Survey of 493 Chinese university students links academic pressure and peer influence to AI misuse

Zhang and Chen investigate why university students may use artificial intelligence in ways that cross academic boundaries and what consequences accompany that behaviour. Their survey draws on the fraud triangle, deterrence and social cognitive perspectives, combining opportunity, pressure, peer context, institutional policy and individual confidence in one structural model. The authors define AI misuse through behaviours such as generating central content or arguments, inserting AI output directly into parts of an assignment, and using AI revisions without checking sources or credibility. This operational definition is broader than a single plagiarism rule.

The researchers sampled nine higher-education institutions in Jiangsu Province: two universities in the 985 group, one 211 institution, four regular undergraduate institutions and two vocational or technical colleges. They received 520 questionnaires from 541 distributed, a reported response rate of 96.12%. After excluding 27 invalid responses, the analytic sample contained 493 students, giving a reported effective rate of 91.13%. Participants completed five-point self-report measures, and the authors used structural equation modelling to examine relationships among proposed antecedents, misuse and self-reported innovation ability.

Academic pressure showed the strongest positive association with AI misuse, with a standardized coefficient of .380. Peer influence was also positive at .268, as was perceived ease of use at .172; the paper reports p values below .001 for all three. Policy deterrence was negatively associated with misuse at -.210, and academic self-efficacy at -.199, again with p values below .001. AI misuse was associated with lower innovation ability at -.423. These estimates describe relationships within the fitted survey model; they are not experimental treatment effects.

Several limitations narrow the conclusion. Data were collected at one time, so the study cannot show that pressure or peers caused misuse, or that misuse later reduced innovation. Every central construct relied on self-report, making recall, interpretation and social-desirability bias possible. Innovation ability was not assessed through independently scored creative work, longitudinal performance or workplace outcomes. The nine institutions broaden the Jiangsu sample but remain within one province. Cultural setting, assessment rules, disciplinary mix and local AI policies may differ in Hong Kong and elsewhere.

For Hong Kong higher education, the findings suggest that integrity policy should be paired with course design and learner support. A clear prohibition may deter misuse, but heavy workload, ambiguous assessment purpose and low confidence can still make outsourcing attractive. Courses can state which AI actions are permitted, require source verification and process notes, and provide staged feedback before high-pressure deadlines. Students can practise solving or drafting before consulting AI, compare generated claims with disciplinary sources, and explain which suggestions they rejected. Institutions should test whether rules are understood consistently across languages and programmes.

The paper supports a risk model, not a verdict about individual students or AI use in general. Pressure, peers, convenience, policy and self-efficacy are plausible intervention points, while the reported innovation relationship needs longitudinal and performance-based replication. Hong Kong studies could combine anonymous surveys with consented usage traces, independently scored projects and follow-up measures, while protecting students from punitive inference based on patterns alone. The most proportionate response is to reduce avoidable pressure, strengthen capability and make expectations concrete, then evaluate whether misuse and independent learning change together.