other side | a fork from the other side | without forks | quotes my website |
| 05 What do you know about the brand? | known, description matches | known, description matches | is misleading the industry | doesn't recognise |
Five questions times four models equals twenty answers. That’s enough to spot a pattern.
Take the test again in a month’s time, using the same questions. A single measurement only tells you how things stand today. It is the second column of dates and the difference between them that is significant.
Please note
Four mistakes that ruin the result
Each of them causes the test to show something different from what it appears to be.
A question about the name of your own companyIt’s about checking your reputation, not a recommendation. A customer who doesn’t know you yet won’t ask that sort of question.
One question per modelThe models respond differently, and the same question can yield slightly different results. One answer is an anecdote.
Following up in the same conversationThe model remembers previous sentences and will suggest a brand that has already been mentioned. Please ask each question in a new conversation.
Changing the questions for the next testA different set of questions yields a different result. You can only compare the same question asked twice.
Border
When a manual test is no longer sufficient
Checking things manually is a good place to start and costs nothing. However, it has a clear limitation, which is easy to spot.
A manual test is sufficient
- Would you like to see what the situation is like, just for once?
- You have one brand and a narrow category
- All you need is an answer like „I’m falling” or „I’m not falling”
- You check once a quarter or less frequently
Manual testing is no longer sufficient
- Do you want to compare results over time, using the same set of data?
- You need to stay ahead of the competition, not just „fall flat on your face”
- Do you have several brands, or do you serve customers?
- You need to show the result to the person responsible for approving the budget
The problem isn’t that it’s labour-intensive, but that it’s repetitive. People unconsciously change the wording, and then the two measurements are no longer comparable. We explain how we tackle this at Aelo in methodology.
FAQ
Questions on assessment
The questions most commonly asked during the first test.
How can I find out what AI is saying about my business?
Ask ChatGPT, Claude, Gemini and Perplexity the same questions about your category, without mentioning the company name, and note down which brands were mentioned and which sources the model referred to. Finally, ask specifically about your brand.
Why shouldn’t I ask for the name of their company?
Because that’s what helps them identify you, not a recommendation. A customer who doesn’t know you yet asks about the category, not your name.
How many questions are enough?
Five questions across four models – that is, twenty answers – are enough to spot a pattern. With such a broad category, it’s worth adding questions about specific services.
Do I need to be logged out?
It’s worth it. The chatbot remembers previous conversations and can suggest a brand that’s already been mentioned. Using private mode or opening a new chat window solves the problem.
What should I do if it turns out I’m not going anywhere?
This is the most common result of the first test. Start with the indicators that allow the model to confirm that the company exists. We describe this in the article when AI doesn’t know your business.
The model is giving false information about my company. What should I do now?
Make a note of exactly what you’re referring to and which model it is. This is usually due to a lack of clear information on your website or to confusion with another company with a similar name.
How often should this test be repeated?
Once a month. There’s no point in doing it more often, as the models need time to detect changes. We discuss this in the article How often should visibility be measured?.
Run the same test automatically
You start with 250 AeloCoins, without a card. That’s enough to start the measurement and test the modules. Once you’ve used them up, you switch to a paid plan. Measurements start straight away, and a full report on your brand is included from the Starter plan onwards. The same set of questions across four models, repeated every month.
