This does not invent prompts. It reads live autocomplete data, the dropdown that appears when someone starts typing, which is a record of what real people actually searched. It pulls those queries for your topic, sorts them into buyer-intent clusters, and assigns each cluster to the engine that owns that moment. What comes back is sometimes uncomfortable: the demand is often not the audience you assumed.
One "AI SEO playbook" is a mistake. Every query below is assigned to the engine that owns that moment.
Rewards citations and third-party corroboration. Open category questions, where your brand gets introduced. Think impressions, not clicks.
Rewards entity authority and surfaces brand mentions. Branded, comparison, and shortlist queries get decided inside the answer box.
Dense citations, sources on display. Mid-funnel research and proof-seeking. Gated expertise does not exist here.
Method, plainly. Two passes. The first is neutral: your topic on its own, plus every a to z continuation, which is an unsteered read of what people search. The second adds question and commercial starters to fill out the clusters. Both go to Google's public autocomplete endpoint, with DuckDuckGo as a cross-check, through a Cloudflare Worker on this domain that exists only to get around the browser's cross-origin rule. Nothing is stored, logged, or kept.
Two honest limits. The intent split is measured on the neutral pass only, because seeding with words like "agency" or "outsource" would manufacture the buying signal it claims to detect. And autocomplete reflects search demand, not AI-prompt volume, no public source exists for the latter, so treat these as the questions behind the prompts. The clustering and engine assignment are judgment, mine, not an API.