26 October 2026
Why one question isn't enough
- The answer changes. The same question, asked twice, can name different businesses. One answer is luck, good or bad.
- Your ChatGPT isn't your customer's. Memory, past chats and location change what you see. A business owner who has asked about their own business many times gets a different answer from a new customer.
- Customers ask dozens of questions. Not just “best gym in Bristol”, but “gym open at 6am near the centre” or “gym with a pool for kids”. A brand can be in some answers and missing from others.
- There's more than one assistant. ChatGPT, Perplexity, Gemini and Claude read different sources and often disagree.
Doing it by hand, properly
- Use a clean session: a temporary chat, or logged out, so memory and past chats don't count.
- Write 10 questions a new customer would ask, without the brand's name, in the city and in the customers' language. These are discovery questions.
- Add 2 brand questions, with the name: “What do you know about [brand] in [city]?”. They show whether AI knows the brand and gets the facts right.
- Ask every question twice, on ChatGPT and on Perplexity at least.
- Write down the position of the brand and of the competitors in each answer.
- Repeat next month with exactly the same questions.
Turning answers into a score
Give each discovery answer points by position: 3 for first, 2 for second or third, 1 for fourth or lower, 0 if the brand isn't named. Add up the points, divide by the maximum (3 per answer) and you get a score from 0 to 100. Score the competitors the same way, in the same answers, and the numbers compare directly. It's the method in our methodology.
The margin of error
Because the answers vary, every score has a margin. With 10 questions asked twice, a move from 29 to 22 can easily be noise: if the margin is ±12, nothing really changed. Before celebrating a rise or worrying about a drop, check that the change is bigger than the margin. More questions and more repeats make the margin smaller.
When to use a tool
By hand, one brand on two assistants means 40 or more answers a month to ask, read and count. For several brands, or every month, a tool does it more consistently. What to look for:
- The same questions every time, so the months compare.
- The assistants' official APIs with web search on, set to the right country, with no history: the answer a new customer gets.
- Repeats and a margin of error, not a single yes or no.
- Competitors scored in the same answers, and the reasons AI gives for them.
- The sources AI cites, which show where to act.
Chatlisted does this for local businesses and agencies, with the questions written for the city and the type of business. Tools such as Otterly and Peec track prompts you choose, across countries; the checker page compares them side by side.
Try it on one brand
The free check asks ChatGPT and Perplexity the questions a business's customers ask, scores the answers and shows who gets recommended instead. It takes about a minute.