In practice · 9 min read

What questions do customers ask AI? How to find out

You cannot see what people ask ChatGPT. You can still build a reliable list of the questions your customers ask. These are the six sources we use for it.

By NXEO · 5 Oct 2026
In short
  • There is no overview of what people ask AI. The best source is you: your phone, your email and your quotes.
  • A question to AI is longer and more personal than a search. Write them down in whole sentences.
  • Thirty good questions are enough to keep you going for a year.

With a search engine you can look up what people search for. There are programs that show how often a word is typed per month. For AI assistants that does not exist. OpenAI, Google and Anthropic do not share which questions are asked.

That looks like a problem. What questions do customers ask AI, if nobody can tell you?

It is not so bad. The questions people put to an AI assistant are the same questions they used to put to you. On the phone, in an email, at the first appointment. They just ask them earlier now, and of someone else. So you have already heard them.

This article lists the six places where you find them, and how to turn them into a list you can work with.

Before you start collecting, it is good to know what you are looking for. A question to AI differs in three ways from what people type into a search bar.

It is longer. In a search engine someone types "dormer cost". To an AI assistant they ask: "What does a dormer of three metres wide cost on a terraced house, and do I need a permit for it?"

It contains a situation. People say who they are and what is going on. "I have a limited company with two employees and want to change bookkeeper." The system uses that information to give a fitting answer.

Follow up questions come. It does not stop at one question. After the answer about cost comes: "And what if my roof is at the front?" And then: "Who does that near Apeldoorn?"

That last one is the most interesting for you. The request for a recommendation often comes third or fourth in a conversation. By then the system knows a lot about the situation, and it chooses a company that fits.

For your list that means: write questions down in whole sentences, with the situation included, the way a person would ask them.

Source 1: your phone

The best source there is. Every new customer who calls asks two or three questions in the first minute. Those are the questions they could not get answered before calling.

Keep a notepad next to the phone for a week. For each call with a new customer, note the first question as literally as possible. Not "asked about price" but "what does it cost to renovate a bathroom of five square metres if the pipework has to stay?"

If colleagues answer the phone, ask them to do the same. After a week you have twenty questions. Half of them occur several times. Those are your most important ones.

Source 2: your email and your contact form

Search your mailbox for the first messages from customers over the past year. The question is there in their own words, including the situation.

Look out for sentences that start with "I was wondering whether" and "Can you also". Those are questions to which the answer evidently could not be found on your website. Each such question is a page or a paragraph that is missing.

Also look at what people enter in the free text field of your contact form. That is often where the most honest questions are.

Source 3: the first appointment

What do you explain again with every new customer? Every business owner has a story they tell ten times a month. How the process runs. Why one material is more expensive than another. What is and is not included in the price.

That story consists of answers to questions. Write them down. These are often the questions of people who are already seriously considering working with you, and so the most valuable.

Think of the objections too. "Is that not far too expensive?" "Could I not do that myself?" "What if it disappoints?" People put those questions to an AI assistant precisely because they do not dare put them to you.

Source 4: Google itself

Google shows in three places what people ask.

The suggestions in the search bar. Start typing a question, for example "what does a dormer cost", and see what Google fills in. Those are searches that occur often.

The block with questions from others. For many searches there is a row of questions people also ask, halfway down the page. Open one and new ones appear. In five minutes you have twenty questions about your trade.

Search Console. If you have set that up, you can see for which searches your website was shown. Filter on searches with words like how, what, why, cost and difference. You then see questions you are already close to.

These questions are shorter than what people ask AI. Use them as a topic and then turn them into a whole sentence.

Source 5: the AI assistant itself

You can ask an AI assistant which questions people ask. That sounds circular, but it works if you go about it properly.

Do not ask "which questions do people ask about dormers". Then you get a general list. Ask: "I own a house and I am considering a dormer. Which questions should I ask before I choose a company?" The answer is a list of a customer's considerations.

A second way: ask an opening question and see which follow up questions the system suggests. Many assistants make suggestions at the bottom of an answer. Those are the paths users often follow.

Use this as a supplement, not as a base. The system knows what people ask in general. You know what your customers ask.

Source 6: places where people ask each other for advice

Forums, groups on social media for a neighbourhood or a hobby, the questions under a product page, the comments under a video. Wherever people ask each other for advice, there are questions in ordinary language.

It pays to look at the questions that get many replies. Those are topics without a clear answer. Whoever does give that answer has something others do not.

From a pile to a list

After a week of collecting you have sixty to a hundred questions. Many duplicates, many half formed. This is how you turn them into a usable list.

Step 1: merge

"What does a dormer cost?", "How expensive is a dormer?" and "Price of fitting a dormer" are the same question. Choose the wording that sounds most like how a person would say it.

Step 2: sort by moment

Put one of three labels on each question.

  • Finding their bearings. The customer wants to understand something. "What is the difference between a prefabricated and a traditional dormer?"
  • Comparing. The customer wants to choose how. "What does a dormer cost and what does that depend on?"
  • Choosing. The customer wants to choose who. "Which company in Apeldoorn fits dormers within one day?"

A good list has a number of all three. Most people have too many of the first kind and too few of the last two.

Step 3: add the situation

Turn each question into a question with a person in it. Not "dormer cost" but "what does a dormer cost for a terraced house from the eighties". Consider which situations occur often among your customers and make a variation per situation.

Step 4: choose thirty

You do not need more for the time being. Thirty questions are enough to write a page every two weeks for a year. And they are enough to measure reliably whether you are mentioned.

Choose them like this:

  • ten that are followed directly by an enquiry;
  • ten that customers are stuck on just before the choice;
  • ten that show you know your trade.

What to do with the list

The list has two functions.

It is your yardstick. Put the questions to ChatGPT, Gemini and Claude and note whether you are mentioned. That way you know where you stand. The method is in is my company mentioned in AI.

It is your writing plan. Every question for which you are not mentioned and for which the answer is not on your website is a page you can write. Start with the questions labelled choosing and comparing. How to build such an answer is covered in an FAQ page AI can quote.

Mistakes we often see

Inventing questions in trade language. You call it a hybrid system with a buffer tank. Your customer calls it "one of those things that hangs next to the boiler". Write the question down the way the customer asks it. Use the trade term in the answer.

Only choosing questions where you come out well. A list of questions so specific that only you can win gives a nice figure and no insight. Include the broad questions too, where you are up against big names.

Skipping the price question. Nearly every customer asks about cost. Almost no business owner wants to write it down. That makes it the question that is easiest to win. A range with an explanation is enough.

Never adjusting the list again. Questions change. A new scheme, a new product, a price rise: questions are added and others drop off. Go through your list every quarter.

Avoiding negative questions. "What are the disadvantages of uPVC window frames?" Your customer asks it anyway. If you do not give the answer, someone else does, and they sell wood.

An example of a list

What does that look like for a company that designs and builds gardens in the Arnhem region? Ten of the thirty, spread across the three moments.

  1. What does it cost to redo a garden of eighty square metres?
  2. How long does building a garden take and in which season is that best done?
  3. What is cheaper to maintain: grass, gravel or paving?
  4. Which plants do well in a shaded garden facing north?
  5. Do I need a permit for a canopy in my back garden?
  6. What should I look for when choosing a landscaper?
  7. What is normally included in a quote for garden construction and what is not?
  8. Which landscaper in Arnhem builds low maintenance gardens?
  9. Who near Velp can design a garden and also build it?
  10. Which landscaping company in Arnhem also does the yearly maintenance?

You answer questions 1 to 5 with explanation. Questions 6 and 7 show that you are honest. Questions 8 to 10 are the questions for which you want to be mentioned, and that only works if the answers to the first seven are on your website.

How to keep the list alive

A list written once and filed away loses its value within a year. A light routine keeps it sharp.

After every unusual call, add a line. If a customer asks something you had not heard before, write it down the same day. One new question per month is plenty.

Each quarter, strike what no longer applies. A scheme that has ended, a product you no longer carry. A shorter list that is right beats a long one that is half outdated.

Once a year, check the wording. Customers adopt new words. What used to be called one thing is suddenly called another, often because of a change in rules or a product that became popular. Your headings should move with them.

Share it with whoever answers the phone. They hear the questions first. A shared document that anyone can add to works better than a list only you maintain.

How we do it

With every new client it starts with this. We draw up a list from the onboarding, the existing website, what we know about the sector and what the systems themselves show. We then put that list to the AI systems daily, for the client and for their competitors.

The list remains the client's, even if they stop. We think that is logical: they are their customers and their questions.

Would you like to see which questions you are losing now? The free baseline check lists them. If you want to know how it continues after that, read what NXEO is or see the frequently asked questions.

Curious how AI mentions your company today?

Enter your website. Within 48 hours the report is in your inbox, free.

Read next

Where can we send the report?

In your inbox within 48 hours. No newsletter, no spam.

Website