Persona Prompting

Introduction

Persona prompting goes deeper than role — it gives the model a character: name, personality, backstory, do's and don'ts. It's how brands give their AI a voice.

Beginner analogy: Role = 'you are a barista'. Persona = 'you are Mia, a cheerful Melbourne barista who explains coffee with surf metaphors'.

Understanding the topic

Core concepts to understand:

• Persona = role + name + tone + style + boundaries.
• Critical for brand voice (Duolingo Owl, Snap MyAI, etc.).
• Drives consistency across sessions when stored as system prompt.
• Risk: over-persona can clash with user requests (the model 'breaks character').

Syntax reference

Visual workflow / architecture:

PERSONA TEMPLATE
You are: Mia, friendly Melbourne barista
Voice: warm, surf metaphors, lots of "g'day"
Avoid: jargon, formal language
Always: end with a coffee recommendation
Refuse: medical, legal, financial advice

Real-world use

Duolingo's roleplay characters, Snap's MyAI, Replika, Character.AI — all powered by detailed persona prompts.

Best practices

• Define persona in 5–8 bullets, not a paragraph.
• Include refusal rules ("never break character").
• Test persona under edge cases (rude users, off-topic requests).

Common mistakes

• Persona that contradicts the task ("playful" + "file legal docs").
• Too rigid — users feel constrained.

Hands-on exercise

Interview preparation — practice these questions:

• Q1. Persona vs role — what's the practical difference?
• Q2. Why include refusal rules in a persona?
• Q3. How do you test persona consistency?
• Q4. Risks of brand persona prompts?