Sustainability disclosures are the backbone of corporate sustainability efforts. If stakeholders want companies to compete over how “good” they are for society, they need a way to see what companies are doing and trying to do. But sustainability disclosures have been widely criticized for their lack of credibility and comparability. To improve them, nonprofits and companies have developed voluntary reporting standards that many firms have now adopted and that have informed mandatory disclosure regulations. These critiques and regulatory debates have proceeded, however, without basic facts about what these disclosures contain and whether standards help. Answering these questions has been difficult because of the non-standardized nature of these disclosures.

This project overcomes these barriers through the use of large language models and provides the most comprehensive analysis of sustainability disclosures to date. We evaluate over 15,000 sustainability disclosure documents issued by more than 2,100 Russell 3000 firms from 1998-2023. We construct measurable indicators based on characteristics that stakeholders have identified as relevant: specificity, quantitative evidence, fluff (~puffery), negative news, and data table and figure counts.

We find five key patterns. First, we document rapid growth in both sustainability reporting and adoption of external assurance and standards. Second, as reporting became mainstream, the reports became less informative: they became less specific and quantitative and included more fluff. Third, longer reporting tenure predicts improvement along our measures, but these differences are largely driven by changes in the composition of firms issuing reports rather than learning-by-doing. Fourth, companies rarely abandon standards and often adopt multiple standards simultaneously. Finally, most governance mechanisms show mixed associations with our measures: they correlate with less fluff, more negative news, and more data tables, but also with less specific and less quantitative evidence. We find no consistent evidence that adoption of voluntary reporting standards leads to improvements along the dimensions we study.

Link.