Emotional Plausibility vs. Emotional Truth: Designing Against Affective Misinformation in Conversational AI

Conversational AI systems increasingly simulate emotional presence, yet remain fundamentally unfeeling. This paper argues that such systems, through their design, propagate affective misinformation: they feel understanding, but do not understand. Drawing on HCI, AI ethics, media studies, and affect theory, we introduce a conceptual distinction between emotional plausibility and emotional truth, and demonstrate how design features like simulated typing, memory recall, affirming tone, and other anthropomorphic cues create the illusion of relational care. We conduct a cross-system design audit of leading chatbots, synthesize real-world harms, and propose five normative principles for literacy-first design. These include counter-anthropomorphic patterns that foster conceptual clarity, and design interventions that aim to mitigate relational misbelief and affective amplification in emotionally charged contexts. Our contributions advance the ethics of AI interface design by foregrounding affective misperception as a site of epistemic risk: one that must be addressed as AI systems become more persuasive, pervasive, and humanlike.