What questions your customers ask AI (and which ones to measure)
By Toni Moral · 13 November 2026 · 5 min read · Leer en español
In a search engine, your customer types three words: "invoicing software small business". In an AI assistant they write a whole sentence, with their situation: "we're 20 people in Manchester, which invoicing tool would you recommend that connects to our bank?".
That difference matters. If you want to know whether AI recommends you, you first need to know what it's being asked. And not every question is equally useful.
Your customers ask at three moments
Before choosing a provider, a customer usually goes through three moments. Each one produces a different kind of question:
- Discover. They don't know yet what they need. "How can I stop doing invoices by hand?".
- Compare. They know what kind of solution they want and want to understand the options. "What's the difference between invoicing software and an ERP?".
- Decide. They're about to choose. "Which invoicing tool would you recommend for a 20-person company?".
Decision questions are the closest to a sale. But discovery and comparison questions count too: they shape the idea the customer has of the market.
AI doesn't name companies in every question
This is what surprised us most in our own measurement. In the latest one for yovara.io, on 4 November, we asked 26 questions without our name. In 20 of them, no company appeared in even half of the answers. AI mostly answered with general advice.
And the other six? Four asked directly for a tool, or for a way to do something with tools: "what tools exist to…?", "is there a system without a subscription to…?", "what's the best way to monitor…?". Another asked how much an analysis costs, and AI answered with providers and their prices. The sixth named a reviews directory that isn't even in our category.
The lesson is simple. When the customer asks for a solution or a price, AI names companies. When they ask how to do something, it usually explains the method without naming anyone. That's why it pays to measure both kinds: some questions tell you whether AI puts you on its list; the others, whether you're the one teaching it the method.
Ask like a customer, not like an SEO expert
The most common mistake when choosing questions is writing them as keywords. "Best invoicing software" is not something anyone says to an assistant.
A few rules help:
- Write full sentences. With a verb, with context and, if it matters, with the customer's situation: size, city, sector.
- Leave your name out. A question with your name measures whether AI knows you, not whether it recommends you. We explain it in how to check if ChatGPT recommends your company.
- Use your customers' words. If they say "tool" and you say "platform", ask like they do. Sales calls and support emails are a good source.
- Ask in your market's language. If you sell in Spain and the UK, you need questions in both languages. The answers change.
How many questions, and which ones
With 5 questions you can't see a pattern. With 100 you won't know where to start. A good starting point is about 25 questions without your name, spread across the three moments, plus a few with your name to check that AI describes you correctly.
To choose them, think about what you sell, not what you'd like to sell. A question about your main product is worth more than one about something secondary. And leave out the ones that have nothing to do with your business, even if they seem interesting.
Once chosen, don't change them
AI answers change from one time to the next, even if you do nothing. If you also change the questions every time you measure, you won't be able to tell whether an improvement is real.
That's why it pays to fix them. The same questions, asked the same way, in every measurement. If you need new questions, add them as a separate group and compare them only with themselves.
What to do with each question
When you have the results, each question will be in one of three situations: AI names you, names others instead of you, or names nobody. Questions where it names nobody are usually the best opportunity, because you don't have to displace any competitor. And when it names others, it pays to read carefully who they are before drawing conclusions.
How Yovara does it
Yovara reads your website and writes about 30 questions the way a customer would ask them before choosing a provider, grouped by stage: discover, compare, decide. About 25 don't name you and about 5 do. You review and edit them before anything is measured. Then they're frozen, so every new measurement is comparable. Each question is asked 3 times in ChatGPT and 3 in Perplexity. We explain it in the methodology.
The first measurement costs 4 credits and measuring again costs 1, with no subscription: prices are on the pricing page. And before paying anything, you can see the questions we'd suggest with the free preview.
Frequently asked questions
What kind of questions do customers ask AI assistants?
Full sentences describing their situation, not keywords. They usually fall into three kinds: to discover what they need, to compare options and to decide which provider to choose.
In which questions does AI recommend companies?
Mostly when the customer asks for a solution: "which tool would you recommend for…?". When they ask how to do something, AI usually answers with general advice without naming anyone.
How many questions do I need to measure my AI visibility?
About 25 questions without your name is a good starting point, spread across discover, compare and decide. With very few, the result depends too much on each question.
Should I include my company's name in the questions?
In a few, to check that AI describes you correctly. But the number that matters comes from the questions without your name, which measure whether AI recommends you when people don't know you.
