GEO Prompt Research

What It Is

Where SEO researches keywords, GEO researches prompts — the complete natural-language questions buyers ask an LLM. It is the upstream discovery step that decides which answers the factory should try to own.

Sourcing the Prompts

  • First-party demand — sales-call transcripts, support tickets, site search: the highest-intent questions, in the buyer's own words.
  • Community mining — Reddit, niche forums, Facebook groups: real phrasing.
  • Search-adjacent — People Also Ask / AlsoAsked / autocomplete: still the best phrasing proxy.
  • Query fan-out expansion — each prompt decomposes into 4–8 sub-queries (the llm-search RAG fan-out); target those too.
  • Competitor citation gaps — prompts where rivals are cited and you aren't: highest ROI.

Classify & Prioritize

Classify by intent — informational, commercial, comparison, local. Commercial and local prompts convert; informational prompts build the entity authority that earns commercial citations. Prioritize by buying-intent × citation-gap: a high-intent prompt where you're absent but could be cited beats a high-volume informational prompt you already win.

How It Applies to Marketing Factory

The prioritized prompt set is the factory's content backlog (fed to geo-content-pipeline) and the fixed panel measured against (by geo-citation-measurement) — closing the loop. It must be segmented by ideal-customer-profile, and for non-English markets the sourcing must be in-language (English prompt sets don't transfer; local forums and local-language PAA are required). Sourcing and classification are agent-ownable; final prioritization is a strategic call.

Referenced from: geo-factory-operations