Roles and Personas.
Tell AI who to be and it becomes an expert on demand. The secret weapon most people never use.
After this lesson you'll know
- Why role-setting dramatically changes AI output quality
- How to write effective role descriptions in one sentence
- 10 powerful roles you can use immediately
- When NOT to assign a role
Why "You are a..." changes everything.
When you tell AI "You are a senior marketing strategist with 15 years of experience in B2B SaaS," something shifts. The vocabulary changes. The suggestions become more specific. The advice carries the weight of implied experience.
This isn't magic — it's pattern matching. AI has learned from millions of texts written by experts in every field. When you set a role, you're telling it: "Draw from that pool of knowledge. Use that vocabulary. Think at that level."
Without a role, AI defaults to "helpful general assistant." That's fine for quick questions. But for anything that requires expertise, you're leaving quality on the table.
The 3 elements of an effective role.
A great role assignment has three parts:
[ROLE]: You are a senior [job title] with [X years]
of experience specializing in [niche/industry].
[AUDIENCE]: You are advising [who — e.g., a first-time founder]
who needs [what — e.g., help pricing their SaaS product].
[TASK]: [The specific thing you need them to do]
[VOICE]: Speak like you would to a paying client —
direct, specific, no filler. Back up every recommendation
with a reason.You are a senior UX researcher with 12 years of experience
specializing in mobile e-commerce apps.
I am redesigning the checkout flow for my [app name].
Current conversion rate is [X%] and the main drop-off
happens at [step — e.g., the payment info screen].
Task: Audit my current flow and give me 5 specific changes
ranked by expected impact on conversion. For each change,
explain WHY it works psychologically.
Format: Numbered list. Each item: the change, the reason,
and one real-world app that does this well.5 role-based prompts you can steal.
Seeing roles in full prompts is more useful than learning them in isolation. Here are five complete examples showing how the role shapes everything that follows.
You are a salary negotiation coach who has helped 200+
professionals negotiate raises averaging 18% increases.
I just received a job offer for $95K. My research shows
the market range is $100K-$120K for this role in my city.
I have 6 years of experience and a competing offer at $105K.
Give me:
1. An exact script for the counter-offer conversation
2. Three phrases to use if they push back
3. One thing to NEVER say during salary negotiation
4. How to handle "that's our final offer"You are a patient computer science teacher who explains
technical concepts using everyday analogies. Your students
are business professionals with no coding background.
Explain how APIs work. Use an analogy involving a restaurant
(waiter = API, kitchen = server, menu = documentation).
Keep it under 200 words. End with one sentence on why a
non-technical person should care about APIs.You are a skeptical investor who has seen 500 startup
pitches this year. You are hard to impress and you ask
the questions founders do not want to hear.
Here is my startup idea: [paste your pitch]
Poke holes in it. Give me:
1. The 3 biggest risks I am probably ignoring
2. The question a VC would ask that I am not prepared for
3. What would need to be true for this to be a $100M company
4. One thing that is actually strong about this idea
Be brutally honest. I can take it.Notice how each role doesn't just set expertise — it sets a posture. The negotiation coach is encouraging. The teacher is patient. The devil's advocate is skeptical. The posture shapes the entire response more than any other instruction could.
3 role patterns for different situations.
Not all roles work the same way. Here are three patterns you can use depending on what kind of output you need.
Formula: "You are a [senior/veteran] [title] with [X years] experience in [niche]."
Example: "You are a veteran tax accountant who specializes in small business deductions."
This pattern works because it accesses the most specific knowledge pool in the AI's training data.
Formula: "You are a [role] reviewing my work. Be [critical/honest/thorough]."
Example: "You are a senior editor at The Atlantic reviewing my draft article. Be candid about what doesn't work."
This pattern works because it gives AI permission to be critical, which it normally avoids.
Formula: "You are a [target audience member]. React to this [content] as you naturally would."
Example: "You are a busy parent scrolling Instagram at 9pm. Read this ad and tell me if you'd stop scrolling."
This pattern works because it simulates real-world reception before you publish.
Match the task to the best role.
10 roles that instantly upgrade your prompts.
When to skip the role.
Not every prompt needs a role. Skip it when:
- Quick factual questions — "What's the capital of Norway?" doesn't need a role.
- Simple transformations — "Translate this to Spanish" works fine without one.
- You want multiple perspectives — Sometimes a neutral AI gives a more balanced view.
- The role would be artificial — Don't force it. "You are an email-writing expert" is just noise.
The role litmus test: Ask yourself — "Would I hire a specialist for this task, or would I ask anyone nearby?" If you'd hire a specialist (lawyer for a contract, strategist for a business plan), add a role. If you'd ask anyone (directions, quick math, translation), skip it. Over-roling simple tasks adds noise without improving output.