Experimental Validation of GPT-5.5-Generated Low-Carbon Mortar Mixture Designs

Reducing Portland cement consumption is an effective strategy for lowering the environmental impact of cementitious materials; however, designing low-carbon mortar mixtures remains a complex engineering task. Inadequate proportioning of low-carbon cementitious mixtures may result in unsuitable workability or uncertain strength development, potentially causing difficulties in handling, placing, and compaction and affecting the consistency and quality of materials used in construction. This study experimentally investigated whether GPT-5.5 can independently generate technically feasible ordinary Portland cement (OPC)–ground granulated blast-furnace slag (GGBFS)-based low-carbon mortar mixtures that satisfy predefined engineering requirements under standard laboratory conditions. A structured engineering prompt specifying target compressive strength, mortar flow, available materials, and design constraints was submitted to GPT-5.5 in three independent conversations. The generated mixtures were manufactured without manual modification and evaluated through mortar flow, 7- and 28-day compressive strength, and mixture-based embodied-carbon assessments. Across the three independent conversations conducted in this study, GPT-5.5 proposed similar mixtures incorporating 60% GGBFS as a replacement for OPC. Although the model underestimated mortar flow and overestimated 7-day compressive strength, all generated mixtures were successfully manufactured without segregation or bleeding and achieved the target 28-day compressive strength of 40 MPa. Compared with the reference mortar, the AI-generated mixtures reduced Portland cement consumption by 60%, embodied carbon by approximately 56%, and embodied carbon intensity by approximately 50–53%. These findings demonstrate that GPT-5.5 can generate technically feasible low-carbon mortar mixtures for preliminary engineering design. However, experimental validation remains essential because the model should be regarded as a decision-support tool rather than a quantitative predictor of engineering performance.