Large language models for sustainability reporting: A systematic review and research agenda

Sustainability reporting refers to the activities by which large companies describe their efforts to maximize the benefits on universal criteria about economic, social, and environmental impacts. This promotes transparency among stakeholders and is considered essential in supporting governance and compliance with international regulations by improving today's business landscape. Among other problems, inaccurate company disclosures, known as Greenwashing, can affect the quality of reports and their unstructured nature poses serious limitations to economic analysts during company evaluation. Efficiently gathering and aligning available reports and other complementary data for a unified framework is an important perspective on modern sustainability analysis. Recent advances in Natural Language Processing (NLP), particularly the rise of transformer-based Large Language Models (LLMs), have enabled new capabilities in semantic information extraction, automated classification, and the detection of misleading claims within unstructured corporate sustainability disclosures. This survey provides a systematic review of NLP and LLM-based approaches in sustainability reporting, with particular attention to methodological trends and unresolved challenges. Adopting a structured review methodology that integrates scientometric mapping and meta-synthesis, we offer an evidence-based taxonomy and a synthesis of current practices, highlighting key gaps and future research directions. The findings highlight past explorations and future directions, demonstrating that LLMs offer remarkable accuracy and innovative solutions for challenges related to traditional sustainability reporting current practices.