Mindful Machines: Understanding How AI's Theory of Mind Capabilities Influence Consumer Response to Product Recommendations

Abstract

Previous research has focused on AI's analytical abilities and how people attribute mental states to AI, but less is known about how AI can accurately respond to human thoughts and emotions. This ability, known as the theory of mind (ToM), may influence how people react to AI recommendations. Based on expectancy violation theory, this study examines how AI's ToM influences users' recommendation acceptance. Six experiments demonstrate that an AI agent's high (vs. low) ToM capabilities lead to higher recommendation acceptance by increasing users' perceptions of the AI agent's social presence. Moreover, we find that product type (virtue vs. vice) moderates the relationship between AI agents' ToM capabilities and recommendation acceptance: for virtue products, a high-ToM agent generates higher acceptance, whereas for vice products, a low-ToM agent generates higher acceptance. Our results are consistent across various consumption scenarios (outfit, restaurant, trip destination, food and snacks choices), different ToM manipulations (text-based and audio-based conversations), different product type manipulations (different products and varying advertising messages about the same product), and diverse participant samples (Chinese and US participants). Our findings contribute to research on AI–human interactions and offer practical implications for designing AI recommendation systems across diverse consumption contexts.