Prompt · Role Prompting · Advanced
Role: Get a Stress-Tested Answer from a Specific Expert
Give the expert a situation. Set the stakes. Get pushback.
An advanced role prompt adds context, ranked priorities and a built-in counter-argument. Use it for decisions with real trade-offs.
About This Prompt
- Author
- Adam Bloom
- AI Technique
- Role Prompting
- Level
- Advanced
- Best For
- Business owners · High-stakes decisions
- Last Updated
- 2026-11-10
- Last Reviewed
- 2026-11-10
Role Prompt Template
Free to use. Copy it, change it, share it. No credit needed.
What to Fill In
- [FIELD]: the area of expertise, such as pricing for a service business.
- [ROLE]: the job the AI plays, such as a finance lead.
- [YEARS] and [TYPE OF BUSINESS]: how seasoned the expert is, and who they have advised.
- [SITUATION]: the pattern the expert has seen before, in one sentence.
- [PRIORITY 1 to 3]: what to protect, ranked. The first one wins a tie.
- Context lines: your size, numbers, limits and past attempts. Use real figures, or leave a line out.
- [MY QUESTION]: the decision you want a view on.
Example: An Advanced Role Prompt on Pricing
Illustrative example, not a client result
Input: Field: pricing for a service business. Role: finance lead. Years: 20. Business type: owner-run service firms. Situation: costs rose and prices stayed flat for two years. Priorities: margin, best clients, no surprises. Business: 12-person consulting firm, about $2M revenue, 28% margin. Constraints: change starts next quarter, no new hires. Tried: small discounts for annual contracts. Question: Should I raise prices 15% next quarter?
Prompt as sent:
Output (shortened):
- Decision: Not “15% or not.” The real decision is how much, for which clients, and when.
- View: Raise prices, but not a flat 15%. Tie the number to your cost growth. Test it on new work first.
- Check first: Cost growth since your last raise. Churn you can absorb before it’s a net loss. Clients below your average rate. What similar firms charge. How much margin your top five clients carry.
- Other side: Holding prices may keep clients through a soft market. If your clients are cutting budgets, a raise could cost more than it earns.
- Risks: A surprise bill (high). Losing a top client (medium). Staff asked to defend a number they didn’t set (low).
- Next: This week, pull costs and rates by client. This month, draft the new rate card and the notice. This quarter, apply it to new clients and give existing clients 60 days’ notice (assumption).
- Would change my mind: If a few top clients carry most of your revenue, I’d phase the raise.
Flag: The 60 days’ notice is my assumption, not a figure from you. I can’t see your churn data, so check it before you decide.
Why the Role Prompt Works
The situation and ranked priorities give the AI something to weigh. The “other side” step forces an objection before the answer. The ask-first rule stops it from guessing at missing facts.
When the Role Prompt Doesn’t Work
- A long prompt can bury the question. Cut any line the answer doesn’t need.
- Arguing the other side surfaces objections. It is not a fact check. Verify any number before you act on it.
- For a simple question, this is overkill. Use the simple version.
Research Behind This Prompt
- Wang et al. (2023), Unleashing the Emergent Cognitive Synergy in Large Language Models: A Task-Solving Agent through Multi-Persona Self-Collaboration. Supports. Solo Performance Prompting had one model play several personas on the same task. It reduced factual errors and kept reasoning strong, but only in GPT-4, not in the weaker models tested.
- Huang et al. (2023), Large Language Models Cannot Self-Correct Reasoning Yet. Challenges. Without outside feedback, models struggled to fix their own reasoning, and performance sometimes got worse. That bears on the “argue the other side” step.
- Ferreira et al. (2026), Beyond Symmetric Agents: Cognitive Diversity and Multi-Agent Debate in Small Language Models. Mixed. In small models, debate beat a single answer by 3 to 7 points where there was room to improve, but at equal compute it tied or lost to simple voting. Persona prompting reduced accuracy.
What the Evidence Says
Mixed. Several personas helped in one study, but only on the strongest model. Models struggled to catch their own reasoning errors. In small models, debate beat a single answer but not simple voting, and personas cost accuracy. Use the steps for structure and the other-side step for objections, then check the facts yourself.
Sources last checked 2026-11-10.
Role Prompt FAQ
Why does the prompt state a situation and priorities?
They give the expert something to weigh. Without ranked priorities, the AI picks its own, and they may not be yours.
Why ask the AI to argue the other side?
It surfaces objections you would otherwise miss. It is not a fact check, so verify any claim before you act.
Is a longer prompt always better?
No. Length helps when the question has real trade-offs. For a simple question, use the simple version.
What does the ask-first rule do?
It lets the AI ask for missing facts instead of guessing. The cap of three questions keeps it from turning into an interview.
Why ask what would change the expert’s mind?
It shows what the advice depends on. If those facts turn out wrong, you know to rethink the plan.
Want this applied to your business?
No pitch. Just a conversation.