ESG Greenwashing Detection: Using AI to Identify the Truthfulness of Enterprises' Environmental Commitments

This article introduces an ESG greenwashing detection project that combines large language models (LLMs) and machine learning technologies, aiming to automatically identify false environmental claims by analyzing enterprises' ESG and CSR reports. This project addresses the high cost and low efficiency of traditional manual audits, providing decision support for investors, regulators, etc., and promoting enterprises to truly fulfill their environmental responsibilities.

Against the backdrop of global climate change, enterprises' ESG performance has received attention, but greenwashing (exaggerated/false environmental claims) has triggered a trust crisis. Its harms include: investors making wrong decisions due to false information; consumers being misled, weakening environmental actions; at the social level, delaying climate response and harming the competitiveness of truly environmentally friendly enterprises. Traditional manual audits are difficult to handle massive reports, so there is an urgent need for automated detection.

The project adopts a hybrid scheme of LLM + traditional machine learning:

• Role of LLM: Identify greenwashing features such as exaggerated and vague expressions in text through prompt engineering;
• Supplementation by traditional ML: Analyze quantitative features like vocabulary frequency and sentence structure to build statistical models;
• Data sources: Process two types of documents—ESG reports (comprehensive assessment of environment, social, and governance) and CSR reports (corporate social responsibility activities).

Identifying greenwashing requires multi-dimensional analysis:

1. Consistency between commitments and actions: Compare commitments and results over different periods to identify 'saying without doing';
2. Language precision: Detect vague words (e.g., 'committed to') and relative descriptions without benchmarks;
3. Sufficiency of data support: Check for third-party verified data, specific values, and industry comparisons;
4. Industry background comparability: Evaluate in combination with industry characteristics to identify statements that deviate from the norm.

This technology has a wide range of applications:

• Investment decision-making: Help ESG investors screen truly sustainable enterprises and avoid green traps;
• Regulatory compliance: Assist regulatory agencies in efficiently reviewing reports and improve regulatory efficiency;
• Media supervision: Help media/NGOs expose greenwashing behaviors and safeguard the public's right to know;
• Enterprise self-audit: Help enterprises ensure the accuracy of reports and avoid doubts caused by inappropriate expressions.

Technical challenges: Scarcity of training data, evolution of greenwashing forms, cross-language and cultural differences, control of misjudgment risks; Ethical considerations: Algorithms need to be transparent, results need manual review, data privacy needs protection, and over-interpretation of detection results should be avoided.

Future directions: Multimodal analysis (integrating text/images), real-time monitoring (social media dynamics), cross-document verification, causal inference; Conclusion: This project provides a new tool for greenwashing detection, promotes enterprises to take real environmental actions, and serves global sustainable development goals.