Greenwashing the future? Computational text analysis of environmental reporting from the fossil fuel industry

Achieving net zero greenhouse gas emissions by mid-century is central to climate policy agendas worldwide. As pressure mounts to show progress towards the energy transition, an increasing number of companies are committing to climate targets and risk engaging in ‘futurewashing’, a new form of misleading communication practice. Pairing the Net Zero Tracker dataset with a novel text corpus, this research uses computational methods to analyze forward-looking discourse in the sustainability reports of 97 fossil fuel companies on the Forbes Global 2000 list. After assessing climate target characteristics, a conventional keyword-based and more sophisticated large language model (LLM) approach are compared to capture future focus. This demonstrates that implementing a dynamic few-shot prompt with Meta's Llama 3.1 405B model outperforms a custom-made dictionary classifier. Using the LLM, the prevalence of forward-looking statements is identified to measure future focus. This analysis finds that while future focus tends to be higher for companies with stronger climate targets, the prevalence of forward-looking statements varies greatly across the fossil fuel industry and suggests inconsistent messaging at the company level. Amid the proliferation of climate targets and the development of mandatory requirements for sustainability reporting, quantitative text analysis of corporate climate communication can contribute to an emerging understanding of ‘futurewashing’ and provide empirical insights to inform policy practitioners.Key policy insightsForward-looking statements are prevalent in corporate sustainability reporting and may serve as a vehicle for so-called ‘futurewashing’.Reporting guidelines should clearly define and delineate acceptable use of forward-looking statements to mitigate the risk of greenwash.Policymakers could reduce verification challenges for environmental disclosures by promoting structured, machine-readable reporting of narrative text and numeric data.Recent developments in machine learning provide opportunities for researchers and regulatory bodies to monitor potentially misleading communication practices.More research is needed on the factors associated with cases of inconsistent messaging by corporations to inform preventative policy measures.