Start Lesson
Here is a prompt that gets a generic answer:
Review my business plan for a new SaaS product.
You get a surface-level summary that reads like a Wikipedia article about business plans. It covers "market opportunity" and "revenue model" in vague terms. Nothing you could not have written yourself.
Now add five words to the front:
Act as a skeptical CTO with 15 years of experience scaling
B2B platforms. Review my business plan for a new SaaS product.
Focus on technical feasibility, architecture risks, and
build-vs-buy decisions.
You get pointed questions about your database architecture, a warning about premature microservices, a cost comparison of building auth vs. using a managed service, and a flag that your timeline assumes zero onboarding time for new engineers.
Same task. The role changed everything. In Lesson 1, you learned the CGC framework -- Context, Goal, Constraints. The role goes into the Context section, and it is often the single highest-impact change you can make.
After this lesson, you will be able to: use the "Act as..." pattern and the persona stack to get domain-expert-level output from any AI model.
When you tell the AI to act as a copywriter, you are narrowing the probability space of its response. Instead of drawing from everything it knows -- writing, marketing, product design, engineering, law, all at once -- it focuses on how a copywriter would approach this specific task. The output becomes sharper, more opinionated, and more useful.
Without a role, the AI defaults to being a generalist. A generalist gives you the average of all possible responses. That is rarely what you want.
A role alone is good. A fully stacked persona is better. There are four layers:
Watch the difference:
| Layer | Basic | Stacked | |---|---|---| | Role | "Act as a financial analyst" | "Act as a financial analyst" | | Expertise | (none) | "with 15 years at a Fortune 500 company" | | Style | (none) | "Be direct, use no jargon" | | Audience | (none) | "Explain to a non-technical board member" |
The stacked version produces output that is specific in expertise, tailored in communication style, and calibrated for the right reader.
Act as a [role] with [years] years of experience in [domain].
Review the following [document/plan/draft]:
---
[Paste your content here]
---
Focus on: [2-3 specific areas to evaluate].
For each issue, explain what is wrong and suggest a fix.
Tone: [direct / diplomatic / technical]. Keep it under [number] words.
Expected output: A structured review with specific issues called out, each paired with a concrete suggestion. Not "this could be improved" but "the pricing section assumes 80% retention, which is above the SaaS benchmark of 65% -- model a conservative scenario at 55%."
Works with Claude, GPT-4, and Gemini. The more specific the role and focus areas, the sharper the review.
I need three different expert perspectives on this decision:
[Describe your decision in 2-3 sentences]
Perspective 1 -- Act as a [role, e.g., CFO]: Focus on
financial risk and ROI.
Perspective 2 -- Act as a [role, e.g., customer]: Focus on
whether this solves a real pain point.
Perspective 3 -- Act as a [role, e.g., competitor]: Focus on
how you would counter this move.
For each perspective, give me: the top concern, the biggest
opportunity, and one question I should answer before deciding.
Format as three separate sections.
Expected output: Three clearly separated sections, each with a distinct angle. The CFO flags cash flow timing. The customer asks why this is better than the free alternative. The competitor identifies the feature gap they would exploit. You get a 360-degree view in one prompt.
Act as a [role] who communicates with [audience type] daily.
Take the following technical content:
---
[Paste technical content here]
---
Rewrite it for [target audience]. Maintain accuracy but
adjust vocabulary, examples, and detail level for someone
who [description of their knowledge level].
Length: under [number] words. Do not oversimplify -- keep
the key nuances but explain them in accessible terms.
Expected output: The same information, reframed for the target reader. A machine learning explanation for engineers becomes a business-impact summary for executives. A legal clause becomes a plain-language FAQ for customers.
Use roles when:
Skip roles when:
Pick a real decision you are working on -- a project direction, a hiring choice, a product feature. Use Template 2 (Multi-Perspective Scan) with these three roles:
I need three different expert perspectives on this decision:
[Your decision here]
Perspective 1 -- Act as a skeptical investor: Focus on
what could go wrong and what proof is missing.
Perspective 2 -- Act as the ideal customer: Focus on
whether this solves a real problem worth paying for.
Perspective 3 -- Act as a journalist covering your industry:
Focus on whether this is newsworthy or derivative.
For each perspective, give me: the top concern, the biggest
opportunity, and one question I should answer before deciding.
Check your output: Each perspective should give you a genuinely different angle. If they all sound the same, your decision description was too vague -- add more specifics about the stakes, the alternatives, and the constraints.
You now have CGC (Lesson 1) and role prompting (this lesson) -- structure plus expertise. In the next lesson, you will add chain-of-thought prompting: forcing the AI to reason through problems step by step instead of jumping to conclusions. This is the technique that turns confident-but-wrong answers into reliable analysis.