How to build a prompt set that actually matters

One of the easiest ways to make AI visibility tracking useless is to track the wrong prompts.

It's tempting to create 500 questions, put them into a dashboard and feel like you're monitoring everything.

You're probably not.

A smaller set of commercially relevant prompts will usually tell you far more than hundreds of random questions.

Don't just convert keywords into prompts

Traditional keyword research is a useful starting point.

But AI search behaviour is different.

People can describe their situation in detail.

Instead of:

email marketing software

someone can ask:

What's the best email marketing platform for an ecommerce company with 20,000 subscribers that needs good automation but isn't as expensive as Klaviyo?

That prompt contains a category, business type, requirement, price concern and competitor.

Those details matter.

Start with customer intent

We normally think about prompt tracking across several groups.

Category prompts

These establish whether AI understands that you belong in your market.

Examples:

What are the best email verification tools?

Which companies offer AI visibility monitoring?

Problem prompts

These start with the customer's problem rather than your product.

How do I reduce email bounce rates?

How can I see whether my company appears in ChatGPT?

These can be extremely valuable because the customer may not know what type of product they need yet.

Comparison prompts

These reveal which brands AI places together.

Compare HubSpot and Salesforce for a small business.

What are the best alternatives to [competitor]?

Requirement prompts

These are more specific and often closer to purchase intent.

What's the best CRM for a recruitment company with 20 employees?

Which AI visibility tool lets me monitor multiple clients?

Price prompts

What are the cheapest AI visibility tracking tools?

Is there a free alternative to [competitor]?

Recommendation prompts

Which platform would you choose for a small agency?

These can reveal particularly interesting brand preferences.

Add variations, but don't go mad

Two prompts that look similar can produce different answers.

For example:

Best CRM software

and:

Which CRM would you recommend for a small UK sales team?

aren't really the same query.

But you also don't need 50 tiny variations of every question.

The aim is to represent meaningful differences in customer intent.

Include non-branded prompts

This is essential.

If all your prompts contain your brand name, you're mostly testing whether AI knows who you are.

That's useful, but it doesn't tell you whether AI discovers you.

A strong prompt set should include questions where the AI has absolutely no obligation to mention your company.

Those are often the most commercially valuable prompts.

Include competitors

Competitor prompts can tell you how the market is being framed.

Track questions such as:

What are the best alternatives to [competitor]?

[Your brand] vs [competitor]

Is there a cheaper alternative to [competitor]?

This helps uncover companies that AI considers similar to you — including competitors you may not have been actively monitoring.

Think about your actual customers

A generic prompt might be:

What's the best project management software?

But perhaps your customers are mostly marketing agencies.

Then this matters more:

What's the best project management platform for a marketing agency managing client work?

Your prompt set should reflect the market you actually want to win.

Organise prompts into groups

Once you have more than a few dozen prompts, tagging becomes useful.

You might use:

• Awareness

• Commercial

• Comparison

• Competitor

• Industry

• Feature

• Problem

• Pricing

Now instead of asking:

What's our AI visibility?

you can ask:

Why is our visibility strong for comparison prompts but weak for industry-specific recommendations?

That's a much more actionable question.

Don't treat your prompt set as finished

Your first prompt set will not be your best prompt set.

You'll discover new questions through:

• customer conversations

• sales calls

• Search Console

• paid search data

• competitor research

• AI answers

• query fan-out

• Reddit and forums

• internal site search

Add useful prompts.

Remove meaningless ones.

Your prompt library should become a living representation of how people research your market.

Because ultimately, tracking more prompts isn't the objective.

Tracking the right questions is.