Persona Following Is Not Selective Control: The Neutrality Gap in LLM User Simulation

2026-09-28

AI Summary

This study addresses the non-selective interference of unspecified attributes when conditioning specific traits in LLM-based user simulation, revealing that neutrality declarations fail to suppress such cross-attribute effects and exposing a selective control gap in persona prompting. To explain this interference, we propose a trait-conditioned completion mechanism and define the neutrality gap, establishing a tri-state diagnostic framework that disentangles attribute following from selective control. Through multi-state comparative experiments integrating black-box auditing with semantic and internal representation analyses, we demonstrate that 51%–81% of items remain influenced by target attributes despite explicit neutrality declarations. These findings confirm that successful attribute following does not entail selective control, providing essential diagnostic tools and theoretical foundations for controllable persona generation in large language models.

Abstract

Persona prompting is widely used to construct user simulations with large language models (LLMs), yet it relies on a largely untested assumption: specifying one user attribute should change that attribute alone. We test this assumption and identify a systematic failure of selective control: across all eight black-box LLMs we audit, changing a target attribute also shifts responses on unspecified, non-target attributes. For example, describing a user as more risk-seeking shifts color choices, even though the prompt never mentions color; we term this cross-attribute influence. Semantic, contextual, and internal analyses collectively suggest that models treat a persona prompt as evidence about the user and extend the inferred profile to unspecified preferences, a process we call trait-conditioned completion. We next ask whether explicitly specifying non-target attributes restores selective control. When a non-target attribute is assigned a clear direction, models generally follow the declaration and suppress the target attribute's influence. However, when the same attribute is declared neutral, the target continues to affect choices across all five open-weight checkpoints, even when the model correctly reports the declared state. This disparity, the neutrality gap, demonstrates that successful persona following does not imply selective persona control, which additionally requires keeping non-target attributes stable. We operationalize this distinction with a three-state diagnostic that leaves the non-target attribute unspecified or declares it directional or neutral; because directional tests can be passed by simply following the stated persona, the neutral state reveals failures they miss. In a post hoc analysis of independent items, neutral declarations leave 51-81% of items target-sensitive, against at most 1 of 320 item-pole comparisons under directional ones.

Problem

• Persona Following
• Selective Control
• User Simulation
• Neutrality Gap
• Cross-attribute Influence

Innovation

• persona following
• selective control
• cross-attribute influence
• neutrality gap
• trait-conditioned completion