["Journal of Consumer Behaviour, EarlyView. ", "\nABSTRACT\nCompanion Artificial Intelligence, also known as AI Companion, is rapidly gaining prominence in the marketplace, particularly in addressing the growing global concern about loneliness. While these AIs are designed to engage in emotionally rich conversations filled with positive affective expressions, their effects on consumer engagement intentions remain underexplored. This article examines how AI‐expressed positive affect drives consumer engagement intentions. It further investigates whether the extent to which consumers feel lonely influences this process. We employ a sequential, two‐phase mixed‐methods approach, in which the first phase involves in‐depth interviews with participants (N = 19) following their two‐week interactions with Replika. Drawing on insights from this phase, we propose a conceptual model grounded in expectation disconfirmation, computers as social actors, emotions as social information, scarcity theories, and the uncanny valley literature. We test the model in the second phase with three experiments. The first experiment (N = 230), utilizing real interaction, demonstrates that consumers are more likely to engage with AI companions that express (vs. lack) positive affect. The second (N = 250) and third (N = 520) experiments use simulated interactions to demonstrate that feelings of eeriness drive the effect of AI‐expressed positive affect on consumer engagement intentions, and that the affect‐eeriness relationship is moderated (strengthened) only for moderately lonely individuals. These findings contribute novel insights to the literature on AI Companions, loneliness, and eeriness. Additionally, it offers actionable guidance for imparting affective elements into AI design and personalizing interactions based on consumers' loneliness.\n"]