Operationalizing sustainability through NLP and large language models: a systematic review of AI-driven requirements engineering

In recent years, sustainability has become a major concern for software-intensive systems. However, sustainability is rarely expressed as quantifiable, traceable, and verifiable requirements. Additionally, rapid advances in natural language processing (NLP) and large language models (LLMs) have revolutionised the automated analysis and generation of textual artifacts, particularly in the field of requirements engineering (RE). Although NLPs and LLMs have advanced significantly, the research community lacks a unified understanding of how AI-driven language technologies may be utilised to operationalise sustainability in RE. Therefore, the purpose of this study is to address this gap through a comprehensive systematic literature review that examines research at the intersection of NLP, LLMs, sustainability, and RE. Using the PRISMA 2020 guidelines, the study analysed 616 peer-reviewed studies published between 2015 and 2025, retrieved from Scopus and Web of Science. The study used systematic review procedures, bibliometric and scientometric analyses to examine trends in research, thematic structures, and methodological evolution. The study examined how NLP and LLM techniques support RE activities, the extent to which sustainability is explicitly modelled as a requirement, and the role of AI in translating sustainability goals into operational engineering artifacts. To combine all these findings, we use a Three-Lens framework. The scientometric analysis and systematic reviews confirm that sustainability is most visibly related in triad analysis. Requirements elicitation and requirements analysis, operationalised as a strong, artifact-driven requirements dimension, are closely integrated with various NLP techniques. On the other hand, the artifact-centric domains at the downstream, i.e., requirements specification and requirements classification, show gaps in their integration with the different sustainability dimensions. The study provides an AI-centric synthesis to guide future research and practice.