A Detailed Study on LLM Biases Concerning Corporate Social Responsibility and Green Supply Chains

Greta Ontrup, Annika Bush, Markus Pauly, Meltem Aksoy · 2025

Read A Detailed Study on LLM Biases Concerning Corporate Social Responsibility and Green Supply Chains by Greta Ontrup, Annika Bush, Markus Pauly, Meltem Aksoy (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-01840.txt.

Introduction

Greta Ontrup, Annika Bush, Markus Pauly, and Meltem Aksoy audit how language models think about corporate responsibility. Companies increasingly use these models in supply chain work, yet models are known to carry biases about how much sustainable and ethical practice matters. The study probes those leanings with validated survey instruments rather than casual prompts.

Several state-of-the-art models answered standardized questionnaires on business ethics and responsibility and on the importance of sustainable supplier and customer relations. Systematic differences between models were significant, and framing the questions with four organizational culture types substantially shifted the answers. The findings matter for anyone letting a model assist decisions in sustainability contexts: the model's corporate values are neither neutral nor stable across vendors and prompts.

Catalog record

CREATORGreta Ontrup, Annika Bush, Markus Pauly, Meltem Aksoy
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE58 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-01840

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

A Detailed Study on LLM Biases Concerning Corporate Social Responsibility and Green Supply Chains Greta Ontrup, Annika Bush, Markus Pauly, Meltem Aksoy Abstract Organizations increasingly use Large Language Models (LLMs) to improve supply chain processes and reduce environmental impacts. However, LLMs have been shown to reproduce biases regarding the prioritization of sustainable business strategies. Thus, it is important to identify underlying training data biases that LLMs pertain regarding the importance and role of sustainable business and supply chain practices. This study investigates how different LLMs respond to validated surveys about the role of ethics and responsibility for businesses, and the importance of sustainable practices and relations with suppliers and customers. Using standardized questionnaires, we systematically analyze responses generated by state-of-the-art LLMs to identify variations. We further evaluate whether differences are augmented by four organizational culture types, thereby evaluating the practical relevance of identified biases. The findings reveal significant systematic differences between models and demonstrate that organizational culture prompts substantially modify LLM responses. The study holds important implications for LLM-assisted decision-making in sustainability contexts. nmark 1 Introduction Large language models (LLMs) have become increasingly influential in organizational decision-making. A specific application field is green supply chain management. Here, LLMs are adopted to enhance the sustainability of operations, assist in the preparation of sustainability reports, or support supplier evaluations ( Zhou et al. 2025 ) . Other examples cover strategic planning, stakeholder communication, informing policy development, and reporting. The widespread adoption of LLMs within organizational settings prompts critical questions regarding how these models interpret and convey principles of sustainability. LLMs are developed using extensive text-based datasets, which inherently mirror the societal values, cultural conventions, and biases contained within their training materials ( Gallegos et al. 2024 ; Aksoy 2025 ) . In addition to training data, other factors such as the underlying model architecture ( Weber et al. 2024 ) , fine-tuning procedures ( Ferrara 2023 ) , instruction-tuning strategies ( Resnik 2025 ) , and the specific prompts used to elicit responses ( Kamruzzaman and Kim 2024 ; Brucks and Toubia 2025 ) also shape model behavior. When deployed for green supply chain management, these embedded aspects can influence how organizations approach environmental management and social responsibility. A recent study found that LLMs exhibit biases related to sustainable development ( Bush et al. 2025 ) . For instance, GPT-4o ( OpenAI 2024 ) responses mirrored skepticism regarding the compatibility between sustainability and AI, while the output by LLaMA reflected techno-optimism ( Bush et al. 2025 ) . This demonstrates that an organization’s choice of an LLM for decision-making could significantly impact its sustainable strategies, as biases inherent in the training data may translate into analysis and output. Consequently, examining how LLMs interpret and respond to sustainability-related concepts is essential, especially as organizations increasingly depend on these technologies to address multifaceted environmental and social issues. In this study, we perform an empirical multi-model evaluation to assess the biases of various LLMs regarding the perceived importance of corporate social sustainability practices and stakeholder relationships within supply chains. To determine the practical implications of these biases, we further prompt the LLMs to assume the perspective of an employee representing one of four prominent organizational culture types. Prior research indicates that organizational culture significantly influences the implementation of sustainable strategies, as distinct cultures are associated with varying capacities, motivations, and processes in corporate social responsibility activities ( Maheshwari et al. 2024 ) . Our analysis examines whether different cultural value orientations amplify or attenuate LLM-generated perspectives on ethics, corporate responsibility, and environmental collaboration with suppliers and customers. The goal is to answer the following research questions: RQ1: To what extent do LLMs display biases when evaluating the importance of corporate sustainability practices and stakeholder relations in green supply chains? RQ2: How does prompting LLMs to adopt specific organizational cultures influence their assessments of the importance of corporate sustainability practices and stakeholder relations in green supply chains? By answering these research questions, we contribute to an interdisciplinary crossroad of research directions, namely, green supply chain management, corporate sustainability, and LLM evaluation/ psychometrics. Specifically, this study contributes to theory building and practical recommendations in these ways: First, by uncovering underlying response patterns in different LLMs – especially in the context of different organizational cultures – we can make predictions about the extent to which the use of LLMs may have potential unintended or undesirable consequences for different organizations. Second, the results add to the research domain of LLM psychometrics, an important methodological approach for evaluating technological outcomes. This work provides significant implications for decision-makers in organizations that increasingly rely on LLMs to support sustainability-related activities, including supplier evaluation, sustainability reporting, and strategic planning. 2 Theoretical Background 2.1 Corporate Social Responsibility and Green Supply Chain In recent years, organizations have increasingly prioritized sustainable business practices within their global strategies. This is often captured under the umbrella term corporate social responsibility (CSR; Ahi and Searcy 2013 ), referring to organizational initiatives that extend beyond economic and legal requirements to promote social good ( McWilliams and Siegel 2001 ) . CSR encompasses a range of activities, including the enhancement of sustainability and environmental performance ( McWilliams et al. 2005 ) . This trend reflects a broader societal paradigm shift toward sustainable development ( World Commission on Environment and Development 1987 ) . The United Nations’ 2030 Agenda for Sustainable Development operationalizes sustainability through 17 Sustainable Development Goals (SDGs), which have become critical objectives in organizational strategy and drive stakeholders to systematically incorporate sustainability considerations into decision-making ( Emilia et al. 2024 ) . Central to this work is the United Nations Sustainable Development Goal (SDG) ("Responsible consumption and production"), which aims to promote the sustainable and efficient use of natural resources, to encourage companies to minimize their social and environmental risks, and to better inform consumers about sustainable consumption ( Federal Ministry for Economic Cooperation and Development (2025) BMZ ) . Central to this work is the United Nations Sustainable Development Goal (SDG) 12 ("Responsible consumption and production"), with one of its most important levers: supply chain management (SCM). Supply chain management is the coordination and integration of the flow of materials, information, and financial resources across the entire supply chain (from suppliers to customers) to optimize overall system performance ( Stroumpoulis and Kopanaki 2022 ) . In the context of sustainability, green supply chain management (GSCM) is discussed as "one of the essential elements for sustainable development and survival" of organizations ( Lee and Joo 2020 , p. 2). Defining GSCM presents challenges, as SCM encompasses a wide range of activities and there is considerable variation regarding the components that should be included in both the definition and implementation of GSCM ( Fahimnia et al. 2015 ; Srivastava 2007 ) . Broadly conceptualized, GSCM refers to initiatives that aim to incorporate environmental aspects into SCM ( Lee and Joo 2020 ) . This encompasses a wide range of tasks including sourcing (e.g., renewable energies), product design (e.g., use of sustainable material), delivery (e.g., emissions), customer and product use (e.g., customer education), recycling (e.g., return of products) or life cycle assessment ( Hassini et al. 2012 ; Srivastava 2007 ; Becker et al. 2025 ; Tokkozhina et al. 2025 )

Read the complete text