Student perceptions and acceptance of a custom designed generative AI tool for Construction Hazard Prevention through Design education

Engineering education increasingly explores Generative AI to support complex, safety-critical learning, yet empirical evidence on student acceptance of domain-specific tools remains limited. This gap is acute in construction safety education, where accurate interpretation of regulatory standards underpins high-stakes hazard decisions. This study investigates undergraduate construction management students’ perceptions of desAIner, a custom GenAI tool combining a Retrieval-Augmented Generation (RAG) architecture anchored to OSHA 29 CFR 1926 with a Human-in-the-Loop (HITL) design to scaffold Construction Hazard Prevention through Design (CHPtD). Sixty-six upper-division students used desAIner during a structured classroom module and completed a 24-item survey operationalizing an extended Technology Acceptance Model (TAM), analyzed using Partial Least Squares Structural Equation Modeling (PLS-SEM). RAG-driven information quality significantly predicted perceived usefulness, establishing regulation-grounded outputs as a precondition for AI utility in high-stakes domains. Learning enhancement emerged as the dominant predictor of overall satisfaction, with usability playing a secondary role, confirming that students prioritize cognitive scaffolding over interface convenience. The model also suggests that perceived usefulness retains a significant direct effect on intention for future professional use, while overall satisfaction exerts a substantially stronger, complementary influence. Students’ intent to adopt domain-specific AI is thus driven primarily by transparent, reliable systems delivering holistically satisfying pedagogical experience.