Conversational AI and the Reconfiguration of Consumer Purchase Decisions:An Anthropomorphic Trust Calibration Approach to AI-Powered Marketing Communication

As generative conversational agents become embedded in brand–consumer dialogic interfaces, scholarly understanding of how such agents shape purchase decisions remains fragmented, under-theorized, and disproportionately focused on utilitarian service recovery rather than persuasive marketing communication. To address this gap, the present research develops and tests the Anthropomorphic Trust Calibration (ATC) model—an integrative framework that links AI-powered marketing communication (AIPMC) characteristics (linguistic alignment, perceived humanness, response latency, and identity transparency) to consumer trust calibration, parasocial engagement, and, ultimately, purchase intention through a partially serial mediation chain. A sequential explanatory mixed-methods design was implemented in three pre-registered phases. In Phase 1, 12,486 publicly available social media conversations involving brand-deployed conversational agents were analyzed through BERTopic topic modeling and VADER sentiment pipelines, yielding five empirically distinct interaction archetypes. In Phase 2, a cross-sectional survey of consumers (n = 684) who had interacted with brand conversational agents in the preceding 90 days was analyzed using partial least squares structural equation modeling (PLS-SEM) with bootstrapped indirect-effect inference. In Phase 3, two pre-registered 2 × 2 scenario-based experiments (N_total = 1,142) isolated the causal effects of anthropomorphic cue intensity and identity transparency on trust calibration and conversion behavior. Across all phases, results demonstrate that linguistic style matching (β = 0.42, p < .001) and earned anthropomorphism achieved through identity disclosure paradoxically produced the highest purchase conversion (Δ = +14.7 percentage points relative to undisclosed human-routed conditions). The ATC model explains 58.3% of variance in purchase intention and surpasses representative baseline extensions of the Technology Acceptance Model by ΔR² = 0.19. Theoretically, the study advances a calibration-centric account of human–machine persuasion that displaces the dominant “more anthropomorphism is better” heuristic and re-specifies the role of identity disclosure as a positive, not merely defensive, lever of brand credibility. Managerially, the findings supply actionable heuristics for the ethical deployment of conversational AI in marketing communication and propose a Transparency–Alignment Matrix (TAMx) for portfolio-level configuration of AIPMC systems.