Toward responsible adoption of LLMs in the energy sector: A mixed-methods study using fuzzy Delphi and cross-cultural SEM

Despite the growing integration of large language models (LLMs) across various industries, the current literature remains limited in key areas, including a predominant focus on adoption intention rather than actual use and post-adoption outcomes, as well as a lack of domain-specific studies, particularly within the engineering and energy sectors. This study addresses these gaps through a two-phase investigation conducted in Malaysia and the United States. The study followed a sequential multistage methodology. First, a literature review and the fuzzy Delphi method (FDM) were employed to identify and validate key LLM adoption factors through expert consensus in the energy sector. These validated factors contributed to the development of an extended UTAUT-based model incorporating Trust Concerned (TC), Ethical Principles (EP), Perceived Anthropomorphism (PANTH), and Perceived Animacy (PAN). The model was then tested using survey data from 509 engineering professionals and analyzed using partial least squares (PLS) structural equation modeling (SEM) and multigroup analysis (MGA). Results revealed that Performance Expectancy (PE), EP, and Facilitating Conditions (FC) significantly influenced usage behavior (USE) in both contexts, while TC demonstrated a negative effect. PAN and PANTH significantly affected Effort Expectancy (EE) in Malaysia; however, PANTH remained insignificant in the US, a difference confirmed by PLS–MGA. Importantly, USE significantly influenced post-adoption outcomes, namely User Satisfaction (USAT) and Quality of Decision-Making (QDM). This study advances theory by integrating socio-ethical and design-based predictors into adoption models and highlights the need for domain-specific, post-adoption, and cross-cultural analyses.