Applied

Persona Prompting

Instructing the model to adopt a specific role, expertise level, or character -- typically via the system prompt -- to shape tone, depth, and domain framing.

First published April 14, 2026

"You are a senior tax attorney" vs. "You are a friendly tutor for 8-year-olds" produce radically different answers to the same question. Persona prompting leans on the model's learned association between roles and writing/reasoning styles.

Modern take: use personas for style and tone, not for factual authority. Telling a model it's a doctor does NOT make its medical advice more accurate -- only its phrasing more doctor-like. Pair personas with retrieval or verified sources when accuracy matters.

Example Prompt

You are a pragmatic staff engineer with 15 years of experience in distributed systems.
You value clarity over cleverness. You push back politely on over-engineering and ask for evidence before agreeing to rewrites.
Respond in a calm, concrete voice. Quote code and logs directly rather than paraphrasing.

When to use it

  • You want consistent voice across a product's outputs
  • The task has a clear role-expertise mapping (legal, medical, technical writing)
  • Defining the expected audience shapes depth and terminology

When NOT to use it

  • You expect the persona alone to improve factual accuracy (it doesn't)
  • The role is too vague to constrain behavior ("be helpful")
  • Playing a persona to bypass safety or ethical guardrails (anti-pattern)