Explainable large language models for software requirements engineering: a systematic literature review

Abstract

Large language models (LLMs) have become increasingly popular in automating language-dependent software engineering activities especially in requirements engineering (RE) because they have an inherent nature of being language-driven. The increasing popularity of LLMs in software engineering applications has led to lack of clarity and understanding regarding the use of explainability within LLM-based RE solutions. There is no existing literature summarizing their techniques, assessment methodologies, issues, or future research directions. This systematic review aims to explore the state-of-the-art techniques, methodologies, and challenges related to using explainability in LLMs for RE applications. The study followed PRISMA guidelines to identify works published between 2018 and 2026 from seven databases that include Springer, ScienceDirect, IEEE Xplore, ACM Digital Library, Wiley, arXiv, and Google Scholar. This survey considered and synthesized 57 primary sources from the total 487 records. The most investigated areas where LLMs are applicable in RE processes include elicitation, ambiguity detection, and traceability. The approaches used for explaining the models are mainly based on prompts and post-hoc methods that do not use any formal approach. In terms of evaluation, task performance-related measures exist but neither of the approaches evaluates the quality of explanations provided by LLMs nor are there any benchmark sets. The challenges related to RE using LLMs include hallucinations, lack of explainability, absence of human-centered design approaches, and unsolved ethical issues. In conclusion, explainability appears to be treated as an afterthought and not as a core element for designing LLM-based RE.

Received 11 June 2026

Accepted 28 August 2026

Published 17 September 2026