This is a Preprint and has not been peer reviewed. This is version 1 of this Preprint.
What Companies Say vs. What Matters: LLM Analysis of Biodiversity Disclosures in Oil and Gas
Downloads
Authors
Abstract
The power system ecosystem encompasses infrastructure intensive industries such as electric utilities, hydropower operators, oil and gas producers, and mining companies supplying critical minerals. These industries share a common challenge: their physical assets interact extensively with natural ecosystems, creating dependencies and impacts that increasingly draw investor and stakeholder scrutiny. Many companies voluntarily disclose nature related commitments through platforms like the Carbon Disclosure Project (CDP), often in response to investor and stakeholder pressure. Yet little is known about whether these disclosures reflect substantive, measurable targets aligned with companies' most material impacts. As nature related financial risks gain attention from investors and lenders, understanding the quality of these commitments becomes critical for capital allocation and policy design.
This paper develops a large language model (LLM) methodology to analyze corporate biodiversity disclosures across infrastructure intensive industries within the power system ecosystem. We demonstrate the approach using oil and gas producers in the United States as a test case, classifying CDP questionnaire responses as goals, commitments, or SMART targets (specific, measurable, achievable, relevant, and time-bound) and assessing their alignment with Global Biodiversity Framework categories and Exploring Natural Capital Opportunities, Risks and Exposure (ENCORE) identified material impacts and dependencies. Our analysis reveals that what companies say is often disconnected from what matters: disclosures are specific but lack measurability and time bound elements, with firms focusing on governance and data related categories rather than their most material ecosystem impacts. Yet peer patterns offer guidance: co-occurrence in disclosure categories reveals common practices that provide roadmaps for companies beginning their target setting journey. These findings point to two paths forward: companies can fill gaps by learning from peers which high materiality areas to prioritize, and can find the right words by using LLMs to articulate their existing actions in alignment with biodiversity frameworks rather than necessarily doing more.
DOI
https://doi.org/10.31223/X58J3J
Subjects
Artificial Intelligence and Robotics, Biodiversity, Oil, Gas, and Energy, Sustainability
Keywords
corporate disclosure, energy sector, biodiversity risk assessment, Large Language Models, natural capital, sustainability reporting, Biodiversity risk assessment, large language models, Energy sector, sustainability reporting
Dates
Published: 2026-02-01 06:02
Last Updated: 2026-02-01 06:02
License
CC-BY Attribution-NonCommercial 4.0 International
Additional Metadata
Data Availability (Reason not available):
The data supporting the findings of this study are being curated and will be made available after publication.
Metrics
Views: 52
Downloads: 3
There are no comments or no comments have been made public for this article.