Cluster-based analysis of university students’ perceptions of generative AI across STEM and non-STEM fields

Generative artificial intelligence (GenAI) is increasingly embedded in university learning, particularly in STEM and related disciplines. Students commonly use tools such as ChatGPT, Grammarly, and Quizlet for writing support, problem-solving, and rapid feedback, yet their evaluations of these tools vary substantially. This study identifies perception-based student profiles and examines how demographic and user characteristics relate to perceived challenges, benefits, and effectiveness of GenAI across STEM and non-STEM fields. Survey responses from 237 U.S. university students were analyzed using a two-stage clustering approach that combined hierarchical clustering (Ward’s method) and K-means clustering across 23 retained perception items. The three-cluster solution was retained based on the hierarchical agglomeration structure and substantive interpretability of the resulting profiles, and profile distinctiveness was subsequently evaluated using a four-item ethical-concern scale excluded from cluster formation. Three profiles emerged: optimists (41.4%), neutrals (37.6%), and skeptics (21.1%). Optimists reported high perceived benefits and effectiveness with comparatively lower concerns, neutrals reported mixed perceptions that balanced perceived benefits with ongoing concerns, and skeptics reported stronger ethical and reliability concerns alongside lower perceived educational value. Multinomial logistic regression showed that AI-use frequency was the strongest independent correlate of profile membership, while gender additionally distinguished optimists from neutrals. The findings support differentiated institutional strategies emphasizing responsible use, clearer guidance, and trust-building.