Stable Personas: Dual-Assessment of Temporal Stability in LLM-Based Human Simulation

TL;DR

This study evaluates the temporal stability of LLM-based personas using dual assessments, revealing stable self-reports but declining observer-rated persona expression over extended conversations.

Contribution

It introduces a dual-assessment framework and provides empirical insights into the stability of LLM personas across multiple conversations and conditions.

Findings

Self-reports of personas are highly stable across conversations.

Observer ratings show a decline in persona expression during extended interactions.

Persona-instructed LLMs produce stable self-reports, but face regression in observable expression.

Abstract

Large Language Models (LLMs) acting as artificial agents offer the potential for scalable behavioral research, yet their validity depends on whether LLMs can maintain stable personas across extended conversations. We address this point using a dual-assessment framework measuring both self-reported characteristics and observer-rated persona expression. Across two experiments testing four persona conditions (default, high, moderate, and low ADHD presentations), seven LLMs, and three semantically equivalent persona prompts, we examine between-conversation stability (3,473 conversations) and within-conversation stability (1,370 conversations and 18 turns). Self-reports remain highly stable both between and within conversations. However, observer ratings reveal a tendency for persona expressions to decline during extended conversations. These findings suggest that persona-instructed LLMs produce stable, persona-aligned self-reports, an important prerequisite for behavioral research, while identifying this regression tendency as a boundary condition for multi-agent social simulation.