Cross-sectional study links student–GAI interaction to educational interaction among nursing students

A cross-sectional survey of 1,020 nursing students found that student interaction with generative artificial intelligence was positively associated with educational interaction involving teachers, peers and course content. The analysis also identified a partial statistical association through self-regulated learning: students who reported stronger self-regulation and more engagement with generative AI tended to report greater educational interaction. These results do not show that AI caused stronger interaction or better clinical learning, because the researchers measured all three constructs at one point in time.

The survey was conducted from January 6 to January 16, 2025, among undergraduate nursing students at Hunan Medical College in Hunan Province, China. Participants were selected through convenience sampling, and electronic questionnaires were distributed through class WeChat groups by trained class leaders. Current nursing students were eligible, while students on leave and fourth-year nursing interns who were not on campus were excluded. Students reported their experiences with generative AI, educational interaction and self-regulated learning during the preceding semester.

Of the 1,573 questionnaires completed, 553 were removed because responses took less than 100 seconds, had no variation across items or followed repeating patterns that the researchers regarded as indicators of questionable response quality. The 1,020 retained responses included 469 first-year students, 334 second-year students and 217 third-year students. Their mean age was 19.09 years, with a standard deviation of 2.40. Women represented 83.2% of the analytical sample, and 98.8% of participants said they had begun using generative AI for course learning by 2024.

The students’ mean student–GAI interaction score was 14.80 with a standard deviation of 2.27. Mean educational interaction was 70.90 with a standard deviation of 8.15, and mean self-regulated learning was 56.00 with a standard deviation of 7.34. All three measures were positively correlated. Student–GAI interaction correlated with educational interaction at r = 0.378, with self-regulated learning at r = 0.236, and self-regulated learning correlated with educational interaction at r = 0.580; all three p values were below 0.01.

A regression model containing student–GAI interaction and self-regulated learning explained 39.8% of the variance in educational interaction, with an adjusted R² of 0.397. Both predictors showed positive associations with educational interaction. Student–GAI interaction had a standardized coefficient of β = 0.256, while self-regulated learning had a standardized coefficient of β = 0.520; both were significant at p < 0.001. Because the model was cross-sectional, these coefficients describe statistical relationships in the survey data rather than changes or causal effects over time.

Structural equation modeling estimated a direct path from student–GAI interaction to educational interaction with a standardized coefficient of 0.26. The standardized indirect path through self-regulated learning was 0.12. On the unstandardized scale, the total estimated association was 1.36, with a bias-corrected bootstrap 95% confidence interval of 1.14 to 1.57. The direct component was 0.92, with a 95% confidence interval of 0.72 to 1.11, while the indirect component through self-regulated learning was 0.44, with a 95% confidence interval of 0.32 to 0.57.

The indirect component accounted for 32.35% of the total estimated association, while the direct component accounted for 67.65%. The researchers therefore interpreted self-regulated learning as a partial statistical mediator. Their interpretation was specifically statistical: the survey cannot establish whether stronger AI interaction precedes stronger self-regulation, whether self-regulation supports educational interaction, or whether other variables influence all three. The mediation tests used 5,000 bootstrap samples, and the confidence interval for the indirect effect did not include zero.

The mediation pattern differed among the three components of educational interaction. Self-regulated learning accounted for 22.15% of the total estimated association with student-student interaction, 45.41% with student-teacher interaction and 41.54% with student-content interaction. The corresponding unstandardized indirect effects were 0.16, 0.17 and 0.11, each with a bootstrap confidence interval that excluded zero. Direct associations also remained significant: they were 0.56 for student-student interaction, 0.21 for student-teacher interaction and 0.15 for student-content interaction.

These results suggest that the self-regulatory pathway was most prominent in teacher-related and content-related interaction in the statistical model. For peer interaction, the direct component represented 77.85% of the total association, compared with 54.59% for student-teacher interaction and 58.46% for student-content interaction. The finding does not mean that AI replaces human contact or that self-regulation has no relevance to peer work. It instead shows how the measured associations were distributed across the three dimensions in this sample of nursing students.

Several technical features qualify the findings. The four-item Student–GAI Interaction Scale had a Cronbach’s alpha of 0.780, with a confirmatory-factor root mean square error of approximation of 0.066, comparative fit index of 0.992 and incremental fit index of 0.992. The 18-item Educational Interaction Scale had an alpha of 0.925 but weaker model fit, including a root mean square error of approximation of 0.099, comparative fit index of 0.872 and Tucker–Lewis index of 0.852. The 15-item self-regulated learning questionnaire had an alpha of 0.919 and a comparative fit index of 0.896.

The study has important limitations. It surveyed students at one Chinese medical university, limiting generalizability, and relied on self-reported measures that could be affected by social-desirability bias. The analysis did not adequately control for prior academic achievement, AI literacy, learning motivation, digital access, the frequency and purpose of AI use, platform type, task complexity, teacher-guided versus self-initiated use, course type, teacher support or institutional permissions. The authors also point out that a high level of AI interaction may reflect adaptive use, convenience, overreliance or course requirements, which were not distinguished. The findings are therefore promising associations from a cross-sectional study, not evidence that introducing generative AI necessarily improves nursing education.

Subject of Research: Cross-sectional mediation analysis of student–generative AI interaction, self-regulated learning and educational interaction among nursing students

Article Title: Generative artificial intelligence use and interpersonal educational interaction among nursing students: self-regulated learning as a statistical mediator

Article References: Li, J., Xu, E., Li, L., Huang, L., Pu, Y., & Huang, R. (2026). Generative artificial intelligence use and interpersonal educational interaction among nursing students: self-regulated learning as a statistical mediator. Frontiers in Psychology, 17, Article 1864535. https://doi.org/10.3389/fpsyg.2026.1864535

Image Credits: AI Generated

DOI: 10.3389/fpsyg.2026.1864535

Keywords: generative artificial intelligence, nursing education, nursing students, student–AI interaction, educational interaction, self-regulated learning, mediation analysis, cross-sectional survey, student-teacher interaction, student-content interaction, Generative, artificial