University student preferences for generative AI subscriptions

Abstract

Subscription design for generative artificial intelligence requires an understanding of which capabilities users value and which trade-offs they accept. This paper examines university student preferences for model capability, context capacity, privacy protection, and monthly price. The empirical material comes from a Chinese competition research report describing ten exploratory interviews and a choice-based conjoint survey with 828 retained respondents. The report applies conditional logit and latent class models to subscription choices that include a free alternative. Its reported willingness-to-pay estimates suggest that expert-level capability and enhanced privacy are important sources of perceived value, whereas context expansion does not show a stable aggregate premium. Three reported preference segments further suggest that capability demand, budget sensitivity, and privacy concerns should be considered jointly. Historical product examples from Anthropic and MiniMax illustrate alternative ways to translate technical capacity into service differentiation. Because individual choice records are unavailable and some statistical entries in the source report show discrepancies, the findings are treated as exploratory reported evidence; precise effect sizes and segment membership require replication. The paper develops implications for capability-based product tiers, credible privacy commitments, and task-specific context provision, while distinguishing candidate subscription prices from empirically optimized prices. Ultimately, this study provides actionable insights for technology developers aiming to align their subscription models with the specific needs and financial constraints of the academic user demographic.