Self-Stigma Is Not a Monolith, but Generic Empathy Is: Persona-Conditioned LLM Support for People Who Use Drugs

2026-06-22

AI Summary

This study addresses the limitation of current dialogue systems that treat self-stigmatizing expressions from individuals who use drugs as homogeneous signals, thereby failing to deliver effective support. The authors propose the first latent profile analysis (LPA)-based typology distinguishing four distinct self-stigma personality profiles. Integrating a sequential Bayesian classifier with a recurrent neural network, the framework accurately identifies an individual’s profile from limited posting history—achieving a macro F1-score of 0.74 with fewer than 30 posts—and leverages this insight to guide large language models in generating personalized responses. While personalized interventions demonstrate greater efficacy in promoting behavioral change, they reveal a tension with generic empathetic strategies in clinical evaluations, highlighting a critical trade-off in the design of supportive conversational agents.

Abstract

Self-stigma predicts treatment avoidance and disengagement among people who use drugs (PWUD), yet conversational systems aiming to provide support typically treat self-stigma expression as a uniform signal. We present a three-phase, proof-of-concept study of a persona-aware approach to LLM support. Latent Profile Analysis (LPA) on indicator-level features from 1,174 self-stigma expressors on Reddit yields a four-persona typology validated against held-out behavioral and linguistic features. Sequential Bayesian and recurrent neural classifiers recover these personas from limited posting histories, substantially outperforming batch and few-shot LLM baselines (macro-F1 = 0.74 at 30 posts). Evaluation by eight clinical experts across three contemporary LLMs revealed a misalignment: persona-matched responses successfully achieved targeted behavioral shifts, yet raters holistically preferred the generic empathy of the persona-neutral baseline. Our findings suggest that holistic empathy judgments and clinically-aligned response design can pull in opposite directions, and that evaluating LLM-based stigma support requires rubrics capable of decomposing the two.

Problem

• self-stigma
• people who use drugs
• conversational systems
• empathy
• persona heterogeneity

Innovation

• persona-conditioned LLM
• self-stigma typology
• Latent Profile Analysis
• empathy misalignment
• behavioral shift