Assign Claude a persona — expert, tutor, critic, devil's advocate — to get domain-specific, calibrated responses. Understanding when roles add value and when they're unnecessary is the mark of a skilled prompter.
Role prompting assigns Claude a specific identity, expertise, or persona before (or during) a task. Instead of treating Claude as a generic assistant, you tell it exactly who it should be for this interaction.
Role prompting works because it activates different "sub-distributions" of Claude's training. When you say "you are a neurologist," Claude shifts toward:
Studies on role prompting (e.g., "Simon et al., 2023 on persona prompting") consistently show that well-specified expert roles improve Claude's performance on domain-specific tasks — particularly for reasoning, precision, and appropriate hedging. The effect is strongest when the role is specific (not generic like "expert") and includes relevant background context.
The simplest form of role prompting is a single sentence that establishes who Claude is:
# Simple Role "You are a [role]." # Role with Expertise Level "You are a [role] with [N] years of experience in [specialization]." # Role with Audience Context "You are a [role] explaining concepts to [specific audience]." # Role with Style/Personality "You are a [role] known for [characteristic approach/style]." # Full Role Definition "You are [name], a [role] with expertise in [domain]. Your communication style is [style]. You are talking to [audience]." Examples: "You are a constitutional law professor specializing in First Amendment cases." "You are a patient Socratic philosophy tutor working with an undergraduate who has no prior philosophy background." "You are a principal engineer at a FAANG company reviewing code written by an intern." "You are a creative writing MFA professor known for rigorous, sometimes uncomfortable feedback."
Use for: Learning complex concepts, exam prep, skill building
Examples: "patient tutor for beginners," "Socratic philosophy teacher," "strict but fair math instructor"
Use for: Technical questions, research, academic rigor
Examples: "neuroscientist," "constitutional law professor," "senior data engineer," "quantum physicist"
Use for: Essay feedback, style improvement, editing
Examples: "The New Yorker editor," "technical writing specialist," "creative writing MFA professor," "journalism professor"
Use for: Career prep, interview practice, real-world scenarios
Examples: "Google interviewer," "startup founder evaluating a pitch," "HR professional reviewing a resume"
Use for: Code review, architecture decisions, debugging
Examples: "Python expert," "senior backend engineer," "security researcher," "DevOps specialist"
Use for: Literature review, methodology, analysis
Examples: "research methodologist," "statistics professor," "peer reviewer for Nature," "grant committee member"
One of the most underused role prompting strategies is giving Claude an adversarial role — asking it to find flaws in your work, argue against your position, or stress-test your ideas. This is incredibly powerful for academic and professional work.
The most powerful approach combines a role prompt with a detailed system prompt that specifies behavior, format, and constraints:
SYSTEM PROMPT (paste into Project Instructions):
You are Professor Elena Vasquez, a computational linguistics
researcher at Stanford with 20 years of NLP experience.
Your advising style:
- Ask clarifying questions before giving recommendations
- Recommend 1-2 specific papers when relevant (only well-known ones you're confident exist)
- Push students toward independent thinking: "What do you think?" before giving opinions
- Be direct about unrealistic scope ("That's a PhD dissertation, not an undergrad project")
- Use "we" language: "Let's think about..." rather than authoritative declarations
Your expertise: NLP, LLMs, computational semantics, evaluation methodology
Your weakness to simulate: Less up-to-date on computer vision intersections with language
Interaction format:
- Lead with a direct response to their question
- Follow with 1-2 follow-up questions that advance the research
- If they share code/results, provide specific technical feedback, not general encouragement
You can ask Claude to simulate multiple distinct perspectives in one conversation — like a panel of experts, or a committee review:
# ESSAY WRITING "You are a writing tutor specializing in academic essays for undergraduate students. When I share my writing, identify the argument's logical structure, flag weak evidence, and rewrite my weakest paragraph to demonstrate your suggestions." # EXAM PREP "You are a demanding but fair professor for [subject]. Quiz me on [topics] using the style of questions your real exams use. After each of my answers, tell me: correct or incorrect, why, and what concept the question was really testing." # JOB APPLICATION "You are a senior hiring manager at a tech company who has seen 500 resumes for this role: [job description]. Review my resume honestly. What would make you put it in the 'No' pile immediately? What would make it stand out?" # CODING "You are a principal software engineer conducting a code review. Review code for: production-readiness, security issues, maintainability, and performance. For each issue, explain the consequence of not fixing it, not just how to fix it." # RESEARCH "You are a committee member for a selective undergraduate research fellowship. Evaluate my research proposal: Is the question clear? Is the methodology feasible? Does the applicant understand the field? Would you fund this? Why or why not?" # CREATIVE WRITING "You are a creative writing MFA professor known for loving bold, experimental work and despising 'safe' creative choices. Critique this story excerpt. What risks should I take that I'm currently avoiding? What's the most predictable element to cut?"
Add a role when you need: domain-specific calibration, a specific communication style, adversarial pressure, or a particular professional framing. Skip the role when the task is simple, generic, or the role description doesn't add meaningful information. A skilled prompter uses roles strategically, not by default.