Integrating artificial intelligence into Requirements Engineering (RE) offers substantial potential to improve efficiency and quality in system development. This paper presents a novel, structured method for systematically integrating and assessing the capabilities of Large Language Models (LLMs) within the RE process. The method comprises five sequential phases: input definition, prompt control selection, model selection, application, and evaluation. The method enables comparative and transparent analysis of multiple LLMs and prompting strategies across various RE task types. The method was validated through a proof-of-concept study focused on developing a football application encompassing 35 distinct requirements. Evaluation involved both quantitative ground-truth metrics (precision, recall, F1-score) and qualitative expert assessments. The results demonstrate that the method enables reproducible, model-independent evaluations of LLM performance and supports the structured integration of LLM-driven automation into RE processes. Overall, this study provides a scalable and scientifically grounded framework for researchers and practitioners aiming to leverage LLM capabilities effectively within systems and software engineering.
Stein et al. (Sun,) studied this question.