Not in our image: rethinking anthropomorphism in expert chatbot design

Abstract This article interrogates how users interpret and respond to anthropomorphic versus minimalist chatbot designs in legal and regulatory advisory domains, contexts where ambiguity is costly and charm rarely billable. Anchored in ten in-depth interviews and supported by probabilistic simulations employing Bayesian inference and Monte Carlo simulation, the study reveals that interface preferences are far from stylistic whimsy. Instead, they reflect deep-seated expectations rooted in professional roles and interactional demands. Practitioners in law, HR, and compliance consistently gravitate toward pared-down, non-human designs and value transparency, cognitive economy, and semantic precision. In contrast, those operating in branding, UX, or emotionally expressive roles tend to welcome anthropomorphic agents, associating them with engagement and affective resonance. The findings advocate for adaptive chatbot architectures: systems that modulate their aesthetic and communicative cues in response to domain norms, user expectations, and interactional context.