Beyond Role Prompting: An Acting Theory-Based Framework for Authentic and Consistent AI Persona Design

Key Visuals

Background / Problem

Many AI character personas in content and entertainment still rely on shallow role prompting (tone and adjective lists), which often fails to sustain coherence and believable behavior. This project investigates acting theory as a rigorous design resource for persona construction, while addressing the structural limitation that AI lacks lived emotional memory.

My Role

I designed the core research framing and conceptual model, including the Stanislavski-Adler reinterpretation strategy, concept selection criteria for AI applicability, and prompt-level operationalization for persona design.

Process / Method

The framework filtered acting concepts using three criteria: operability in prompt text, structural fit with AI constraints, and contribution to authenticity (consistency, contextual responsiveness, human-likeness). Key constructs were translated into prompt modules: Given Circumstances, Super-objective, Magic If, and Units & Objectives. A comparative pilot test (Claude Sonnet 4.6) contrasted baseline role prompting vs framework prompting using identical Hamlet utterance probes.

Result / Outcome

Compared to baseline role prompting, the framework condition showed stronger cross-turn consistency, better context-sensitive goal shifts, and more authentic-seeming internal tension in responses. The work establishes a transferable prompt framework for AI character design and is positioned for conference submission and follow-up validation studies.