Affect-aware human-agent conversations: towards empathic LLM interaction
Abstract Large Language Models can produce supportive responses, but their apparent empathic behavior is usually grounded only in explicit textual input. This paper presents Empathic Extended Prompting a real-time affect-aware framework that enriches human--LLM conversation with facial affective descriptors to support perceived empathic attunement. The system integrates FaceReader-based affective descriptors into the LLM context, including emotional intensity, valence, and arousal. We evaluated the approach in a within-subjects study with N=20 participants, comparing an affect-aware experimental condition with an empathic text-only baseline. Results favored the affect-aware condition on key perceived qualities of the interaction, particularly emotional intelligence and mood responsiveness. These findings suggest that real-time affective context may add relational grounding to LLM conversations without reducing usability or eliciting uncanny-like perceptions.
Authors
- Emanuele Frontoni (ORCID: https://orcid.org/0000-0002-8893-9244)
- Lorenzo Stacchio (ORCID: https://orcid.org/0000-0002-9341-7651)
- Alessandro Galdelli (ORCID: https://orcid.org/0000-0002-4140-6424)
- Andrea Gaggioli (ORCID: https://orcid.org/0000-0001-7818-7598)
- Maurizio Mauri (ORCID: https://orcid.org/0000-0003-2229-9384)
- Andrea Ubaldi
Institutions
- Università Cattolica del Sacro Cuore (IT)
- Marche Polytechnic University (IT)
- University of Macerata (IT)
Publication Details
- Journal
- Journal of Ambient Intelligence and Humanized Computing
- Published
- 2026-11-17
- DOI
- https://doi.org/10.1007/s12652-026-05141-z
- Primary Topic
- Emotion and Mood Recognition
- Type
- article
- Field-Weighted Citation Impact
- 0.00