This study investigates how two under-explored conversational features – AI-expressed emotions and grounding through lexical entrainment – jointly foster consumer trust, decision comfort and service satisfaction in human–AI service encounters. By integrating feeling-as-information and trust theories, it addresses the persistent gap between AI efficiency and users’ perception that “the machine doesn’t understand or care.”
Four preregistered online experiments on the Credamo platform (total N = 737) tested the main, mediating and interactive effects of the two features across coffee-ordering, financial-consulting and psychological-counseling scenarios. Studies 1–2 manipulated each feature independently; Studies 3–4 employed a 2 × 2 design to examine their synergy and boundary conditions.
Both AI-expressed emotions and grounding independently elevated service satisfaction (e.g. F(1, 198) = 18.21 for emotions; F(1, 98) = 66.64 for grounding) and interacted to boost perceived ability when grounding cues were weak. A sequential double-mediator model was supported: each feature increased perceived ability/integrity → decision comfort → satisfaction. Interaction analyses confirmed moderated mediation on ability (F(1, 236) = 13.17).
This article proposes strategies that offer actionable recommendations for designing AI service agents that are more effective and better aligned with consumer expectations.
The research (1) unites affective and cognitive trust pathways to explain consumer responses to conversational AI, (2) demonstrates that “warmth” (emotional expression) and “understanding” (grounding) are complementary rather than substitutive design levers and (3) reveals context-dependent boundary conditions, showing integrity gains from emotions only in risk-laden services. These insights extend the AI-marketing theory and provide actionable guidance for designing empathetic yet reliable service agents.
