[Purpose/significance] Systematically identifying the hierarchical structure and evolutionary patterns of users′ role cognition of Generative Artificial Intelligence (GenAI) in academic topic selection not only enriches the theoretical framework of human-AI collaboration but also offers implications for layered interaction design of GenAI and for cultivating students′ academic innovation capabilities in the AI for science era. [Method/process] Based on the Computers Are Social Actors theory and role theory framework, this study employed a longitudinal interpretative phenomenological analysis to track 13 university students′ interactions with GenAI and their cognitive experiences during academic topic selection over three months through 36 in-depth interviews. [Result/conclusion] This study, drawing on Bloom′s taxonomy, developed a multi-level 3M (Minion, Mate, and Mentor) role spectrum comprising four hierarchical layers of role type, role dimension, specific role, and role manifestation. Users′ role perceptions of GenAI exhibit complex diversity; GenAI′s role positioning undergoes dynamic evolution as research progresses, and users′ role perceptions of GenAI present three dynamic evolutionary patterns: elevation, expansion, and adaptation.