Key points are not available for this paper at this time.
In the digital economy, AI companion platforms integrate upstream AI resources, platform governance, and downstream user engagement, yet sustained emotional interaction does not necessarily translate into paid conversion. To examine this engagement–monetization paradox, this study proposes the Emotion–Barrier–Behavior (EBB) framework as a context-specific integration of relational value, platform frictions, post-adoption continuance, and payment evaluation. A mixed-method design was employed. First, 12,144 online reviews of Xingye, a Chinese AI companion platform, were analyzed through text mining and latent Dirichlet allocation to identify context-specific emotional experiences and usage barriers. Second, structural equation modeling was conducted using survey data from 321 adult respondents. Perceived emotional utility was positively associated with continuance intention but had no significant direct association with willingness to pay; a significant indirect association was observed through continuance intention. Perceived usage barriers were negatively associated with willingness to pay but positively associated with continuance intention and emotional utility in the hypothesized model. However, these barrier-related associations became nonsignificant after controlling for an unmeasured latent method construct and varied across barrier dimensions; they should therefore be interpreted as method-sensitive baseline associations rather than robust structural effects. These cross-sectional findings suggest that continued use and payment intention may diverge in AI companion platforms. The study therefore frames sustainable monetization not as revenue persistence alone, but as commercially viable value capture aligned with SDG 3-related user well-being and SDG 12-related responsible digital consumption and platform practices. These SDG-related outcomes were not directly measured, and broader generalization to AI-driven digital platform ecosystems requires further validation.
Lu et al. (Fri,) studied this question.