A framework for BIM–AI interaction in sustainable construction site management

Sustainability-oriented construction site management requires the retrieval, comparison and verification of information distributed across building information models, technical documents and certification requirements. In current practice, these activities are often performed manually through model schedules, document inspection and spreadsheet checks, resulting in time-consuming and error-prone workflows. This paper proposes and preliminarily evaluates a building information modelling and artificial intelligence interaction framework for supporting sustainability-related construction site tasks through natural language query automation. The framework connects model-based product and quantity information, construction documents, a semantic knowledge layer and large language model-based agents to enable access to specific project information. The framework was evaluated on representative sample portions of the BIM model and project documentation through three sustainability-related tasks: construction and demolition waste management, building reuse assessment and environmental product declaration identification. The outputs were validated against manual document checks and reference values, using a weighted correctness score. The results show an overall reliability of 75.0% across the evaluated queries, with higher performance in building reuse assessment and environmental product declaration checking than in waste-related model queries. The main limitations concerned phase- aware model filtering and element-level traceability. The contribution of the paper is a task-based framework and evaluation protocol that shows how product and process information in the built environment can be connected with artificial intelligence-supported querying to reduce manual information retrieval and support sustainability decision-making.