Prompt volumes: how often is a question asked to AI?

Prompt volumes are the number of times a particular question (prompt) is asked to an AI assistant. It is the AI version of the search volume you know from Google. The honest answer: reliable prompt volumes do not exist yet. ChatGPT, Gemini, Perplexity and Claude do not share those figures. What you see are estimates. You can still work out which questions matter for your business, even without exact numbers.

What exactly are prompt volumes?

With Google, you have been able to look up how often a keyword is searched for years. For example "plumber Zwolle". Marketers use that to decide where to focus.

AI assistants work differently. People do not type loose keywords, they type whole questions. For example: "My boiler is leaking and I live in Zwolle. Which plumber can come today?" A question like that is almost always unique. The same topic is asked in thousands of ways. That makes "how often is this prompt asked" a tricky question, even if you had all the data.

Why are there no reliable figures (yet)?

There are three reasons:

• The AI companies do not share it. OpenAI, Google, Anthropic and Perplexity do not publish search volumes per question. There is no public source like the one that exists for Google keywords.
• Questions are long and personal. Two people with the same need phrase their question very differently. Counting exact sentences tells you little.
• Conversations continue. Someone asks a question, gets an answer and asks a follow-up. The question that leads to a recommendation is often the third or fourth in the conversation.

So where do tools get their estimates?

Some tools do show prompt volumes. These are usually based on estimates, for example:

• Panels. A group of users agrees to have their usage measured. That is then extrapolated to all users.
• Derived from search volumes. The search volume of a Google keyword is translated into an estimate for AI questions on the same topic.
• Models and assumptions. A mix of sources with their own calculation rules.

Figures like these can help you tell big topics from small ones. Treat them as a direction, not as fact. For local and niche questions in particular, the estimates are often uncertain. That is why HelloGeoLabs does not show prompt volumes. We measure what AI answers to the questions you care about.

How to choose the right questions anyway

You do not need exact numbers to choose good prompts. Use these sources:

1. Google Search Console. Under Performance, look at queries of several words and questions with "how", "what", "which" or "best". Long queries resemble AI questions. They show what people really want to know.
2. Bing Webmaster Tools. You can see queries here too. Useful, because Microsoft Copilot relies on Bing.
3. Your own customers. Which questions do you get by phone, email, chat and in sales conversations? Those are exactly the questions people also ask ChatGPT.
4. Your sales team or front desk. Ask which comparisons customers make. "Which is better, A or B?" is a typical AI question.
5. Google suggestions. The questions under "People also ask" give a good picture of common variations.

Our article on which questions customers ask AI has more examples.

From loose questions to a fixed set of prompts

Because every question is phrased differently, it works best to think in topics. For each topic, pick one or two questions that sum it up well. An online shop selling garden furniture could choose this set, for example:

• "Which online shop has good sustainable garden furniture?"
• "Where can I buy a lounge set that can stand the rain?"
• "What is a reliable online shop for garden furniture in the Netherlands?"

Also vary who is asking. A first-time buyer asks something different from an experienced one. Read more in prompts with personas.

Prioritising without figures

Give each question a score on three points:

• Value. Does this question lead to a customer or a purchase?
• Frequency based on your own experience. Do you hear this question often?
• Opportunity. Are you already mentioned, or does AI mostly name competitors?

Start with high-value questions where you are not mentioned yet. That is where the biggest gains are.

Frequently asked questions

Can I look up prompt volumes for free somewhere?
No, there is no official public source. Use Google search volumes as a rough indication for a topic, not for an exact question.

Should I track lots of questions?
A small, well-chosen set is better than hundreds of random questions. Ten to twenty questions that really fit your customers tell you more.

Why do I get something different to the same question every time?
AI answers vary each time and per model. That is why it makes sense to measure regularly. See why AI answers differ.

Start with the questions that really matter

Want to know how AI currently answers your customers' questions? With the free AI check we ask five customer questions to ChatGPT, Gemini, Perplexity and Claude. Want to follow your own set of prompts every week? Take a look at GEO tracking. Read more about measuring in how to measure whether AI mentions your business.