The impact of generative artificial intelligence on academic development of Chinese students in humanities and social sciences

Synthesis: This large-scale survey of humanities and social sciences (HSS) students in China examines how Generative AI reshapes academic development across four dimensions: usage patterns, effects on learning processes and performance, challenges, and preferred curricular integration approaches. Over half of respondents reported enhanced learning motivation, independent thinking, and Creativity, though a substantial minority saw little change or decline. A larger majority reported academic performance gains, though the authors caution these may partly reflect limitations in conventional Assessment practices.

Key Findings

Usage patterns: HSS students use GenAI primarily for writing assistance, information synthesis, and idea generation — tasks that align closely with HSS learning outcomes expressed through written and interpretive forms. Variations emerged by discipline and duration of GenAI experience, with modest gender differences.

Learning processes: More than half of students perceived enhanced motivation, independent thinking, and creativity. However, a substantial minority reported little change or even decline, suggesting that Personalized Learning approaches to GenAI integration may be necessary.

Academic performance: A notably larger majority reported academic gains, though these may partly reflect assessment practices ill-equipped to distinguish AI-assisted from independent work — a challenge related to Academic Integrity.

Challenges: Limited accuracy and Over-Reliance emerged as the most pressing concerns. While an overwhelming majority valued ethical considerations, only slightly more than half were satisfied with Privacy protections. Students favored partial or optional GenAI integration into curricula.

What this means for practice

• Learners. Treat GenAI as an entry point rather than an endpoint: nearly 70% reported high or very high willingness to use it for initial exploration of innovation-related problems, and more than half used it for research proposal development and brainstorming, so reserve interpretive judgment and argument construction for yourself.
• Learners. Check output against sources before building on it: limited accuracy was the most cited challenge (81.42%), 67.10% reported encountering inaccurate information frequently or very frequently, and only a further 31.04% encountered it occasionally.
• Learners. Monitor how your own thinking changes: 65.03% reported increased active-learning motivation but 12.57% reported a decline, and views on independent thinking split between 22.08% significant improvement and 34.21% improvement against 23.39% decline and 2.73% significant decline — build tool-free stretches into writing and analysis work.
• Administrators. Tailor guidance by discipline instead of issuing one-size-fits-all policy: 65.68% agreed GenAI aligns with their discipline while 31.26% were neutral, and privacy satisfaction was divided (56.94% satisfied, 37.27% neutral, 5.80% dissatisfied), pointing to field-specific faculty development rather than a single institution-wide mandate.
• Administrators. Repair assessment validity before reading reported gains as learning: a notably larger majority reported academic performance gains, but the authors caution these may partly reflect conventional Assessment practices that cannot distinguish AI-assisted from independent work, making nuanced Academic Integrity policy a prerequisite rather than an afterthought.

Limitations

• Self-report is the only outcome measure and the authors expect bias in both directions — students may over- or under-estimate the learning benefits of AI use — while the exclusive use of one anonymized questionnaire constrained interpretive depth.
• The sample is imbalanced: of 915 valid respondents, 736 were female (80.44%) and education-related majors accounted for 48.31%, which limits external validity and the precision of the gender and cross-discipline contrasts.
• No causal or longitudinal claim is supportable: the comparisons across groups with different durations of GenAI use were between different participants, so the design cannot capture within-individual change over time.
• Coverage limits: roughly 1000 responses collected anonymously in January 2025 through Wenjuanxing were reduced to 915 valid HSS responses, and discipline-level analysis was restricted to four fields (education, economics and management, arts, and law) because of sample-size limits.

Citation

Fan, L., & Liu, F. (2026). The impact of generative artificial intelligence on academic development of Chinese students in humanities and social sciences.