AI reveals which sustainability issues companies actually emphasize

Every loaf of bread carries a hidden ledger of environmental and social costs: a carbon footprint from farming and transport, a supply chain stretching across continents, and a workforce whose welfare depends on corporate policy. As climate change, food safety, and labor rights draw intensifying public scrutiny, the disclosures companies publish about environmental, social, and governance (ESG) issues have become central to how investors, regulators, and consumers judge corporate behavior. Yet a persistent question has haunted sustainability research: do companies actually talk about the issues that matter most financially to their industry, or do they simply produce volumes of text that sound responsible without addressing material risks?

A research team at Kyushu University in Fukuoka, Japan, has now built an artificial intelligence method designed to answer precisely that question. Rather than counting how often sustainability buzzwords appear in corporate filings, the new approach measures how closely what companies disclose matches the ESG issues that are financially material to their specific industry. Applied to Japan’s food and beverage sector, the method revealed a striking pattern: firms place far greater emphasis on social-related topics, particularly product quality and customer welfare, while environmental disclosure remains comparatively limited. The findings were published in the journal Corporate Social Responsibility and Environmental Management on September 16, 2026.

The study’s first author, Siyu Shen, a graduate student at Kyushu University’s Graduate School of Economics, explained the limitation of conventional text-analysis approaches in corporate sustainability research. High-quality ESG disclosure helps stakeholders assess business performance and risks, she noted, yet simply counting mentions of an issue, or judging whether the surrounding language sounds positive or negative, reveals little about how seriously a company actually treats that issue. A firm could mention climate change dozens of times in vague, aspirational terms while providing almost no concrete data, and traditional keyword-based methods would score it highly. The Kyushu team set out to close that gap.

Determining which issues genuinely matter most to a given industry requires an authoritative, industry-specific benchmark. The researchers turned to the Sustainability Accounting Standards Board, now maintained under the IFRS Foundation, which identifies the sustainability issues most likely to affect financial performance within each industry. Food and beverage companies offered a compelling test case: their reliance on agricultural raw materials and exposure to global supply chains make ESG risks directly relevant to costs, operational stability, and long-term profitability. Water scarcity, greenhouse gas emissions, labor conditions among suppliers, and food safety are not abstract concerns for such firms but tangible financial variables.

To process the enormous volume of corporate text, the team analyzed 1,623 annual securities reports filed by 340 publicly listed Japanese food and beverage companies between 2020 and 2024. The analytical engine was a multilingual AI model called Text Match Pre-Trained Transformer, or TMPT, designed for semantic text matching. In a first step, TMPT locates passages in each report that relate in meaning to specific ESG issues. Crucially, the model does not depend on exact keyword matches; it captures underlying concepts even when they are expressed in different words, so a discussion of carbon dioxide emissions and a reference to greenhouse gas reductions are recognized as addressing the same material issue.

The model then rates the semantic relevance of each passage. A statement reporting the exact tonnage of emissions a company cut, for example, scores substantially higher than a general pledge to work toward carbon neutrality. This scoring architecture allows researchers to systematically compare, company by company, year by year, and issue by issue, how closely corporate disclosures address the matters that SASB identifies as financially material. The result is a quantitative map of disclosure quality rather than a mere tally of sustainability vocabulary, offering a far more discriminating instrument for evaluating corporate communication.

Professor Hidemichi Fujii of Kyushu University’s Faculty of Economics, who supervises Shen, highlighted a further methodological advantage of the approach. TMPT operates through zero-shot analysis, he explained, meaning it does not require researchers to manually write prompts or provide subjective guidance. That characteristic makes the analysis more standardized and reproducible than methods that depend on human-crafted rules or labeled training examples, a significant consideration when findings are meant to inform regulatory design and investment decisions across different markets and languages.

The analysis produced several findings with immediate relevance for investors and policymakers. Japanese food and beverage companies consistently emphasized social issues, especially product quality and customer welfare, while giving comparatively less attention to environmental matters such as water and energy management, and to governance concerns such as supply chain oversight. More revealing still, disclosure priorities shifted noticeably after 2022. Greenhouse gas emissions showed the sharpest increase in disclosure emphasis, followed by human capital and supply chain management. The rise in climate-related disclosure was significantly greater among companies listed on Japan’s Prime Market, particularly during 2023 and 2024, a timing that coincided with new climate reporting requirements aligned with the Task Force on Climate-related Financial Disclosures, the international framework for reporting climate-related financial risks and opportunities, which became mandatory for Prime Market-listed firms.

That correlation between regulatory change and disclosure behavior carries a powerful implication: clear, industry-specific requirements from regulators can measurably influence which ESG issues companies choose to prioritize. Shen suggested that the findings demonstrate how disclosure regimes shape corporate communication, and expressed hope that the study will help firms identify their material sustainability topics and allocate limited disclosure resources more effectively. For standard setters designing sustainability reporting rules, the evidence suggests that specificity matters; vague expectations produce vague reporting, while concrete, financially anchored requirements appear to drive substantive shifts in what companies disclose.

The Kyushu team views the framework as extensible well beyond Japan’s food and beverage sector, with potential application to other industries and countries where ESG reporting regimes are evolving rapidly. The researchers plan to refine the method’s accuracy by developing a human-coded benchmark, incorporating corporate sustainability reports alongside annual securities filings, and assessing not just the presence but the depth and credibility of corporate disclosures. As AI-driven text analysis matures, tools of this kind could become standard instruments for separating meaningful sustainability accountability from polished but shallow corporate language, giving investors and regulators a reproducible way to ask not just what companies say, but whether they say anything that matters.

Subject of Research: AI-based measurement of financially material ESG disclosure in Japanese food and beverage companies

Article Title: AI helps reveal what sustainability issues companies emphasize

Article References: AI helps reveal what sustainability issues companies emphasize. Original publication (Bibliographic details unavailable.)

Image Credits: AI Generated

DOI: Not provided

Keywords: ESG disclosure, artificial intelligence, natural language processing, corporate sustainability, SASB standards, Japanese companies, food and beverage industry, greenhouse gas emissions, TCFD framework, sustainability reporting, text mining, Kyushu University