Abstract
<p>As people increasingly turn to artificial intelligence (AI) for empathy, new modes of human–AI empathic interaction have emerged. While prior research on empathic AI has focused almost exclusively on text-based empathy, we explored how vocal delivery of empathetic messages might enhance or undermine AI versus human empathy. Across five studies (N = 1,584) we examined how the mode of expression (voice vs. text) and the empathy source (human vs. AI) jointly shape perceptions of empathic quality and experiences of feeling heard across different interaction contexts. In interactions involving pre-recorded empathic messages, vocal delivery reduced perceptions of quality and experiences of feeling heard for both human and AI empathizers, though the reduction was stronger for AI. This reduction was eliminated for human empathizers by giving empathy recipients the option to first opt in or consent to receiving empathy from a human empathizer, though the reduction remained for AI voices in this pre-recorded context. The reduction for AI voices also disappeared, and sometimes reversed into a benefit, in more immersive and personally relevant interaction contexts such as extended, live interaction with an AI chatbot. We further explored how the mode of expression impacts feelings of uncanniness, and the mediating role uncanniness plays in shaping feelings of being empathized with and feeling heard. Together, these findings demonstrate that whether voice helps or hurts AI empathy critically depends on the interaction context, and highlight uncanniness as one factor that can undercut spoken empathy in some contexts.</p>