Role Prompting

Introduction

Role prompting assigns the model a job title or expertise level. "You are a senior pen-tester reviewing AWS IAM policies" produces very different output than "explain IAM". The role primes vocabulary, depth and tone.

Beginner analogy: Asking your dentist about teeth vs asking a random friend. Same question, very different answer because the role sets expertise.

Understanding the topic

Core concepts to understand:

• Role lives in the system message ideally.
• Concrete roles ("senior security auditor") beat vague ones ("expert").
• Role + audience ("explain to a CFO") shapes both depth and language.
• Roles can be invented ("You are PromptGPT, a specialist in...") but realism helps.

Syntax reference

Visual workflow / architecture:

Same input, different roles:

Role: junior dev    → simplified, learning-focused
Role: senior arch.  → tradeoffs, scale concerns
Role: CISO         → risk, compliance, audit framing

Real-world use

Customer-support bots assign 'polite, empathetic support agent'; code reviewers assign 'principal engineer reviewing for production'. Roles are everywhere in deployed prompts.

Best practices

• State role at the very start of the system message.
• Anchor role with 1 line of expertise + 1 line of constraints.
• Pair role with audience for double the effect.

Hands-on exercise

Interview preparation — practice these questions:

• Q1. Why does role prompting work?
• Q2. Where in the prompt does a role belong?
• Q3. Difference between role and persona prompting?
• Q4. Can role prompting cause hallucinations?