Artificial Intelligence (AI) chatbots are increasingly used to compose emotionally supportive messages, yet research has focused primarily on how these messages are received rather than on how relying on AI affects the sender. Theories of cognitive offloading and motivated empathy avoidance suggest that delegating empathic work to AI may reduce emotional engagement. The present study examined whether different degrees of human involvement in AI-assisted message generation shape users’ emotional and cognitive responses toward a distressed friend. In a 4-condition × 4-scenario within-subjects experiment, 267 adults (Mage = 45; 49% female) responded to scenarios involving a close friend’s distress. Each scenario was paired with one message-generation format: copying an AI-generated message, choosing among AI-generated options, editing an AI-generated message, or writing a message without AI assistance. After composing each message, participants reported compassion, empathy, cognitive engagement, closeness, effort, and self-efficacy. Linear mixed-effects models showed that message-generation condition significantly predicted all outcomes. Greater human involvement produced higher responses overall: passive AI use yielded the lowest outcomes, whereas choosing AI options showed only modest gains. Notably, editing an AI-generated message were statistically indistinguishable from fully self-written messages. Within the edit condition, greater modification of the AI template was associated with higher compassion, closeness, cognitive empathy, and cognitive engagement. These findings show that human involvement—not AI assistance itself—drives emotional and cognitive self-effects in AI-supported communication. Active engagement preserves the psychological benefits of empathic responding, whereas passive reliance diminishes them.

Authors:
Daniela Kuhr, University of Lübeck, Germany
Olga Stavrova, University of Mannheim, Germany

About the Presenter(s)
Daniela Kuhr is currently pursuing a doctoral degree in Psychology. Her research examines how advances in artificial intelligence and large language model chatbots influence human cognition, emotions, relationships, and social well-being.

See this presentation on the full schedule – Tuesday Schedule

Bookmark