This episode turns AI into a research and ideation engine without letting it lie to you. The throughline: AI is fast at synthesizing and structuring, but it has no idea what your real customers said, so output quality is capped by the real artifacts you feed it. Retrieval over recall.
Job 1 - Audience research. ICP vs buyer persona (HubSpot); Jobs-to-be-Done from Clayton Christensen (FullStory); voice-of-customer and the spine technique, review mining (ReadsToLeads, Valchanova). Mine the 3-star reviews. Tools: SparkToro (2025 redesign), AnswerThePublic, Perplexity. Reported conversion lifts from mined language.
Job 2 - Competitor teardowns. Search intent and SERP briefs; content gap analysis (Single Grain, Yotpo 2026); reviews as competitive intelligence. Find what they're NOT saying.
Job 3 - Positioning. April Dunford five components (canvas); Schwartz's 5 stages of awareness (LeadGen); the "so what?" and name-swap tests.
Job 4 - Content briefs. The keystone deliverable. Anatomy and templates: RankUp, AgencyAnalytics, SEOmonitor.
Three pitfalls. AI fabricates stats and citations (rising fake-citation rates, phys.org); deep research (OpenAI, Perplexity, Gemini) makes verification possible, not output true. Generic personas (UX Psychology, Britopian, ACM). And research theater. Ground in real artifacts (Claude Projects, vs Custom GPTs) and route every finding into the brief.