building-evidence-based-personaslisted

# Building Evidence-Based Personas ## Overview A persona is a **behavioral model** of a user type, built from research — not a demographic profile and not a fictional character. Its value is capturing **goals** (stable end-states and motivations) distinct from **tasks** (the changeable steps to reach them). Design for goals. ## AI guardrail **Do not generate personas from an AI's general knowledge.** AI personas are "a generic stereotype based on averaged internet content" — blind to real context of use and B2B complexity, and they describe idealized rather than actual behavior. AI may help *structure* a persona *from your own research data*; it must not be the source of the user. **If you have neither research nor a named human team's stated assumptions, STOP and ask for them — never generate personas (not even labeled "proto") from your own knowledge.** ## Choose the right rigor | Type | Built from | Use when | |------|-----------|----------| | **Proto-persona** | A **named human team's** stated assumptions (never the model's), no new research | Aligning beliefs fast / low maturity — **label it a hypothesis and validate later**; if no human assumptions exist, ask rather than generate | | **Qualitative persona** (sweet spot) | ~5–30 interviews, coded into patterns | Most teams, most decisions | | **Statistical persona** | Qual + large survey (100–500+), clustered | Large orgs with statistical expertise and need for population proportions | Proto-personas *start a conv