A new way to analyze ESG reports: A theme based model supported by AI

Abstract

Sustainability reports are often long, complex, and poorly standardized, making it difficult to extract comparable and meaningful information across companies. Despite increasing regulatory pressure, including the EU's Corporate Sustainability Reporting Directive (CSRD) and European Sustainability Reporting Standards (ESRS), the quality, clarity, and usefulness of ESG disclosures remain questionable. Neither manual nor automated analyses have fully resolved these challenges. This study introduces a transparent and replicable methodological framework for organizing heterogeneous ESG reports into structured thematic summaries. Rather than applying predefined categories, the approach begins with the reports themselves: ESG disclosures are reviewed and similar passages grouped inductively into seven shared ESG themes. Theme-specific keywords and contextual filters are then derived to support AI-assisted retrieval across a larger dataset. The framework was applied to 393 sustainability reports from publicly listed EU companies. The findings demonstrate that the inductively derived thematic structure can be applied consistently across large datasets while preserving interpretability. The framework produces structured summaries showing which ESG topics are addressed, how extensively, and at what level of specificity, providing a foundation that users can begin to interpret and compare in a manner of their choosing. Developing and validating formal comparison methods lies beyond the scope of this study and represents a direction for future research.