A Buyer-Intent Prompt Set for Measuring AI Search Visibility

Every AI-visibility number you'll ever report is downstream of one artifact: the prompt set. If the prompts don't resemble what buyers actually ask, you're measuring visibility in conversations that never happen. If they quietly change, your trend line is fiction. The set deserves the same discipline as a database schema.

The intent mix

• Problem prompts — the pain, no category vocabulary: 'my team double-books client calls constantly'
• Category prompts — 'best scheduling tools for B2B sales teams'
• Comparison prompts — 'X vs Y for round-robin scheduling' using the competitors buyers actually shortlist
• Pricing prompts — 'cheapest way to get team scheduling with CRM sync'
• Objection prompts — 'is it safe to give a scheduling tool calendar access?'

Five intents, five prompts each is a defensible 25. Weight toward the intents your funnel actually converts from, and write prompts in buyer language — messy, specific, occasionally misspelled — not in your category's marketing vocabulary.

Inclusion rules that keep the set honest

1. No prompt names your brand — brand-inclusive prompts measure recall, not visibility, and belong in a separate set
2. Every prompt must be answerable by naming vendors — 'how does scheduling software work' can't produce a mention and wastes a slot
3. Each prompt gets an id and an intent tag that never change
4. The set has a version; any addition, removal or rewording bumps it, and cross-version comparisons are flagged in reporting

A filled example — fictional, for shape only

For Lumina Desk (fictional), prompt LD-C3, intent 'comparison', reads: 'calendly alternatives that do round-robin for a 12-person sales team'. It earns its slot because it's specific enough to force vendor names, generic enough that a real buyer would type it, and it doesn't contain the brand. Twenty-four siblings like it make a set worth trusting.

A failure worth checking

Brand leakage is the quiet invalidator: a well-meaning teammate adds 'is Lumina Desk good for agencies?' to the set, mentions jump, and the chart celebrates. The set now partially measures whether engines echo your own name back at you. Audit for it explicitly — the rule is mechanical: if the prompt contains the brand, it's in the wrong file.