Should AI adapt its empathy to match each person's personality?

Chatbots today offer generic empathy regardless of who they're talking to. This work introduces PersonaEmp, a dataset of long-term user-AI interactions paired with personality profiles, and PereGRM, a reward model that learns to evaluate empathetic responses tailored to specific traits. The approach consistently outperformed baselines, suggesting LLMs can learn to adjust emotional strategies—not just say the right words, but say them the right way for each person.