Proceedings of the Annual Meeting of the Cognitive Science Society

Large language models, particularly in their conversational deployments, invite users to relate to them as if they possessed inner perspectives. This paper conceptualizes "perceived subjectivity" as a socially consequential interactional effect co-constituted in human-AI encounters rather than an intrinsic property of AI or the user's attribution error bias. Through interdisciplinary synthesis, we distinguish perceived subjectivity from phenomenal consciousness, showing how interface features, prompts, histories and user dispositions jointly stabilize impressions of mindedness. This conceptualization shifts the focus from metaphysical debates about machine consciousness and sentience to the interactional dynamics that reorganize trust, emotional dependence, and epistemic authority. By treating perceived subjectivity as a design-mediated effect, we provide pragmatic implications for implementing "productive friction", deliberate safeguard mechanisms that restore critical distance and preserve user autonomy. This approach aims to enable designers and regulators to target interactional conditions to reduce manipulation, deception and epistemic deference risks in affective conversational systems.