Improving Software Development Traceability with Structured Prompting
Dae‐Kyoo Kim · Journal of Computer Information Systems · 2025
Generative large language models (LLMs) are increasingly integrated into software development to enhance efficiency across various phases, from requirements analysis to testing. However, the design of effective prompts for leveraging these AI capabilities in software development remains underexplored. This study aims to bridge this gap by developing prompt designs that enhance the traceability of software artifacts created by AI models. Employing design principles adapted from social sciences, the study introduces structured prompting that significantly improves traceability measures compared to unstructured prompting. Structured prompts were notably effective in the Design Modeling and Testing phases, showing improvements of 32.21 and 26.35 respectively with an overall average improvement of 23.92. Despite these advances, challenges persist, such as ensuring AI models adhere strictly to structured prompts, which can sometimes lead to outputs being overly specific or not fully compliant with the prompt instructions.