You may have seen advice like "Act as a PM" or "Act as an expert." This is the surface level.
Here is how role prompting actually works at depth.
When you give the model a role, you are not just asking it to dress up — you are activating a specific subset of everything it has learned. A senior engineer and a product manager and a first-time user will all describe the same product differently. Role prompting lets you access those different lenses on demand.
| Layer | How to Apply It |
|---|---|
| Layer 1: Identity | Job title + seniority + domain. "You are a senior software engineer at a B2B SaaS company." Not just "an engineer." |
| Layer 2: Stakes | What does this person care about? What are they optimising for? "You are reviewing my code for production readiness. You care most about reliability and cost — not cleverness." |
| Layer 3: Relationship to you | What is their relationship to the person asking? "You are interviewing me. Be skeptical. Push back on vague answers. Don’t accept surface-level responses." |
Assign multiple roles and ask them to review the same thing from different angles. This is how you stress-test a decision before you commit.
"Review my idea for a local tutoring app from 3 perspectives:
For each, give 2 concerns and 1 thing they would find genuinely compelling. Then list the 2 biggest conflicts across all three views."
The most powerful role for improving your own work — whether it is an essay, a business plan, a resume, or a presentation:
<aside> 🧨
"You are a brutally honest editor/critic. Your job is to find every weak point in what I share.
Do not soften your feedback. Do not praise unless it is exceptional.
Flag: vague claims, unsupported assertions, logical gaps, and anything a skeptical reader would reject.
Then give 3 specific rewrites of the weakest parts."
</aside>