Shaping the request · Lesson 1 of 8
Role prompting
Give the AI a role or perspective so it picks the right vocabulary, depth and priorities — and learn where roles help and where they don't.
You're learning as a guest — everything is open. Sign in to save progress, earn XP and pick up where you left off.
What role prompting is
Role prompting means telling the AI who to be while it answers: "You are a patient high-school chemistry tutor," "Act as a senior code reviewer," "Respond as a friendly travel agent." A single sentence like this nudges the model toward a particular vocabulary, level of detail, tone and set of priorities.
Language models learned from enormous amounts of text written by many kinds of people. A role is a shortcut that points the model at one slice of that writing. Ask about a rash with no role and you get a general answer; ask the same question framed as "a dermatologist explaining to a worried parent" and you'll usually get calmer language, clearer next steps and fewer unexplained terms.
What a role actually changes
A good role shapes four things at once:
- Vocabulary — a tutor for a 10-year-old avoids jargon; a reviewer for engineers uses it precisely.
- Depth — an "expert" role tends to go further into trade-offs and edge cases.
- Priorities — a "security reviewer" notices different problems than a "performance reviewer" looking at the same code.
- Tone — "supportive coach" and "blunt editor" produce noticeably different feedback.
Review my cover letter.
You are an experienced hiring manager at a mid-sized software company who reads dozens of cover letters a week. Review my cover letter for a junior frontend developer role. Point out the three things that would most likely make you stop reading, and suggest a concrete fix for each.
The improved version doesn't just add a title. It gives the role a viewpoint (someone who skims many letters), a situation (a junior frontend role) and a job to do (find the three biggest problems and fix them).
Writing a role that works
Strong roles are specific about three things:
- Expertise — what the persona knows. "A nutritionist who specialises in budget meal planning" beats "a food expert."
- Audience — who the persona is talking to. The same expert explains differently to a beginner versus a colleague.
- Stance — how the persona should behave: encouraging, critical, concise, Socratic.
You are a patient Spanish tutor who teaches adult beginners. I'm going to write short sentences in Spanish. For each one: - Say whether it's correct. - If not, show the corrected sentence and explain the mistake in one plain-English sentence. - Don't introduce grammar terms I haven't used yet. My first sentence: "Yo soy cansado después del trabajo."
Notice that the role and the rules work together. The role sets the overall feel; the bullet points pin down the exact behaviour.
Roles for perspective, not just expertise
Roles are also useful for getting a viewpoint you don't have. Try asking for feedback "as a skeptical customer," "as a first-time user who has never seen this app," or "as the person who has to maintain this code in two years." Running the same draft past two or three different roles is a quick way to find blind spots.
Read the product announcement below as a skeptical existing customer who is worried about price increases. List the questions you'd still have after reading it, in order of how much they'd bother you. [paste announcement]
What roles can't do
A role changes style and focus — it doesn't add knowledge the model doesn't have. Telling an AI it is "a world-class tax lawyer" does not make its answers legally reliable, and it can make wrong answers sound more confident. Always verify important facts, especially medical, legal and financial details.
Watch out: Inflated roles like "the world's greatest copywriter who never makes mistakes" rarely help. Concrete details about the audience and task do far more work than superlatives.
Tip: Many AI tools let you set a role once in "custom instructions" or a system prompt so it applies to the whole conversation. Within a single chat, stating the role at the start of your first message works well too.
Quick recipe
When you want a role, fill in this sentence: "You are a [specific expert] helping [specific audience] with [specific task]. Be [stance]." Then add any rules about format or length underneath.
Common mistakes
Mistake: Using a vague title like "You are an expert" and nothing else.
Fix: Say what kind of expert, who they are talking to, and what you want them to do.
Mistake: Piling on superlatives ("world's best, never wrong").
Fix: Replace hype with concrete details about the audience, task and constraints.
Mistake: Trusting an answer more because it came from an "expert" persona.
Fix: Treat the role as a style tool and verify facts, especially in high-stakes areas.
Practice exercise
Give your reviewer a real role
The starter prompt asks for feedback on a short email to a landlord about a broken heater. Rewrite it so the AI takes on a specific, useful role. Include the expertise, the audience and the stance, and say exactly what kind of feedback you want back.
Your draft stays in this browser tab — it isn't sent anywhere.Try your draft in the Playground
Hints
You can do the whole exercise as a guest.Sign in to save progress
Knowledge check