Artificial Intelligence-Assisted Thematic Analysis: A Contribution to Architectural Research

Abstract

Architectural research involves comprehensive analysis of diverse data, including user satisfaction data, interview excerpts, plans, and photographs. Analyzing such qualitative material is costly and time-consuming. While architectural programming firms often use qualitative data analysis software, many practitioners rely on informal methods. This article explores AI-assisted content analysis by comparing it with human-centered analysis in two case studies. Efficiency, accuracy, and replicability were evaluated as key criteria. Case Study 1 utilized open-ended responses from a survey of 356 school staff about school premises, informing design guideline development for public school renovations in Quebec, Canada. Case Study 2 analyzed 33 interviews with teleworking mothers involving architectural plans, dwelling photographs, and participant profiles aimed at enhancing home office environments. The human-centered analysis included manual transcript splitting and coding, alongside reviewing photographs and plans. Our findings show that AI-assisted analysis excels in efficiently disassembling and reassembling data, while having some shortcomings in compiling and interpreting data. It also demands considerable expertise in prompt formulation, which hinders accuracy and reliability. Human-centered analysis, although inefficient, requires data familiarity, knowledge of evidence, and research experience. All three are absent from the models' training. In conclusion, the potential of artificial intelligence to streamline qualitative analysis is discussed.