
Tell the AI who it is before asking what you want. Assigning a role shapes the tone, vocabulary, depth, and perspective of every response that follows.
TLDR
Start your prompt with "You are a [specific role with relevant context]." A well-defined role primes the model to respond with the right expertise, tone, and assumptions for your task.
Choose a specific role, not a vague one
"You are an expert" gives the model nothing to work with. "You are a senior UX designer who specializes in mobile onboarding" tells it exactly what expertise to draw on. The more specific the role, the more targeted the response.
Add context to the role
Include the role's perspective, audience, or constraints: "You are a B2B sales consultant who has closed deals in SaaS and financial services." This context shapes what the model emphasizes and assumes.
Set the role before anything else
Place the role at the very beginning of your prompt. Everything that follows will be interpreted through that lens. Setting it after your question reduces its effect.
Match the role to the task
Use a role that genuinely relates to what you are asking. A copywriter role for writing tasks, a data scientist role for analysis tasks. Mismatched roles produce generic output because there is no relevant expertise to activate.
Reinforce the role in long conversations
In multi-turn conversations, roles can drift. If the model starts responding out of character, bring it back: "Staying in your role as [X], now help me with..."
Example prompt
E-commerce optimization: getting expert-level CRO advice rather than generic suggestions
You are a conversion rate optimization specialist who has run A/B tests for e-commerce brands with over $1M in annual revenue. I am launching a checkout page redesign and want to reduce cart abandonment. Review the following checkout flow and identify the three highest-impact changes I should test first. For each, explain the psychological principle behind the recommendation. [Paste your checkout flow description here]
Getting expert-level depth
When you need more than surface-level answers, a specific expert role activates relevant knowledge, terminology, and frameworks that a generic prompt would not.
Controlling communication style
Different roles communicate differently. A "senior engineer" role produces technical precision. A "product marketer" role produces clearer, benefit-focused language. Use the role to control how, not just what, the AI communicates.
Maintaining consistency across a session
Setting a role at the start of a conversation keeps responses consistent in tone and perspective, which is useful when writing a document, conducting a simulated interview, or reviewing something from one angle.
Using generic roles
"You are a helpful assistant" gives the model nothing to work with. Specificity is everything: what kind of expert, in what domain, with what constraints or audience.
Setting the role after the task
If you describe the task first and add the role at the end, the model may not apply it fully. Always set the role first.
Mismatching role and task
Asking a "legal expert" to write casual social copy creates tension that usually produces mediocre results. Match the role's natural output style to what you actually need.
Yes, consistently. Assigning a specific, relevant role improves response quality across domains, especially for tasks requiring specialized knowledge, tone consistency, or a particular perspective. The key word is specific: vague roles produce vague improvements.
Yes, and it is more durable there. Setting a role in the system prompt means it persists across the entire conversation. Setting it in a user message works but may need reinforcement in longer sessions.
Ask the model to respond as multiple roles sequentially: "First respond as a skeptical investor, then as a supportive product manager." This is a useful technique for getting balanced feedback on a decision.
Bottom line
A specific, well-matched role shapes the expertise, tone, and depth of everything that follows. The more specific the role, the better the result. Set it first, before anything else.
Prompt packages that apply this technique directly.
Content Marketing Strategy
Most content marketing fails because it produces content without a strategy.
See promptsCopywriting
Most copy fails because it talks about the product instead of the reader's problem, buries the benefit in generic claims, and never gives the reader a clear reason to act now.
See promptsTechnical Documentation
Creating comprehensive technical documentation can overwhelm teams, leaving critical gaps or unclear instructions.
See prompts