Perceived dehumanization in Human–AI service encounters: A framework for understanding chatbot-mediated customer experience

The rapid integration of artificial intelligence into service interactions is fundamentally altering the ways in which organizations engage with their customers. Conversational agents, including chatbots, are now widely deployed to handle inquiries, guide purchase journeys, and assist users across a range of tasks. While these technologies enhance service efficiency and availability, they raise substantive questions about the relational quality of digital experiences. In certain cases, customers may feel that an interaction fails to acknowledge their individuality, an experience the literature has begun to describe as perceived dehumanization. Despite growing scholarly interest in this phenomenon, existing theoretical frameworks remain fragmented and do not adequately account for the mechanisms through which such perceptions emerge in AI-mediated interactions. This article proposes a processual architecture to explain the emergence of perceived dehumanization in human-AI service encounters. The framework rests on four sequential mechanisms with explicit transition conditions: expectation activation, interactional evaluation, attribution formation, and identity-level interpretation. Perceived dehumanization is defined as a reflexive experiential state manifesting through four analytically distinct indicators: perceived lack of personalization, perceived emotional attenuation, perceived interactional coldness, and perceived digital isolation. A crossed typology (passive/active intentionality x felt/functional reception) and five contextual moderators complete the architecture. Three contributions are advanced. The framework specifies the precise conditions under which perceived dehumanization emerges, stabilizes, or does not occur, a specification irreducible to existing variance-based approaches. It generates five falsifiable propositions whose predictions contrast structurally with those of prior models. It provides a differentiated diagnostic grid by configuration, absent from prior conceptualizations.