Towards trustworthy LLM-assisted EPD comparability assessment for circular construction materials

Environmental Product Declarations (EPDs) are increasingly used to support construction material selection, but their comparability depends on how system boundaries and allocation rules are interpreted. Although large language models (LLMs) are increasingly explored for sustainability assessment, limited attention has been given to how methodological ambiguity affects LLM reasoning reliability and comparability judgement. This study investigates how LLMs interpret A1–A3 carbon emissions in EPDs for virgin and recycled construction materials. Using virgin aggregates and recycled concrete aggregates as case materials, GPT-5.3, Qwen 3.6-Plus, and Gemini 3 were evaluated through individual EPD interpretation and pairwise comparability assessment tasks under zero-shot and rule-grounded prompting. The results show that A1–A3 emissions for recycled materials may not represent the same process boundaries as those for virgin materials because recycling-related burdens can be allocated to different EPD modules. Zero-shot LLMs often produced stable but conservative comparability judgements, frequently defaulting to non-comparability when methodological conditions were difficult to synthesise. Rule-grounded prompting improved judgement accuracy and reasoning quality for GPT-5.3 and Qwen 3.6-Plus, but reduced response stability and increased output variability, revealing an accuracy–stability trade-off. These findings show that LLM-assisted EPD interpretation requires evaluation beyond final-answer accuracy. Current LLMs should be used as supportive tools rather than standalone decision-making systems, and clearer, more harmonised methodological representation within EPDs is needed to improve AI-assisted interpretation.