Role prompting assigns a persona to the AI: "You are a senior Python developer." But does it improve output?
Testing in found that generic personas like "Act as an expert" rarely help. The model already tries to be helpful. What works is specific, detailed personas with explicit constraints. "Be an expert" is vague. "Review this code for SQL injection vulnerabilities" is actionable.