As generative artificial intelligence (AI) tools such as generative pretrained transformer–based language models gain prominence, the construction industry faces adoption challenges that differ from deterministic tools. Generative AI (GenAI) is probabilistic, multifunctional, and cognitively demanding, requiring prompting skills, interpretability, and data governance—factors not fully captured by traditional technology acceptance model (TAM) and unified theory of acceptance and use of technology (UTAUT) applications. Addressing this gap, this study assesses how US construction professionals perceive both the benefits and complexities of GenAI, identifies which applications they prioritize, and segments adoption profiles across the workforce. A nationwide survey of 97 companies revealed three benefit dimensions—Real-Time Construction Intelligence, Regulatory and Analytical Automation, and Design Acceleration and Enhancement—and three complexity dimensions—Organizational and Change Management, Data and Customization, and Integration and Cultural Alignment—through exploratory factor analysis. Weighted ranking placed Design Time Reduction, Building Information Model (BIM) Enhancement, Risk Assessment Support, Real-Time Cost Control, and Site-Safety Monitoring as top benefits, while Risk Assessment Challenges, Organizational Data Availability, and Real-Time Deployment emerged as leading barriers. Benefits were rated 11.53 points higher than complexities t(96)=12.98, p<0.001; d=1.318 and were positively correlated (r=0.613, p<0.001). Cluster analysis revealed four perception profiles, with the most critical being training moderated polarization and highly familiar users. Theoretically, the findings refine TAM/UTAUT by showing that usefulness manifests as decision support, analytical automation, and design acceleration, while ease of use depends on interpretability, organizational readiness, and role alignment. Practically, firms should begin with a governed retrieval layer that enables GenAI to cite project sources, complement it with role-based prompt-and-verify training, and pilot bounded tasks under human review before scaling.
Sadeh et al. (Fri,) studied this question.