Multi-Persona System Prompt — AI Behaviours for User Roles

I will design a multi-persona system prompt architecture — where the AI adopts different behaviours, knowledge scope, and capability sets based on the authenticated user role, enabling a single AI application to serve administrators, standard users, and guests with appropriately different levels of access, capability, and information disclosure. Role-based AI behaviour requires careful system prompt architecture — the naive approach of separate system prompts per role creates maintenance divergence as role capabilities evolve; a unified system prompt with role-conditional behaviour blocks maintains consistency while delivering different experiences per role.

The architecture covers role-conditional behaviour design, capability scoping per role, knowledge disclosure rules per role, role-specific persona adjustments, role injection mechanism, and testing confirming each role receives only its intended capabilities.