One Persona, Many Cues, Different Results: How Sociodemographic Cues Impact LLM Personalization

TL;DR

This study examines how different sociodemographic cues used as personas influence LLM responses, revealing significant variability and cautioning against reliance on single cues for bias analysis.

Contribution

It compares multiple sociodemographic cues across various LLMs and tasks, highlighting the impact of cue variation on bias and response differences.

Findings

Responses vary significantly across different sociodemographic cues.

Single cues can lead to misleading conclusions about bias.

Cue choice affects the interpretation of LLM behavior and fairness.

Abstract

Personalization of LLMs by sociodemographic subgroup often improves user experience, but can also introduce or amplify biases and unfair outcomes across groups. Prior work has employed so-called personas, sociodemographic user attributes conveyed to a model, to study bias in LLMs by relying on a single cue to prompt a persona, such as user names or explicit attribute mentions. This disregards LLM sensitivity to prompt variation and the rarity of some cues in real interactions (external validity). We compare six commonly used persona cues across seven open and proprietary LLMs on four writing and advice tasks. While cues are overall highly correlated, they produce substantial variance in responses across personas that can change findings on persona-induced differences and bias. We therefore caution against claims based on single persona cues, especially when they are overly explicit and have low external validity.