Answers are grounded in this paper and cite their sources.
This two-site mixed-methods study examined a four-session classroom career-guidance package that limited generative AI to verification, counterexamples, uncertainty checks and action-plan review. Four vocational-college classes participated (N = 240; paired n = 231) without a comparison group or prospective feasibility thresholds. Retention was 96.3%; core-work completion was 81.3%–85.0% of enrolment. Technical recovery, incomplete work and ambiguous declining AI use identified redesign needs. Observed score change was +20.97 on an assessment closely aligned with instruction, including task materials listed in the instructional knowledge base; this is non-causal attainment, not demonstrated transfer. A separate validation cohort showed a seven-day change of +0.18. Of 1,499 interactions, 1,330 were jointly classifiable in a blinded artefact audit and matched student-recorded handling states; 169 lacked joint confirmation. Integrated case records distinguished written performance from workload, access and reported action. The study specifies site-supported design conditions for subsequent controlled and practitioner-led evaluation.
No takes yet. Share an insight, caveat, or question.
XIONG et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: