Answers are grounded in this paper and cite their sources.
LLM-powered conversational agents increasingly use empathetic, agreeable styles to build rapport, yet whether this fosters genuine trust remains unclear. This study examined how three AI modes (Empathic Correction, Sycophantic Agreement, Neutral Control) shaped perceived empathy, authenticity, intimacy, and trust, testing whether authenticity mediates the empathy-trust link. Fifty participants completed a within-subjects experiment (150 observations) with a conversational AI across all three modes. MANOVA revealed significant effects of AI mode on empathy, authenticity, and intimacy, but not trust. Mediation analysis revealed a Trust Paradox: Sycophantic Agreement increased trust via heightened empathy while decreasing it via diminished authenticity; these effects cancelled at the surface, masking two offsetting mechanisms. Sycophancy maximized perceived warmth while minimizing genuineness; Empathic Correction showed the reverse. Findings suggest AI systems prioritizing agreement over honesty risk building trust on an unstable foundation, and that surface-level metrics can obscure meaningful human-AI trade-offs. Implications for trustworthy conversational AI design are discussed.
No takes yet. Share an insight, caveat, or question.
Zaman et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: