Perceived classmate GenAI use was associated with lower trust, while perceived AI literacy attenuated the direct link

Zhang, Geng and Qi investigate how students evaluate a peer whom they believe uses generative AI. The researchers collected online questionnaires in April 2026 from students at Henan Normal University and Henan Women's Vocational College. The survey distributed 450 questionnaires, of which 406 were retained as valid, a 90.2% valid-response rate. Participants were aged 17 to 23, with a mean age of 19.39; 88 were male and 318 were female. Each participant recalled a specific classmate who had used AI and rated that person's AI use, AI literacy, warmth, competence and trustworthiness.

The distinction between reality and perception is central. The study did not inspect the classmate's prompts, assignments or account history, and it did not test that person's actual AI literacy. All variables describe one observer's perception of one recalled target. In the bivariate results, perceived AI use correlated negatively with interpersonal trust at about r = -0.19. Perceived warmth and competence correlated positively with trust at about r = 0.71 and r = 0.68 respectively. These associations describe how judgments clustered in this sample, not verified behaviour by the recalled classmates.

The modeled paths were consistent with the proposed social-perception account. Greater perceived AI use was associated with lower perceived warmth, with a standardized coefficient of -0.15, and lower competence, with a coefficient of -0.18. Warmth and competence were in turn positively associated with trust, at 0.48 and 0.29. A negative direct association between perceived AI use and trust remained at -0.10. The paper therefore reports indirect statistical paths through warmth and competence, but a cross-sectional mediation model does not establish a temporal or causal mechanism.

Perceived AI literacy of the target moderated the direct association. At low perceived literacy, the AI-use-to-trust slope was approximately -0.23; at the mean it was -0.10; and at high perceived literacy it was 0.03 and not statistically significant. However, the indices of moderated mediation were not significant. The careful interpretation is that perceived literacy attenuated the remaining direct association in this model, not that literacy removed the warmth and competence pathways or made an AI-using student objectively more trustworthy.

Several design limits prevent a causal headline. AI use was not randomized or manipulated, there was no before-and-after trust measure, and participants may have recalled a particularly visible, disliked or suspected peer. A question about extensive AI use may also cue concerns about outsourcing. The sample came from two mainland Chinese institutions, was predominantly female and used a single-source recall method. Reverse causation, common-method bias and unmeasured academic or interpersonal characteristics remain plausible explanations.

For Hong Kong, the study raises a practical question about disclosure and peer collaboration rather than a basis for punishing AI users. Universities could test short disclosure statements that distinguish brainstorming, feedback and verification from unacknowledged completion, then examine whether they improve calibrated trust. Replication should use standardized scenarios or observed practice, include Cantonese, Chinese and English settings, broaden disciplines and gender representation, and measure trust longitudinally. The defensible conclusion is narrow: perceived classmate GenAI use was associated with lower reported trust in this sample, while causation and generalizability remain unproven.