Choose which prompts to track
By the end of this playbook you will have a prompt set that mirrors how real customers ask about your category, balanced across the funnel and your key personas. Prompts define everything downstream, so getting them right is what makes your Share of Voice and visibility metrics mean something.
This is the strategy, not the mechanics
For how to add, schedule, and edit prompts, see the build a prompt set guide. This page is about deciding which prompts belong in the set in the first place.Prompts define everything downstream
Your prompt set is the population every metric is calculated over. Organic Share of Voice reads non-branded prompts, while branded prompts are used for reputation and accuracy. Choose prompts that flatter you and the numbers will look good while telling you nothing. Choose prompts a real buyer would type into an assistant and the numbers become a map of where you win and lose.
Cover real customer intents and funnel stages
A buyer asks different questions at different stages. Track all of them, because winning the comparison stage matters little if you are absent when someone first asks what tools exist. Spread your set across the funnel rather than clustering at one stage.
A decision table of prompt categories
| Category | Funnel stage | Example | Why track it |
|---|---|---|---|
| Category discovery | Top | "best project management tools for startups" | Where new buyers first meet the category; high reach, hard to win |
| Use-case fit | Middle | "project management tool for a remote design team" | Qualified intent; where a strong fit beats a bigger brand |
| Comparison | Bottom | "Acme vs Beta for sprint planning" | Buyers are deciding; your positioning shows up directly |
| Brand-defining | Bottom | "is Acme good for startups" | Confirms what engines say about you specifically |
| Problem or job | Top to middle | "how to keep a startup team on track" | Captures buyers who do not yet know the category by name |
Segment by persona and region
The right answer differs by who is asking and from where. If you sell to both founders and agency owners, write prompts for each, because engines may name you for one and not the other. If you operate in distinct markets, add region-specific phrasing where it changes the answer. Keep segments as separate, labelled prompts so you can read visibility per persona instead of a blurred average.
How many prompts, and the brand-versus-category balance
Start with ten to twenty prompts. That is enough to cover the funnel and a couple of personas without making each run slow or the trend noisy to read. Grow deliberately as you find gaps, not all at once.
Balance brand-defining prompts against category prompts. Brand prompts ("is Acme good for startups") are easy to win and tell you what engines say about you, but they flatter your numbers. Category prompts ("best project management tools for startups") are harder and where competitors live, but they reflect real demand. A healthy set leans toward category and use-case prompts, with a few brand prompts for signal. If most of your set is brand-defining, your mention rate will look great and your organic visibility read will be too narrow.
Worked example
Acme sells project management software to startups. Its first set of fifteen prompts breaks down as five category discovery prompts ("best project management tools for startups" and variants), four use-case prompts split across founder and agency personas, three comparison prompts including "Acme vs Beta for sprint planning", two brand-defining prompts, and one problem-framed prompt. Reading visibility per category shows Acme is strong on comparison and brand prompts but nearly absent on category discovery, which becomes the focus of the improve Share of Voice work.
What good looks like
A well-chosen prompt set has:
- Prompts that sound like customer questions, not your tagline.
- Coverage across discovery, use-case, comparison, and brand stages.
- Separate, labelled prompts for each persona or region you sell into.
- Ten to twenty prompts, weighted toward category over brand-defining ones.
Common mistakes
Stacking the set with brand prompts. You will win them and learn nothing about whether you show up where buyers actually look.
Averaging personas together. A blended view hides that you win founders and lose agencies. Keep them separate.