The illusion of empathy: Evaluating AI-generated outputs in moments that matter

This study investigates how anthropomorphism and source attribution shape perceptions of creativity, authenticity, and moral respect in emotionally significant communication. Drawing on theories of human-machine communication and symbolic value, we examine whether messages are evaluated differently depending on who – or what – is believed to have authored them. Across two experimental studies, we manipulated both the emotional context (childbirth vs. terminal illness) and the attributed message source (close friend, florist, Google, or ChatGPT). Study 1 used a within-subject design to compare message evaluations before and after source disclosure; Study 2 disclosed the source at the outset. Results show that emotional proximity significantly enhances perceived communicative value, while attribution to artificial or emotionally distant sources reduces it. Anthropomorphic cues temporarily elevate AI evaluations but collapse upon disclosure, particularly in high-stakes contexts. Attribution to ChatGPT led to the steepest declines in authenticity and moral respect, underscoring the symbolic and ethical limitations of AI in relationally charged settings. Our findings contribute to the literature on AI-human interaction by theorizing anthropomorphism as a double-edged attributional mechanism and offer practical insights for the deployment of generative AI in domains requiring emotional sensitivity, care, and symbolic coherence.