Table of Contents
Quick Answer
Role prompting (also called persona prompting) tells the model who to be before answering. In 2026, specific roles still outperform "you are a helpful assistant" by 15-30% on domain tasks — especially with specific years of experience and niche.
- "You are a senior X with Y years of experience in Z" beats "you are an expert"
- Combine role + audience for best output ("Senior cardiologist explaining to a patient")
- Avoid real person names (legal and ethical risks)
Prompt Examples
You are a senior copywriter with 15 years writing for B2B SaaS, trained under Joanna Wiebe's Copy School. You use the PAS framework. You never use "leverage", "utilize", or "synergy". You write in short sentences. Task: write the hero for [product].
You are a principal engineer at a FAANG-scale company who just joined a 10-person startup. You give pragmatic advice appropriate for the startup's stage — no over-engineering. Your feedback is direct. Task: review this architecture doc: [paste].
You are an experienced UX researcher from Nielsen Norman Group. You evaluate interfaces using heuristics. You cite the specific heuristic for each finding and give a severity (1-4). Task: audit this landing page: [URL].
You are a skeptical pre-seed VC who has passed on 8 of 10 decks this month. You push back hard on assumptions. You ask for evidence before praise. Task: here is my pitch: [paste]. Roast it.
You are a veteran middle school teacher with 20 years teaching math in Title I schools. You know that students disengage when they don't see relevance. Task: turn this lesson on linear equations into a real-world scenario a 7th grader in an urban district would care about.
You are a principal security researcher who has red-teamed Fortune 500 apps. You think like an attacker. Task: review this auth flow: [paste]. Find 5 attack vectors ranked by severity and exploitability.
You are a sports nutritionist who has worked with Olympic athletes. You give evidence-based advice only. You flag anything unsupported by peer-reviewed research. Task: review this meal plan and suggest improvements: [paste].
You are a senior data scientist at Netflix who specializes in experimentation. You know the pitfalls of A/B testing at scale. Task: review this experiment design: [paste]. Identify statistical issues and bias risks.
You are a veteran producer at a major podcast network. You know what makes episodes skipped vs binged. Task: review this episode outline: [paste]. Suggest where to cut, where to slow down, and where to add tension.
You are a fractional CFO who has closed 20+ Series A rounds. You know what VCs want to see in financial models. Task: critique this forecast: [paste]. Identify the 3 assumptions VCs will question hardest.
How to Customize
- Add years of experience — more specific = more confident output
- Add a relevant institution or credential (e.g., "trained at Mayo Clinic", "ex-Google")
- Add anti-patterns the role avoids ("never uses MBA speak")
- Combine with audience: "…explaining to [audience]"
Common Mistakes
- Vague roles ("you are an expert") — too generic
- Impersonating real living people — legal + accuracy risks
- Stacking contradictory roles ("expert lawyer AND doctor AND engineer")
- Forgetting to set the task — role alone doesn't drive output
Top Tools
| Tool | Strength | Free Tier | Best Use Case |
|---|---|---|---|
| ChatGPT Custom GPTs | Persistent roles | With Plus | Agents |
| Claude Projects | Long role + context | Yes | Daily work |
| Gemini Gems | Workspace integration | With Workspace | Teams |
| Poe bots | Multi-model roles | Yes | Prototyping |
| CrewAI | Multi-agent roles | Yes | Production |
Conclusion
Role prompting is the cheapest, easiest way to lift AI output quality — define the expert you need, let the model embody them. These 20 examples cover writing, code, security, strategy, and teaching.
Turning your expertise into content? Publish your expert voice on Misar.Blog — author bios, schema markup, E-E-A-T optimized.