The study involved 186 graduates assigned as twelve intact cohorts, six per condition, with allocation crossed over institution, facilitator and schedule. Experts, who did not know which programme participants had completed, gave a global rating from 0 to 100 to recruitment-task materials produced on site, under invigilation and without access to AI, both before and after the programmes. This design meant the assessments captured performance under controlled conditions rather than perceptions of learning.

Among the 168 participants with post-test data, representing overall attrition of 9.7 percent, the adjusted difference favoured the AI-supported programme by 5.38 points on the 0–100 scale, with a 95 percent confidence interval of 2.76 to 8.01 and a standardised effect size of d = 0.60. The result was stable across eight different statistical specifications, suggesting it was not an artefact of a particular modelling choice.

By contrast, four newly developed self-report scales showed smaller differences, with effect sizes ranging from d = 0.12 to 0.29, and none remained statistically significant after Holm correction for multiple comparisons. The researchers noted that each scale could only detect differences of about d = 0.4, meaning the study may have been underpowered to identify smaller effects on self-reported outcomes. Notably, the effect on expert ratings exceeded the mean effect on self-report measures by 0.37 standard deviations, with a confidence interval of 0.10 to 0.63.

The authors flagged important limitations. The expert ratings were made without a pre-specified rubric, which limits how they can be interpreted. Component analyses were described as exploratory, and the design cannot separate the contribution of AI itself from the structured practice and feedback embedded in the programme. The findings therefore establish an association between the combined AI-supported, structured approach and better-rated task performance, rather than isolating AI as the active ingredient.

Subject of Research: Effects of an AI-supported structured human resource development programme on expert-rated recruitment task quality among education graduates in China

Article Title: Structured human resource development with AI support is associated with higher expert rated recruitment task quality among education graduates

Article References: He, Y., Haja Maideen, M., & Liang, Z. (2026). Structured human resource development with AI support is associated with higher expert rated recruitment task quality among education graduates. Scientific Reports. https://doi.org/10.1038/s41598-026-76662-y

Image Credits: AI Generated

DOI: 10.1038/s41598-026-76662-y

Keywords: generative AI, graduate employability, human resource development, career guidance, education graduates, expert ratings, cluster allocation, self-report measures, Guangdong China, recruitment tasks, Structured, human