Constraint Saturation in Persona-Based AI: A Pilot Study of Behavioral Constraint Density and Perceived Human-Likeness

Abstract

This pilot study investigates whether increasing explicit behavioral constraint density in persona-based conversational agents can reduce perceived human-likeness even when persona fidelity is maintained. We compare five persona prompting conditions across 360 generated responses organized into 72 paired persona-scenario blocks. A length-matched comparison between a verbose persona specification and a heavy behavioral-constraint condition provides the primary test. AI and independent human smoke evaluations show directional evidence consistent with Constraint Saturation: heavy behavioral constraints tend to reduce human-likeness and autonomy while increasing template-like behavior, without a corresponding collapse in persona fidelity. Full-dataset objective text analysis further shows increased opening-pattern reuse under the heavy-constraint condition. These findings are exploratory rather than confirmatory and motivate a “Rich Persona, Sparse Constraints” design principle for persona-based agents.