<p>While generative AI (GenAI) tools such as DeepSeek are increasingly embedded in higher education, existing research has predominantly focused on accuracy, bias, and adoption, with limited attention to the nature of human–GenAI interaction in learning. This study examines how nursing students interact with GenAI and how distinct interaction patterns shape cognitive, behavioral, and affective outcomes, as well as ethical risks in professional training contexts. A cross-sectional mixed-methods design was employed, integrating a rapid review, survey data (<i>n</i> = 401), and qualitative interviews (<i>n</i> = 30). Quantitative data were analyzed using descriptive statistics, correlations, and regression, while qualitative data underwent thematic analysis to identify recurring interaction patterns. Four dominant interaction patterns emerged: (1) passive consumption, characterized by uncritical acceptance of outputs; (2) efficiency-driven use, where GenAI functions as a productivity shortcut; (3) dependent interaction, reflecting overreliance and reduced self-directed learning; and (4) critical engagement, involving verification and reflective use. These patterns are associated with divergent outcomes. Efficiency-driven and dependent interactions were associated with greater perceived task efficiency and performance, but were also associated with lower critical thinking, reduced motivation, and increased academic dishonesty. In contrast, critical engagement supports deeper learning but demands higher AI literacy and effort. Ethical risks—including misinformation, fabricated references, and misalignment with clinical standards—are most pronounced under passive and dependent use. The findings suggest that educational outcomes are shaped not merely by GenAI adoption but by interaction quality. This study advances a shift toward interaction-centered frameworks, emphasizing the need for structured, critical, and ethically grounded GenAI engagement in higher education.</p>