How to choose which prompts to track for AI visibility

Short answer: track the questions real buyers ask an assistant before they choose a product like yours. Group them into intents (one buying need each), write two or three natural phrasings per intent, cover every buyer stage from "I have a problem" to "X or Y?", and keep questions that name your brand in a separate, branded group. Agree the set before you see the results, so nobody picks flattering prompts.

Why does the prompt set matter so much?

An AI visibility number is only ever "how often we were mentioned for these questions". Change the questions and the number changes, often dramatically.

Two prompt sets for the same brand can give opposite stories:

• "Best project management tool for creative agencies" → you appear in most answers.
• "Best project management tool" → you appear in almost none.

Both are true. The question is which one your buyers actually ask. A good prompt set reflects real demand, not the questions you already win.

Start from intents, not keywords

An intent is one thing a buyer is trying to find out, such as "find affordable payroll software for a small UK business". Each intent gets a few phrasings: the different ways someone would type or say it to an assistant.

Why several phrasings? Assistants are sensitive to wording. Measuring only one phrasing tells you about that sentence, not about the need behind it.

Write phrasings the way people talk to assistants: full sentences, with context ("for a 10-person startup", "in the UK", "that works with Xero"). Assistant prompts tend to be longer and more specific than search keywords. "payroll software UK?" is a keyword with a question mark, not a prompt.

Cover every buyer stage

Spread intents across the stages of a buying decision:

1. Problem-aware: "How do small businesses handle payroll without an accountant?"
2. Category: "What types of payroll software are there for small businesses?"
3. Recommendation: "What's the best payroll software for a UK startup?"
4. Comparison: "Brand A vs Brand B for payroll", "Alternatives to Brand A".
5. Branded: "Is [your brand] good for payroll?", "[Your brand] pricing".

Recommendation and comparison intents are usually where visibility affects sales most directly. Problem-aware intents tell you whether you're part of the conversation early.

Keep branded and non-branded separate

If a prompt names your brand, the assistant will almost certainly mention you. Mixing those prompts into your main number inflates it.

• Non-branded intents answer "do we get discovered?" They are your headline visibility.
• Branded intents answer "how are we described when people ask about us?" Track them for accuracy and sentiment, not for discovery. See what to do when AI gets your brand wrong.

Watch for a non-branded intent that quietly names you ("Is there anything like [your brand] but cheaper?"). It belongs in the branded group.

Where do good prompts come from?

• Sales and support conversations. The questions prospects ask before buying are the best source.
• Search Console. Longer, question-style queries that already bring people to your site.
• "People also ask" boxes and related searches for your category.
• Community threads (Reddit, forums, Slack groups) where people ask for recommendations.
• Your own site. The problems your product pages claim to solve suggest the intents you should be visible for.

How many intents should you track?

Enough to cover your main use cases without spreading your sample thin. For most brands, 10 to 30 intents with 2 or 3 phrasings each is a good range. Remember that each intent needs enough answers to give a usable rate. See why one AI answer tells you almost nothing.

Keep it fair and stable

• Agree the set first, then look at results. Choosing prompts after seeing which ones you win is how dashboards end up flattering.
• Make the set visible to everyone who reads the numbers, including clients if you are an agency.
• Change it deliberately. Adding or removing intents changes the headline number. Note the date, and compare like with like.

Frequently asked questions

Should I include competitor names in prompts?

Yes, in comparison intents ("A vs B", "alternatives to A"), because buyers ask them. Keep them separate from general recommendation intents so you can see each clearly.

Should prompts include a location?

If your market is local or national, yes. Assistants often tailor answers to location, and "best accountant" means something different in Leeds and in Lisbon.

How often should I update my prompt set?

Review it every quarter, or when you launch a product or enter a new market. Avoid changing it every week, or you can't compare one month with the next.

Can I use the same prompt set as a competitor?

You can share a set to compare fairly. That's how share of voice works: the same questions, measured the same way, for every brand. See how to measure AI search visibility.