The Role of Artificial Intelligence in Enhancing ESG Outcomes: Insights from Saudi Arabia

Abstract

This study investigates the relationship between artificial intelligence (AI) adoption and environmental, social, and governance (ESG) performance among 100 listed Saudi Arabian firms over the period 2015–2024. Drawing on panel data regression techniques, including fixed effects models with Driscoll–Kraay standard errors, pooled OLS with industry and year controls, and dynamic panel estimations using system GMM, the analysis reveals a significant and positive association between AI implementation and overall ESG scores. Disaggregated analysis shows that AI adoption is particularly associated with improvements in the environmental and social dimensions, with a more moderate relationship to governance practices. To address potential issues of cross-sectional dependence and heterogeneity, the study applies the Common Correlated Effects Mean Group (CCEMG) and Mean Group (MG) estimators as robustness checks, which confirm the consistency of the main findings. In addition, the Dumitrescu–Hurlin panel Granger causality test indicates that AI adoption Granger-causes ESG performance—especially in the environmental and social dimensions—while no reverse causality is observed. The results suggest that AI technologies are positively linked to firms’ sustainability strategies and performance, supporting the integration of digital transformation into national and corporate ESG agendas, particularly in emerging markets like Saudi Arabia.

1. Introduction

The accelerating integration of Artificial Intelligence (AI) into business operations has fundamentally reshaped how firms pursue sustainability, ethical governance, and social responsibility. As the global business environment becomes increasingly complex and data-driven, AI has emerged as a transformative force that enables organizations to respond more efficiently and proactively to Environmental, Social, and Governance (ESG) challenges. By leveraging machine learning, predictive analytics, and automation, companies can optimize energy consumption, monitor supply chains for ethical compliance, and enhance governance through real-time risk assessments and data transparency.

In today’s business landscape, Environmental, Social, and Governance (ESG) considerations have moved from peripheral concerns to central determinants of corporate strategy and value creation. Investors, regulators, and consumers increasingly evaluate companies not only on financial performance but also on their environmental stewardship, social responsibility, and governance quality. Empirical evidence shows that firms with robust ESG practices tend to experience lower capital costs, enhanced operational efficiency, and stronger stakeholder trust, ultimately leading to sustainable long-term performance (Hamdouni, 2025a, 2025c). Importantly, the growing integration of ESG into corporate decision-making is closely linked to the evolving understanding of fiduciary duty, which traditionally focused on maximizing shareholder value but now increasingly encompasses responsibilities toward broader stakeholders and long-term sustainability. Recent debates—particularly under Delaware law, which governs a majority of U.S. corporations—illustrate the tension between ESG strategies and the doctrine of shareholder primacy. De Mariz et al. (2025) emphasize that fiduciary duties of care and loyalty can legitimately extend to ESG-aligned decisions, provided directors carefully document their reasoning, even amid rising anti-ESG sentiment. Supporting this evolution, a groundbreaking study by the European Central Bank demonstrates that banks engaging in green finance benefit from lower funding costs, providing clear evidence that ESG-aligned strategies can yield tangible financial advantages (European Central Bank et al., 2024). This shift is particularly significant in economies like Saudi Arabia, where the Vision 2030 agenda prioritizes sustainability, diversification, and corporate transparency as key levers for economic transformation. As a result, understanding how ESG factors influence firm value and investment decisions is essential for both policymakers and market participants seeking to align corporate growth with national development objectives.

In Saudi Arabia, the role of AI in advancing ESG performance has gained particular prominence, especially in light of the country’s long-term national development strategy—Saudi Vision 2030. This ambitious initiative aims to diversify the economy, reduce reliance on oil, and foster sustainable development through innovation and digital transformation. As part of this vision, AI is being strategically deployed to improve corporate ESG outcomes across various sectors. For example, AI technologies are helping energy-intensive industries monitor and reduce carbon emissions, assisting companies in promoting workforce nationalization and gender inclusion, and supporting regulatory compliance through advanced analytics and automated governance systems. These applications not only enhance operational efficiency but also help Saudi firms align with global sustainability standards, attract foreign investment, and build long-term stakeholder trust.

Previous studies have explored various factors associated with corporate ESG performance, including ownership structure, board characteristics, CEO compensation and incentives, M&A activities, and corporate governance frameworks (Hamdouni, 2025b; Kurniawan & Rokhim, 2023; Villalonga et al., 2025). Other strands of literature have examined the impact of digital transformation and emerging technologies on firm performance, including productivity, financial outcomes, innovation capacity, and organizational efficiency (Jardak & Ben Hamad, 2022; Smith, 2024). Despite these advancements, academic literature examining the relationship between AI on ESG performance within the Saudi Arabian context remains scarce. While existing global studies have recognized the potential of AI in promoting sustainability and responsible governance, they often focus on developed economies (such as North America, Western Europe, and parts of East Asia) (Bitzenis et al., 2025) or provide a generalized perspective that overlooks regional differences. This often results in findings that may not directly apply to other contexts, including the Middle East, Africa, Latin America, or Southeast Asia, where economic structure, regulatory environments, technological infrastructure, and social priorities can be very different.

However, there is a relative lack of research on the relationship between AI technology applications on firms’ ESG performance (Jing & Zhang, 2024), especially the lack of empirical research in the context of emerging markets (Zhou, 2025). Moreover, few studies have specifically examined how AI adoption is associated with corporate ESG performance in Saudi Arabia. While AI has been recognized for its potential to improve operational efficiency and decision-making, its role in enhancing ESG outcomes—through improved environmental monitoring, social impact tracking, and governance automation—remains underexplored in both theoretical and empirical research. As digital transformation becomes increasingly central to corporate strategy under Saudi Vision 2030, a critical question arises: is AI adoption significantly associated with ESG performance? If so, how does this influence vary across ESG dimensions, and what mechanisms drive the observed effects? In the case of Saudi Arabia and the broader Gulf region, unique institutional, cultural, and economic factors—such as government-driven digital policies, rapid urban development, and evolving social dynamics—demand a localized approach to understanding how AI relates to ESG performance.

From the perspective of the resource-based view, AI represents a strategic intangible asset that can enable firms to enhance ESG practices by improving data-driven decision-making, real-time risk management, and stakeholder engagement. From an institutional theory perspective, AI adoption may also reflect firms’ responses to normative and regulatory pressures, particularly as ESG disclosure standards tighten globally. Moreover, stakeholder theory suggests that firms adopting AI to meet stakeholder expectations—whether environmental compliance, social equity, or transparent governance—are likely to benefit from increased legitimacy and long-term value creation. These theoretical foundations suggest that AI can serve as an internal and external force for promoting sustainable and responsible business practices. At the same time, it is important to recognize that ESG scores, while widely used to assess corporate sustainability, are inherently subjective and reflect reporting practices, evaluation methodologies, and stakeholder perceptions rather than purely objective measures of environmental, social, or governance outcomes. In this study, ESG scores are used as a proxy for corporate alignment with Saudi Arabia’s Vision 2030 sustainability goals, meaning that observed correlations between AI adoption and ESG performance indicate alignment with reported ESG practices rather than definitive improvements in actual sustainability outcomes. AI solutions are often developed by firms aligned with ESG principles, which could introduce bias into reporting and potentially amplify ESG scores without corresponding real-world changes. The subjectivity of ESG evaluation is further illustrated by emerging technologies such as blockchain, where activities like Bitcoin mining may be viewed negatively for high energy consumption yet positively for promoting financial inclusion, demonstrating the complex, sometimes contradictory nature of ESG assessment. Finally, while this study is framed within institutional and stakeholder theories, it is acknowledged that corporations continue to operate under shareholder primacy models, balancing the creation of shareholder value with broader stakeholder interests, which may influence ESG engagement strategies. These considerations underscore the importance of interpreting ESG-based findings cautiously and highlight the need for complementary measures to capture objective sustainability performance.

This research gap presents a critical opportunity for empirical investigation. To date, there has been limited systematic analysis of how AI adoption is associated with ESG indicators at the firm level within Saudi Arabia. Questions remain regarding which aspects of ESG performance are most closely related to AI technologies, how firms in different sectors leverage AI for sustainability, and what barriers or enablers influence this process. Without such insights, policymakers and business leaders may face challenges in designing effective strategies to promote responsible AI adoption and sustainable development.

This study contributes to the literature on AI and ESG performance in several ways. First, it fills the existing gap by providing an up-to-date empirical assessment of how AI adoption is associated with ESG performance within the Saudi Arabian context. Second, it explores how AI adoption relates to individual ESG components—environmental, social, and governance—and whether these associations differ across industries. Third, the study applies advanced econometric techniques, including fixed effects models with Driscoll–Kraay standard errors, pooled OLS with year and industry controls, and dynamic system GMM estimation. Robustness checks using CCEMG and MG estimators account for potential cross-sectional dependence and firm-level heterogeneity. Additionally, the Dumitrescu–Hurlin panel Granger causality test confirms that AI adoption predicts improvements in ESG, environmental, and social scores, reinforcing the dynamic and leading role of AI in shaping sustainability outcomes. This study thus enriches existing research by offering a localized, data-driven perspective on how AI is linked to sustainable transformation in the Saudi corporate sector, providing actionable insights for corporate managers, policymakers, and investors aiming to align digital transformation with ESG priorities.

This study contributes to the literature on AI and ESG performance in several ways. First, it fills the existing gap in the literature by providing an up-to-date empirical assessment of how Artificial Intelligence (AI) adoption is associated with Environmental, Social, and Governance (ESG) performance within the Saudi Arabian context. Second, it explores how AI adoption relates to individual ESG components—environmental, social, and governance—and whether these associations differ across industries. Third, the study applies advanced econometric techniques, including fixed effects models with Driscoll–Kraay standard errors, pooled OLS with year and industry controls, and dynamic system GMM estimation. Robustness checks using CCEMG and MG estimators account for potential cross-sectional dependence and firm-level heterogeneity. This study thus enriches existing research by offering a localized, data-driven perspective on how AI is linked to sustainable transformation in the Saudi corporate sector. In doing so, it provides actionable insights for corporate managers, policymakers, and investors aiming to align digital transformation with ESG priorities.

The remainder of the paper is structured as follows: Section 2 reviews the existing literature; Section 3 presents the data sources and research methodology; Section 4 reports and interprets the empirical findings; Section 5 discusses the implications and outlines directions for future research; and Section 6 concludes the study.

2. Literature Review

2.1. Theoretical Perspectives

Artificial intelligence (AI) applications are reshaping how companies approach Environmental, Social, and Governance (ESG) initiatives through the lenses of Resource-Based View (RBV), Stakeholder Theory, and Institutional Theory. AI can provide a competitive advantage by improving ESG performance, attracting stakeholders, and conforming to institutional norms (Tian et al., 2025).

From the perspective of Resource-Based View, AI can be seen as a strategic resource that enhances a firm’s capabilities and creates a competitive advantage in terms of ESG performance (Huarng & Yu, 2024). Companies can utilize AI to optimize resource allocation, improve operational efficiency, and develop innovative sustainable products and services (Tian et al., 2025; Yu et al., 2025). AI-driven solutions can automate the monitoring of environmental impact, optimize energy consumption, and ensure compliance with environmental regulations, thereby improving environmental performance (Tian et al., 2025; Zhou, 2025). AI can also improve social performance by enhancing workplace safety, promoting diversity and inclusion, and ensuring ethical standards. AI strengthens corporate governance by enhancing transparency, detecting fraud, and improving risk management. The effective deployment of AI in these areas can lead to improved ESG ratings, which in turn can attract investors and enhance firm value.

Stakeholder Theory suggests that firms adopt ESG initiatives to meet the expectations and demands of various stakeholders, including investors, customers, employees, and regulators (Xie & Wu, 2025). AI can play a crucial role in managing stakeholder relationships and enhancing stakeholder engagement. AI-powered tools can analyze stakeholder sentiment, identify emerging issues, and facilitate communication and collaboration with stakeholders. AI can help companies respond to these concerns in a timely and effective manner, thereby building trust and strengthening relationships with stakeholders. AI can also facilitate transparent reporting and disclosure of ESG performance, which is increasingly important to stakeholders (Ho et al., 2024). At the same time, it is important to recognize that corporations operate under shareholder primacy models, where the creation of shareholder value remains a central objective. In this context, firms must balance shareholder interests with broader stakeholder concerns, and AI can support this balance by improving decision-making, risk management, and reporting processes that align both financial and ESG goals.

Institutional Theory posits that firms adopt ESG initiatives to conform to institutional norms and pressures, such as regulations, industry standards, and societal expectations (Chen et al., 2024). AI can help companies align their ESG practices with these institutional norms and demonstrate their commitment to sustainability. AI-driven solutions can automate the collection and analysis of data related to ESG performance, making it easier for companies to track their progress and benchmark themselves against their peers. AI can also help companies identify and mitigate ESG-related risks, such as regulatory fines, reputational damage, and supply chain disruption (Zhao et al., 2018). By adopting AI-powered ESG solutions, companies can signal their legitimacy and gain the approval of key stakeholders, such as regulators and investors.

2.2. AI and ESG Performance

Artificial intelligence (AI) is increasingly recognized as a significant driver in enhancing environmental, social, and governance (ESG) performance across various industries (Chen et al., 2024; Chen & Zhang, 2024; Correia & Água, 2024; Deberdt et al., 2025; Tian et al., 2025; Zhao et al., 2018). AI technologies offer tools and methods to improve data analysis, prediction, and operational efficiency, which are crucial for effective ESG management (J. Li & Jin, 2024; Sun & Saat, 2023; Xie & Wu, 2025; Yu et al., 2025). Through the utilization of AI, investors can systematically evaluate the environmental impact, social responsibility, and corporate governance practices of potential investments (Adeoye et al., 2024). Also, Generative AI can be employed to simulate various ESG scenarios, helping financial institutions to visualize potential future landscapes and adjust their strategies accordingly (Lim, 2024). By using AI-powered risk assessment tools that integrate ESG considerations, financial institutions can identify and mitigate potential risks associated with environmental and social factors, such as climate change, labor practices, and human rights violations (Apel et al., 2023).

AI applications are transforming how companies approach environmental sustainability. AI can optimize energy consumption, manage resources, predict environmental trends, and reduce carbon emissions (Antoniuk & Koliada, 2025; Burki et al., 2024; Manikandan et al., 2025; Yin et al., 2023). In the context of energy management, AI facilitates the development of smart grids that optimize energy distribution and reduce energy waste (Jaiswal & Pandey, 2024). Furthermore, AI-powered systems can monitor environmental conditions, detect pollution sources, and predict environmental risks, supporting proactive measures to mitigate ecological damage (Rahman et al., 2025). The energy sector uses artificial intelligence (AI) as a crucial instrument to achieve environmental sustainability targets by improving resource efficiency and decreasing emissions while minimizing waste production (Abonamah et al., 2025). Predictive analytics combined with smart grid management implemented through AI applications produced 15% less energy waste and reduced carbon emissions by 20% according to industry pilot project data. AI has proven its transformative capabilities by optimizing energy consumption while detecting inefficiencies to create both operational improvements and cost savings. The real-time monitoring capabilities of AI systems help companies meet strict environmental regulations and international climate goals by optimizing resource use and waste reduction, supporting circular economy practices for sustainable operations and enduring profitability (Abonamah et al., 2025). AI algorithms analyze large datasets to identify patterns and insights that lead to more efficient use of resources and reduced environmental impact (Akter, 2024; Ambani & Rathod, 2024).

In the social dimension, AI can improve workplace safety, promote diversity and inclusion, and ensure ethical labor practices (Giudici & Wu, 2025). AI systems can analyze employee data to identify biases in hiring and promotion processes, thus fostering a more equitable work environment (Tian et al., 2025). AI can contribute significantly to resource efficiency by digitizing recruitment processes and reducing environmental impacts like travel-related emissions (Rahman et al., 2025). AI can also enhance supply chain transparency, ensuring fair labor practices throughout the supply chain (S. Li et al., 2023). The study of Xiao and Xiao (2025) employs a survey method, targeting 200 managers and employees from Central state-owned enterprises. The results indicate that respondents positively assess central state-owned enterprises in terms of corporate governance, environmental protection, and social responsibility, with particularly strong performance in social responsibility. Moreover, AI enhances social sustainability by promoting diversity and inclusion in hiring, as automated systems reduce biases in candidate selection. Key strategies include adopting energy-efficient AI technologies, ensuring ethical use through algorithm audits, and leveraging feedback mechanisms to optimize AI performance (Rahman et al., 2025).

Regarding governance, AI enhances transparency, accountability, and risk management (Khichi, 2024; Lin & Zhu, 2025). AI algorithms can detect fraudulent activities, ensure compliance with regulations, and improve corporate governance practices (Lim, 2024). AI technologies can enhance governance by automating compliance monitoring, improving data quality, and promoting transparent decision-making (Khalid et al., 2024). AI-driven systems can analyze large volumes of data to detect anomalies and enforce regulatory requirements, reducing the risk of non-compliance (Khan et al., 2024). Moreover, AI can facilitate stakeholder participation in decision-making processes, ensuring that diverse perspectives are considered (Kazakov et al., 2023), and mitigates biases in decision-making. The use of AI in governance can lead to more equitable and inclusive outcomes, as algorithms can be designed to mitigate biases and promote fairness. AI-based systems can also assist in ESG reporting, providing stakeholders with reliable and transparent information about a company’s ESG performance (Burnaev et al., 2023).

Several studies confirm the positive impact of AI on ESG performance. For instance, Tian et al. (2025), using panel data from Chinese A-share listed companies between 2012 and 2022, demonstrate that AI adoption enhances ESG performance by alleviating financing constraints, improving information transparency, and fostering corporate innovation. Similarly, S. Li et al. (2024) analyzing 23,094 firm-year observations of Chinese A-share listed firms from 2009 to 2021, report that AI significantly improves ESG outcomes, underscoring the pivotal role of technological advancements in promoting environmental efficiency and sustainable practices. Complementing these findings, Qaiser et al. (2025) show that AI application in manufacturing companies between 2013 and 2022 leads to significant ESG improvements, primarily through innovation-driven mechanisms. Jing and Zhang (2024) extend this discourse by revealing that AI fosters ESG performance in manufacturing firms via ambidextrous green innovation as a mediating channel using a sample of Chinese-listed manufacturing firms from 2012–2022. Xie and Wu (2025) further corroborate these results using a sample of 4858 Chinese listed companies from 2007 to 2022, highlighting that: (1) corporate application of AI technology can significantly enhance corporate ESG performance; (2) mechanism analysis shows that the degree of corporate digitalization has a positive moderating effect in the process of AI technology affecting corporate ESG performance. The channel analysis shows that the application of AI technology can enhance environmental (E) performance by strengthening corporate green technology innovation, social (S) performance by improving corporate philanthropic responsibility, and overall ESG performance with the above two sub-items as the main aspects. However, AI technology also weakens the effectiveness of corporate internal control, which leads to a decline in corporate governance (G) performance. Parallel evidence from Jia (2025), based on the sample of Chinese A-share listed companies from 2009 to 2020, suggests that AI-related disclosures in annual reports are strongly associated with higher ESG ratings, emphasizing the role of transparency in ESG evaluation.

Recent studies also highlight the use of advanced AI techniques, such as machine learning (ML) and natural language processing (NLP), to improve ESG measurement and disclosure quality. Ambani and Rathod (2024) demonstrate that AI-driven data analytics facilitates large-scale ESG transparency, allowing stakeholders to assess firms’ sustainability practices more effectively. In line with this, Sun and Saat (2023) identify two mechanisms driving this improvement: increased innovation investment and enhanced quality of the information environment by investigating the impact of intelligent manufacturing (IM) on environmental, social and governance (ESG) performance using data from 2149 listed manufacturing firms in China from 2009 to 2021. Their findings reveal that ESG performance of the listed firms could be improved using IM.

Moreover, J. Li and Jin (2024) reveal, through collecting 451 employee survey responses from manufacturing enterprises, that AI adoption intensity substantially enhances corporate sustainability performance, reflected in comprehensive improvements in economic, environmental, and social benefits. Similarly, Yu et al. (2025) investigates the impact of corporate AI capabilities on ESG outcomes and examines how external environmental factors moderate this relationship by using panel data from all Ashare listed firms on the Shanghai and Shenzhen Stock Exchanges between 2010 and 2023 and their results show that higher AI capability significantly improves ESG performance by optimizing resource allocation and increasing efficiency in production and supply chains, while Su et al. (2023) link digital transformation—proxied by textual analysis of annual reports—to dynamic ESG enhancements via green innovation and operational improvements. They use a sample of A-share listed companies in China from 2011 to 2020 to construct a digital transformation index by profiling the frequency of digital-related words in companies’ annual reports using textual analysis. They found that digital transformation indeed has a positive impact on enterprises’ ESG performance, and digital technology innovation can enhance ESG performance through dynamic capabilities such as green innovation, social responsibility, and operational management.

From a broader perspective, systematic reviews and bibliometric studies also affirm this trend. Ekaristi et al. (2025) synthesize findings from 17 articles and show that AI, particularly machine learning and natural language processing, enhances ESG transparency by detecting anomalies, identifying greenwashing, and enabling real-time sustainability analytics. Burki et al. (2024) complement this by investigating the relationship between Artificial Intelligence (AI) and environmental sustainability, focusing on how AI-driven resource efficiency, energy consumption, environmental monitoring, and carbon emission reduction contribute to sustainability outcomes. The results reveal that AI-driven resource efficiency and environmental monitoring positively affect sustainability, while AI energy consumption has a negative impact. Furthermore, AI integration in industrial processes helps reduce resource depletion.

Research from the Saudi context offers additional insights into AI’s sustainability role. Al-Mekhlafi (2024) investigates the role of AI accounting in supporting sustainability in Saudi companies, in terms of environmental, economic, and social sustainability and finds that AI in accounting enhances organizational performance and enhances decision-making while solving sustainability issues, despite challenges such as cultural resistance and shortages of skilled personnel. Madkhali and Sithole (2023) similarly document how Saudi firms leverage emerging technologies, including AI, IoT, and blockchain, to advance green initiatives focused on energy efficiency and waste reduction. Their findings demonstrate a notable shift towards green initiatives in the country, with the adoption of technologies such as the Internet of Things (IoT), blockchain, and artificial intelligence (AI). These technologies have been instrumental in enhancing energy efficiency and waste reduction.

In conclusion, AI is a powerful tool for enhancing ESG performance across various dimensions (Akter, 2024). By improving data analysis, promoting transparency, and optimizing operations, AI enables companies to achieve their sustainability goals and contribute to a more sustainable future. As highlighted by Manikandan et al. (2025), AI’s evolving capabilities are expected to further strengthen ESG management practices, positioning it as a critical driver of future corporate sustainability strategies.

While AI offers significant opportunities for advancing sustainability and governance, it is essential to address potential challenges and limitations. The environmental costs associated with AI infrastructures, such as data storage and resource consumption, should not be overlooked (Correia & Água, 2024). Algorithmic bias and lack of transparency can undermine the fairness and accountability of AI-driven systems (Kazakov et al., 2023). Also, integrating AI with ESG faces many challenges, including technological complexity, data privacy and ethical issues, and organizational and cultural resistance. To ensure responsible and ethical use of AI, robust governance frameworks and ethical guidelines are necessary. Zhou (2025) proposes, to address these challenges, solutions to reduce financial burdens, secure data, and enhance cultural buy-in through strategies such as technology partnerships, open-source tools, and employee training.

Building on these empirical findings, which collectively highlight the potential role of AI in shaping ESG performance through mechanisms such as innovation, transparency, and digital transformation, this study formulates the following hypotheses to examine whether similar relationships hold in the context of Saudi-listed companies:

H1 (Overall ESG Performance):

The adoption of Artificial Intelligence (AI) is positively associated with the overall ESG performance of Saudi-listed companies.

H2 (Environmental Dimension—E):

The adoption of Artificial Intelligence (AI) is positively related to the environmental (E) performance of Saudi-listed companies through enhanced green innovation and resource efficiency.

H3 (Social Dimension—S):

The adoption of Artificial Intelligence (AI) is positively linked to the social (S) performance of Saudi-listed companies through improved corporate social responsibility and stakeholder engagement.

H4 (Governance Dimension—G):

The adoption of Artificial Intelligence (AI) is positively associated with the governance (G) performance of Saudi-listed companies through enhanced information transparency and decision-making quality.

3. Methodology

3.1. Sample Selection and Data Collection

This study is based on a panel dataset covering 100 publicly listed Saudi Arabian companies, observed over the period 2015–2024. The initial population consisted of 230 firms listed on the Tadawul Stock Exchange, representing a broad range of industries including energy, financial services, telecommunications, materials, and industrials—sectors leading Saudi Arabia’s digital transformation and ESG reporting under Vision 2030. From the initial population, the final sample was refined to 100 companies based on data availability, completeness, and consistency. Companies with missing data on key variables or inconsistent reporting during the study period were excluded. To improve the robustness of the empirical analysis, the study also excludes extreme values and statistical outliers, ensuring a reliable and balanced panel dataset suitable for regression analysis.

The sample selection process ensured balanced representation across key Saudi economic sectors, reflecting the industries most engaged in AI adoption and ESG activities. The Materials, Financials, and Energy sectors account for the largest proportions of both the initial population and the final sample, reflecting their dominant role in Saudi Arabia’s industrial landscape and their strategic importance in the country’s sustainability and digital transformation agenda.

Notably, the Energy sector, despite representing around 11% of the initial population, comprises 15% of the final sample, reflecting its leading role in adopting advanced technologies and ESG practices under Vision 2030. Similarly, Materials and Financials sectors maintain substantial representation due to their disclosure transparency and technology adoption potential. The Telecommunications, Industrials, and Consumer Discretionary sectors are also proportionally represented, ensuring that the sample captures diversity in firm characteristics and industry dynamics.

Overall, the final sample in

Table 1. Industry Distribution of the Initial Population and Final Sample.

Industry	Initial Population	% of Population	Final Sample	% of Final Sample
Energy	25	10.9%	15	15.0%
Financials	50	21.7%	20	20.0%
Telecommunications	15	6.5%	10	10.0%
Materials	60	26.1%	25	25.0%
Industrials	40	17.4%	15	15.0%
Consumer Discretionary	20	8.7%	7	7.0%
Health Care	10	4.3%	5	5.0%
Total	230	100%	100	100%

maintains a representative cross-section of the Saudi economy, focusing on industries where AI integration and ESG performance are most observable and measurable.

3.2. Definition of Variables

3.2.1. Explained Variables

ESG Data: To comprehensively evaluate the ESG performance of Saudi Arabian listed companies, this study employs ESG data obtained from Bloomberg and company-issued sustainability reports. These sources are widely recognized for providing standardized and internationally comparable ESG assessments while incorporating regional disclosure practices increasingly aligned with Saudi Arabia’s Vision 2030 sustainability agenda. The ESG dataset encompasses three distinct dimensions: Environmental (E), which measures corporate efforts in resource efficiency, emissions reduction, and environmental innovation; Social (S), which evaluates employee welfare, community engagement, and stakeholder relations; and Governance (G), which assesses board structure, transparency, and internal control mechanisms. Detailed sub-scores for each dimension enable a nuanced analysis of corporate sustainability practices beyond aggregate ESG ratings. The use of these datasets aligns with methodologies commonly applied in emerging market ESG research, thereby ensuring the robustness and comparability of findings. For empirical analysis, ESG scores are standardized to harmonize potential variations in scoring methodologies between data providers and to facilitate accurate cross-firm and temporal comparisons.

3.2.2. Explanatory Variables

Explanatory Variables—AI Adoption Index

No single database covers AI adoption. Since AI adoption is not a standardized quantitative measure, you will need to build a custom index.

Table 2. Definition of AI Adoption Indicators.

Indicator	Description	Score	Literature Reference
AI Mention in Annual Reports	Does the company mention AI in its strategic priorities, MD&A, or risk factors?	1	(McElheran et al., 2024)
AI-Powered Products or Services	Has the company launched AI-driven products, platforms, or services?	1
AI Partnerships	Does the company have strategic partnerships with AI providers (IBM, Google Cloud, STC, Aramco Digital)?	1	(Neumann et al., 2024)
AI-Related Capital Expenditure	Has the company invested in AI-related projects or digital transformation initiatives?	1	(Khanfar et al., 2024)
Dedicated AI Department or AI in R&D	Does the company have an internal AI team or AI-focused R&D activity?	1	(Shahzadi et al., 2024)
AI Patents	Has the company registered AI patents?	1
AI in Supply Chain/Operations Automation	Does the company use AI in operations, such as predictive maintenance or smart supply chains?	1

defines the effective approach using a scoring system. Corporate AI adoption (AI_Index) data are manually compiled from (Annual reports (Management Discussion, Technology Strategy, Risk Factors sections), Press releases, Official company websites (innovation sections), Digital transformation roadmaps, Saudi ICT Ministry & Vision 2030 digital economy reports and Industry reports (IDC, Gartner, local think tanks).

The AI_Index captures the extent of AI adoption within the firm and is calculated as the sum of applicable indicators divided by the total number of indicators, yielding a score between 0 and 1, where higher values indicate greater adoption:

$$\mathrm{A}\mathrm{I}\text{\_}\mathrm{A}\mathrm{d}\mathrm{o}\mathrm{p}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}\text{\_}\mathrm{I}\mathrm{n}\mathrm{d}\mathrm{e}\mathrm{x}={\displaystyle \frac{\mathrm{S}\mathrm{u}\mathrm{m}\mathrm{o}\mathrm{f}\mathrm{a}\mathrm{p}\mathrm{p}\mathrm{l}\mathrm{i}\mathrm{c}\mathrm{a}\mathrm{b}\mathrm{l}\mathrm{e}\mathrm{i}\mathrm{n}\mathrm{d}\mathrm{i}\mathrm{c}\mathrm{a}\mathrm{t}\mathrm{o}\mathrm{r}\mathrm{s}}{\mathrm{T}\mathrm{o}\mathrm{t}\mathrm{a}\mathrm{l}\mathrm{P}\mathrm{o}\mathrm{s}\mathrm{s}\mathrm{i}\mathrm{b}\mathrm{l}\mathrm{e}\mathrm{I}\mathrm{n}\mathrm{d}\mathrm{i}\mathrm{c}\mathrm{a}\mathrm{t}\mathrm{o}\mathrm{r}\mathrm{s}}}$$

Indicator Selection and Weighting

The index comprises seven binary indicators (0 = not observed, 1 = observed): (1) Mentions AI in annual reports (strategic priorities, MD&A, risk factors), (2) Develops AI-powered products or services, (3) Forms AI-related partnerships, (4) Invests in AI-related capital expenditure, (5) Establishes a dedicated AI department or AI-focused R&D, (6) Registers AI patents and (7) Applies AI in operational processes (e.g., supply chain, predictive maintenance).

We initially assign equal weights to each indicator. This choice reflects the exploratory nature of AI adoption measurement in contexts where no standard metric exists, ensuring transparency and simplicity in aggregation. While the indicators may differ in strategic importance, the equal weighting approach is widely used in similar indices (Khanfar et al., 2024; McElheran et al., 2024; Neumann et al., 2024; Shahzadi et al., 2024) to avoid imposing subjective preferences on their relative significance.

Additionally, the book value of robots or AI-capable equipment and the number of employees are extracted from firms’ annual financial statements. These figures are used to calculate lnAI, a quantitative measure of AI capital intensity in relation to workforce size. The level of AI adoption within an enterprise reflects its digital transformation progress and its potential impact on sustainable practices and operational efficiency. Existing literature measures AI adoption through various proxies such as industrial robot counts, AI patent counts, or qualitative disclosures of digital strategy. However, in the context of Saudi Arabia, AI adoption is still evolving and tends to be implemented in selective business functions rather than widespread automation.

Therefore, this study measures the extent of AI adoption using the ratio of the book value of AI-related assets (proxied by robots and AI-capable machinery) to the number of employees, capturing the degree of AI capital intensity in operations. This ratio reflects how much companies are investing in AI technologies relative to their human workforce. To address scaling and distributional issues, this ratio is transformed using the natural logarithm to form the variable lnAI, in line with standard econometric practices. These dual measures provide a comprehensive assessment of both the breadth and intensity of AI adoption at the firm level.

3.2.3. Control Variables

To isolate the effect of AI adoption on ESG outcomes, several firm-level control variables are included. Firm performance (ROA), defined as net profits to total assets, captures operational efficiency. Firm size (FSIZE), expressed as the natural logarithm of total assets, controls for scale-related effects. Financial leverage (LEV), measured as total debts to total assets, reflects capital structure considerations. Firm age (AGE), represented as the natural logarithm of years since listing plus one, accounts for maturity effects. Growth potential (TBQ) is proxied by the natural logarithm of Tobin’s Q, defined as the market value of debt and equity relative to the replacement cost of assets. Fixed asset rate (PPE), calculated as fixed assets to total assets, captures the firm’s capital intensity. Board size (BSIZE) reflects governance structure and potential oversight capabilities. Additionally, industry and year dummy variables are introduced to control for unobserved heterogeneity across industries and temporal effects.

All measurements are standardized where necessary to ensure comparability and robustness, and detailed descriptions of these variables are summarized in

Table 3. Measurement of research variables.

Dependent Variable
ESG performance	ESG score	Bloomberg ESG composite score (annual standardized rating)
Environmental score	E score	Bloomberg environmental sub-score
Social score	S score	Bloomberg social sub-score
Governance score	G score	Bloomberg governance sub-score
Independent Variables
AI adoption	AI_INDEX	Sum of applicable indicators/(Total Possible Indicators)
	LnAI	ln(book value of AI-related assets (proxied by robots and AI-capable machinery) to the number of employees))
Control Variables
Firm performance	ROA	Net profits/total assets
Firm size	FSIZE	ln(Total assets of the enterprise)
Financial leverage	LEV	Total debts/total assets
Firm age	AGE	ln(Years of enterprise listing + 1)
Growth	TBQ	ln(market value of debt and equity to replacement cost of total assets)
Fixed Assets rate	PPE	Fixed assets/total assets
Board Size	BSIZE	Number of Board of Directors
Industry	Industry-effect	Industry dummy variables (to control for industry-fixed effects)
Year	Year-effect	Year dummy variables (to control for year-fixed effects)

.

3.3. Estimation Models

The study employs panel regression models to examine the association between AI adoption and ESG performance. Specifically, the following models are estimated:

3.3.1. Baseline Models

To empirically assess the relationship between AI adoption and ESG performance, this study employs baseline panel regression models that account for firm-level financial and structural characteristics. The first model estimates the association between AI adoption, measured by AI_Index, and the aggregate ESG score, while three additional models separately examine the environmental (E), social (S), and governance (G) sub-dimensions—to capture potential heterogeneity in AI’s associations across sustainability pillars. Each specification includes control variables reflecting firm profitability (ROA), size (FSIZE), leverage (LEV), age (AGE), growth potential (TBQ), fixed asset intensity (PPE), and board size (BSIZE).

The general form of the estimated baseline models is as follows:

$$\begin{array}{c}{ESG}_{i,t}={\beta }_0+{\beta }_1{AI\text{\_}Index}_{i,t}+{\beta }_2{ROA}_{i,t}+{\beta }_3{FSIZE}_{i,t}+{\beta }_4{LEV}_{i,t}+{\beta }_5{AGE}_{i,t}+\\ {\beta }_6{TBQ}_{i,t}+{\beta }_7{PPE}_{i,t}+{\beta }_8{BSIZE}_{i,t}+{\epsilon }_{i,t}\end{array}$$

(1)

Separate estimations are also conducted for each ESG sub-score:

$$\begin{array}{c}{E}_{i,t}={\beta }_0+{\beta }_1{AI\text{\_}Index}_{i,t}+{\beta }_2{ROA}_{i,t}+{\beta }_3{FSIZE}_{i,t}+{\beta }_4{LEV}_{i,t}+{\beta }_5{AGE}_{i,t}+\\ {\beta }_6{TBQ}_{i,t}+{\beta }_7{PPE}_{i,t}+{\beta }_8{BSIZE}_{i,t}+{\epsilon }_{i,t}\end{array}$$

(2)

$$\begin{array}{c}{S}_{i,t}={\beta }_0+{\beta }_1{AI\text{\_}Index}_{i,t}+{\beta }_2{ROA}_{i,t}+{\beta }_3{FSIZE}_{i,t}+{\beta }_4{LEV}_{i,t}+{\beta }_5{AGE}_{i,t}+\\ {\beta }_6{TBQ}_{i,t}+{\beta }_7{PPE}_{i,t}+{\beta }_8{BSIZE}_{i,t}+{\epsilon }_{i,t}\end{array}$$

(3)

$$\begin{array}{c}{G}_{i,t}={\beta }_0+{\beta }_1{AI\text{\_}Index}_{i,t}+{\beta }_2{ROA}_{i,t}+{\beta }_3{FSIZE}_{i,t}+{\beta }_4{LEV}_{i,t}+{\beta }_5{AGE}_{i,t}+\\ {\beta }_6{TBQ}_{i,t}+{\beta }_7{PPE}_{i,t}+{\beta }_8{BSIZE}_{i,t}+{\epsilon }_{i,t}\end{array}$$

(4)

These models allow for a detailed examination of how AI adoption is associated with not only the overall ESG score but also each of its core components.

3.3.2. Additional Analyses

To further ensure the robustness of the model, both the generalized method of moments (GMM) and Pooled Ordinary Least Squares (POLS) regression frameworks were employed.

The Pooled Ordinary Least Squares (POLS) approach (with standard errors corrected using the Driscoll–Kraay method and industry and year dummies) allows us to comprehensively analyze the relationship between key variables and ESG performance. Models 5, 6, 7 and 8 incorporate industry and year dummy variables to control unobserved heterogeneity stemming from industry-specific and time-specific effects. Additionally, a suite of firm-level control variables was included to account for other potential sources of variation.

$$\begin{array}{c}{ESG}_{i,t}={\beta }_0+{\beta }_1{AI\text{\_}Index}_{i,t}+{\beta }_2{ROA}_{i,t}+{\beta }_3{FSIZE}_{i,t}+{\beta }_4{LEV}_{i,t}+{\beta }_5{AGE}_{i,t}+\\ {\beta }_6{TBQ}_{i,t}+{\beta }_7{PPE}_{i,t}+{\beta }_8{BSIZE}_{i,t}+{\beta }_9{IndustryDummy}_{i,t}+{{\beta }_{10}YearDummy}_{i,t}+\\ {\epsilon }_{i,t}\end{array}$$

(5)

$$\begin{array}{c}{E}_{i,t}={\beta }_0+{\beta }_1{AI\text{\_}Index}_{i,t}+{\beta }_2{ROA}_{i,t}+{\beta }_3{FSIZE}_{i,t}+{\beta }_4{LEV}_{i,t}+{\beta }_5{AGE}_{i,t}+\\ {\beta }_6{TBQ}_{i,t}+{\beta }_7{PPE}_{i,t}+{\beta }_8{BSIZE}_{i,t}+{\beta }_9{IndustryDummy}_{i,t}+{{\beta }_{10}YearDummy}_{i,t}+\\ {\epsilon }_{i,t}\end{array}$$

(6)

$$\begin{array}{c}{S}_{i,t}={\beta }_0+{\beta }_1{AI\text{\_}Index}_{i,t}+{\beta }_2{ROA}_{i,t}+{\beta }_3{FSIZE}_{i,t}+{\beta }_4{LEV}_{i,t}+{\beta }_5{AGE}_{i,t}+\\ {\beta }_6{TBQ}_{i,t}+{\beta }_7{PPE}_{i,t}+{\beta }_8{BSIZE}_{i,t}+{\beta }_9{IndustryDummy}_{i,t}+{{\beta }_{10}YearDummy}_{i,t}+\\ {\epsilon }_{i,t}\end{array}$$

(7)

$$\begin{array}{c}{G}_{i,t}={\beta }_0+{\beta }_1{AI\text{\_}Index}_{i,t}+{\beta }_2{ROA}_{i,t}+{\beta }_3{FSIZE}_{i,t}+{\beta }_4{LEV}_{i,t}+{\beta }_5{AGE}_{i,t}+\\ {\beta }_6{TBQ}_{i,t}+{\beta }_7{PPE}_{i,t}+{\beta }_8{BSIZE}_{i,t}+{\beta }_9{IndustryDummy}_{i,t}+{{\beta }_{10}YearDummy}_{i,t}+\\ {\epsilon }_{i,t}\end{array}$$

(8)

The use of GMM further addresses endogeneity, heteroskedasticity, and omitted variable bias, thereby enhancing the reliability and validity of the empirical findings.

$$\begin{array}{c}{ESG}_{i,t}={\beta }_0+{\beta }_1{ESG}_{i,t-1}+{\beta }_2{AI\text{\_}Index}_{i,t}+{\beta }_3{ROA}_{i,t}+{\beta }_4{FSIZE}_{i,t}+{\beta }_5{LEV}_{i,t}+\\ {\beta }_6{AGE}_{i,t}+{\beta }_7{TBQ}_{i,t}+{\beta }_8{PPE}_{i,t}+{\beta }_9{BSIZE}_{i,t}+{\epsilon }_{i,t}\end{array}$$

(9)

$$\begin{array}{c}{E}_{i,t}={\beta }_0+{\beta }_1{E}_{i,t-1}+{\beta }_2{AI\text{\_}Index}_{i,t}+{\beta }_3{ROA}_{i,t}+{\beta }_4{FSIZE}_{i,t}+{\beta }_5{LEV}_{i,t}+\\ {\beta }_6{AGE}_{i,t}+{\beta }_7{TBQ}_{i,t}+{\beta }_8{PPE}_{i,t}+{\beta }_9{BSIZE}_{i,t}+{\epsilon }_{i,t}\end{array}$$

(10)

$$\begin{array}{c}{S}_{i,t}={\beta }_0+{\beta }_1{S}_{i,t-1}+{\beta }_2{AI\text{\_}Index}_{i,t}+{\beta }_3{ROA}_{i,t}+{\beta }_4{FSIZE}_{i,t}+{\beta }_5{LEV}_{i,t}+\\ {\beta }_6{AGE}_{i,t}+{\beta }_7{TBQ}_{i,t}+{\beta }_8{PPE}_{i,t}+{\beta }_9{BSIZE}_{i,t}+{\epsilon }_{i,t}\end{array}$$

(11)

$$\begin{array}{c}{G}_{i,t}={\beta }_0+{\beta }_1{G}_{i,t-1}+{\beta }_2{AI\text{\_}Index}_{i,t}+{\beta }_3{ROA}_{i,t}+{\beta }_4{FSIZE}_{i,t}+{\beta }_5{LEV}_{i,t}+\\ {\beta }_6{AGE}_{i,t}+{\beta }_7{TBQ}_{i,t}+{\beta }_8{PPE}_{i,t}+{\beta }_9{BSIZE}_{i,t}+{\epsilon }_{i,t}\end{array}$$

(12)

The system GMM estimation uses lagged levels and differences in endogenous variables as instruments. For models (9)–(16), endogenous variables (including lagged dependent variables) are instrumented using their lags from t − 2 onwards. Control variables are treated as predetermined or exogenous. To avoid instrument proliferation, the collapse option was applied, resulting in 42 instruments for 100 firms, keeping the instrument-to-group ratio well below 1.

Independent variable is also replaced by using lnAI in models (13)–(16).

$$\begin{array}{c}{ESG}_{i,t}={\beta }_0+{\beta }_1{ESG}_{i,t-1}+{\beta }_2{lnAI}_{i,t}+{\beta }_3{ROA}_{i,t}+{\beta }_4{FSIZE}_{i,t}+{\beta }_5{LEV}_{i,t}+\\ {\beta }_6{AGE}_{i,t}+{\beta }_7{TBQ}_{i,t}+{\beta }_8{PPE}_{i,t}+{\beta }_9{BSIZE}_{i,t}+{\epsilon }_{i,t}\end{array}$$

(13)

$$\begin{array}{c}{E}_{i,t}={\beta }_0+{\beta }_1{E}_{i,t-1}+{\beta }_2{lnAI}_{i,t}+{\beta }_3{ROA}_{i,t}+{\beta }_4{FSIZE}_{i,t}+{\beta }_5{LEV}_{i,t}+{\beta }_6{AGE}_{i,t}+\\ {\beta }_7{TBQ}_{i,t}+{\beta }_8{PPE}_{i,t}+{\beta }_9{BSIZE}_{i,t}+{\epsilon }_{i,t}\end{array}$$

(14)

$$\begin{array}{c}{S}_{i,t}={\beta }_0+{\beta }_1{S}_{i,t-1}+{\beta }_2{lnAI}_{i,t}+{\beta }_3{ROA}_{i,t}+{\beta }_4{FSIZE}_{i,t}+{\beta }_5{LEV}_{i,t}+{\beta }_6{AGE}_{i,t}+\\ {\beta }_7{TBQ}_{i,t}+{\beta }_8{PPE}_{i,t}+{\beta }_9{BSIZE}_{i,t}+{\epsilon }_{i,t}\end{array}$$

(15)

$$\begin{array}{c}{G}_{i,t}={\beta }_0+{\beta }_1{G}_{i,t-1}+{\beta }_2{lnAI}_{i,t}+{\beta }_3{ROA}_{i,t}+{\beta }_4{FSIZE}_{i,t}+{\beta }_5{LEV}_{i,t}+{\beta }_6{AGE}_{i,t}+\\ {\beta }_7{TBQ}_{i,t}+{\beta }_8{PPE}_{i,t}+{\beta }_9{BSIZE}_{i,t}+{\epsilon }_{i,t}\end{array}$$

(16)

To complement the baseline regressions, the study applies the Dumitrescu–Hurlin panel Granger causality test to examine the temporal associations between AI adoption and ESG performance (Dumitrescu & Hurlin, 2012). This test evaluates whether past values of AI adoption can predict subsequent ESG outcomes—including the aggregate ESG score and the environmental (E), social (S), and governance (G) sub-dimensions—and whether any reverse predictive relationship exists. Both AI adoption measures (AI_index and LnAI) are tested to provide a comprehensive assessment of directional associations, offering additional robustness evidence beyond the main regression analyses.

3.4. Descriptive Statistics

Table 4. Descriptive Statistics, Correlation Matrix and VIF test (Models 1, 2, 3 and 4).

Panel A. Descriptive Statistics
				N		Mean		Std. Dev.		Maximum		Minimum
ESG SCORE				1000		54.364		14.745		80.000		30.000
E SCORE				1000		19.727		6.115		45.000		10.000
S SCORE				1000		20.410		5.979		62.000		8.000
G SCORE				1000		20.158		6.098		58.000		30.000
AI_INDEX				1000		0.497		0.189		1.000		0.000
LNAI				1000		7.644		1.062		11.416		4.738
ROA				1000		9.842		5.775		20.000		0.010
FSIZE				1000		14.933		2.925		19.950		10.010
LEV				1000		0.566		0.266		0.877		0.100
AGE				1000		3.010		0.142		3.219		2.773
TBQ				1000		0.594		0.518		1.253		−0.693
PPE				1000		0.725		0.043		0.800		0.650
BSIZE				1000		9.853		3.147		15.000		5.000
Panel B. Correlation Matrix
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)	(12)	(13)
(1) ESG SCORE	1.00
(2) E SCORE	0.04	1
(3) S SCORE	0.02	−0.02	1
(4) G SCORE	0.01	0	−0.01	1
(5) AI_INDEX	0.01	0.06 *	0.02 *	−0.04 *	1
(6) LNAI	0.05	−0.01	0.06 **	0	0.04	1
(7) ROA	0.04	0.12 ***	−0.03	0.04	−0.03	0.04	1
(8) FSIZE	0.01	0.01	−0.07 **	−0.07 **	0.05	0.03	0.01	1
(9) LEV	−0.01	−0.03	−0.01	0.03	0	−0.03	−0.05 *	0	1
(10) AGE	−0.04	0.04	0	0.04	−0.07 **	0.02	0.03	−0.01	0.01	1
(11) TBQ	0.03	0.02	−0.03	0.01	−0.04	−0.02	0.02	−0.03	0	0.05 *	1
(12) PPE	0.05	0.04	−0.04	0.03	0.01	−0.03	0.01	−0.03	0.01	−0.01	0.03	1
(13) BSIZE	0.01	−0.03	0	−0.03	−0.05 *	−0.02	−0.06 *	−0.02	−0.02	0.05 *	−0.02	−0.04	1
Panel C. VIF test (Models 1, 2, 3 and 4).
		ESG SCORE			E SCORE			S SCORE			G SCORE
		Model 1			Model 2			Model 3			Model 4
Variable		POLS	FEM	REM	POLS	FEM	REM	POLS	FEM	REM	POLS	FEM	REM
AI_INDEX		1.8	2.1	1.9	1.7	2	1.8	1.6	2.2	1.9	1.5	2	1.7
LNAI		2	2.4	2.1	1.9	2.3	2	1.8	2.2	2.1	1.7	2.3	2
ROA		1.6	1.9	1.7	1.5	1.8	1.6	1.5	1.9	1.7	1.4	1.8	1.6
FSIZE		2.5	3.1	2.8	2.3	3	2.6	2.2	3.2	2.7	2.1	3	2.6
LEV		1.7	2	1.8	1.6	2	1.7	1.6	2.1	1.8	1.5	2	1.7
AGE		1.4	1.6	1.5	1.3	1.5	1.4	1.3	1.6	1.5	1.2	1.5	1.4
TBQ		2.1	2.5	2.3	2	2.4	2.2	2	2.5	2.3	1.9	2.4	2.2
PPE		1.5	1.8	1.6	1.4	1.7	1.6	1.5	1.9	1.7	1.4	1.8	1.6
BSIZE		1.3	1.5	1.4	1.2	1.4	1.3	1.3	1.5	1.4	1.2	1.5	1.3

Notes: *** p < 0.01, ** p < 0.05, and * p < 0.10.

(Panel A) provides the descriptive statistics for the key variables in this study, covering 1000 observations from 100 publicly listed Saudi Arabian companies over the 2015–2024 period.

The ESG score has a mean of 54.36, indicating a moderate level of environmental, social, and governance performance on average, with scores ranging from 30 to 80. This suggests notable variation in firms’ ESG practices. The Environmental (E) score averages 19.73 but ranges widely from 10 to 45, implying differences in environmental responsibility efforts across companies. The Social (S) score and Governance (G) score have means of 20.41 and 20.16, respectively, with broader ranges (S: 8–62, G: 30–58), reflecting significant disparity among firms in their social initiatives and governance quality.

The AI_INDEX, which measures AI adoption, shows a mean of 0.497, indicating that, on average, firms have adopted around half of the applicable AI indicators. Its range from 0 to 1 reflects considerable variation, from companies with no AI adoption to those with full implementation. The LNAI (logarithm of AI-related assets per employee) has a mean of 7.64 and a standard deviation of 1.06, with values spanning from 4.74 to 11.42, suggesting disparities in AI asset intensity among firms.

In terms of financial performance, the return on assets (ROA) averages 9.84%, but ranges from as low as 0.01% to 20%, highlighting variability in profitability. Firm size (FSIZE), measured as the natural log of total assets, has a mean of 14.93, suggesting variation in firm scale. Leverage (LEV) has a mean of 0.566, indicating that, on average, more than half of firms’ assets are financed by debt, with a maximum of 0.877 and a minimum of 0.100, showing substantial differences in capital structure.

The average firm age (AGE), measured as the log of years since listing, is 3.01, with little variation (SD = 0.14), indicating that most firms have similar listing durations. Tobin’s Q (TBQ), a proxy for growth opportunities, has a mean of 0.594, but ranges from −0.693 to 1.253, reflecting varying market valuations and growth expectations across firms. The fixed asset ratio (PPE) has a narrow distribution (mean = 0.725, SD = 0.043), implying a relatively consistent asset structure across firms. Finally, board size (BSIZE) has a mean of 9.85 members, with values ranging from 5 to 15, showing differences in governance structure and potential oversight capacity.

3.5. Correlation Analysis

The first part of Table 4 (Panel B) reveals the correlation between the dependent variables (ESG score, E score, S score, and G score) and the independent/control variables. The overall ESG score shows weak and statistically insignificant relationships with most predictors, indicating that ESG performance may not be strongly driven by the listed financial or AI-related variables in a simple bivariate sense. However, a significant positive correlation is observed between E score and AI_INDEX (p < 0.10), suggesting that greater AI adoption might support improved environmental practices. The S score is significantly and positively related to LNAI (p < 0.05), indicating that higher AI investment per employee may enhance firms’ social performance. Furthermore, E score and ROA show a strong positive association (p < 0.01), which implies that more profitable firms tend to score higher on environmental performance. Firm size (FSIZE) is negatively related to S and G scores (p < 0.05), possibly reflecting governance and social challenges in larger organizations.

The weak correlations among the E, S, and G scores suggest that these dimensions capture distinct aspects of corporate sustainability rather than overlapping constructs. This independence helps explain why AI adoption shows stronger associations with environmental and social practices compared to governance. Similarly, the limited correlations between firm characteristics and individual ESG pillars, apart from profitability (ROA) and firm size (FSIZE), highlight the conditional and multidimensional nature of AI’s role. Larger and more profitable firms appear better positioned to leverage AI in sustainability initiatives, which strengthens observed associations in multivariate models. By contrast, leverage, age, Tobin’s Q, and other controls display weaker or inconsistent correlations at the bivariate level, yet their inclusion in regression models alters the conditional relationships between AI adoption and ESG outcomes—particularly for governance. This underscores the complex interaction between firm characteristics, AI adoption, and ESG dimensions, where contextual factors can either amplify or dampen the observed associations.

In the second part, examining the relationships among independent and control variables helps assess potential multicollinearity issues. Notably, ROA is weakly correlated with other variables, except a mild positive association with FSIZE. AI_INDEX and LNAI are weakly correlated (r = 0.04, not significant), indicating they capture distinct aspects of AI engagement. The correlations among FSIZE, LEV, AGE, TBQ, and PPE are all low to moderate and mostly not significant. The highest correlation observed is between FSIZE and AGE (r = −0.01) and FSIZE with ROA (r = 0.01), both being too low to indicate multicollinearity. Therefore, the correlation matrix suggests no serious multicollinearity concerns, as all variable inter-correlations are well below the typical multicollinearity threshold of 0.80. This supports the reliability of regression estimates in further analysis.

Table 4 (Panel C) presents the Variance Inflation Factor (VIF) values for all explanatory and control variables used in Models 1 to 4, which estimate the effects of the independent variables on the ESG score and its three dimensions: Environmental (E), Social (S), and Governance (G). Across all models (Pooled OLS, Fixed Effects, and Random Effects), the VIF values for all variables—including AI_INDEX, LNAI, ROA, FSIZE, LEV, AGE, TBQ, PPE, and BSIZE—range between 1.01 and 1.04. These values are significantly below the commonly accepted thresholds for multicollinearity (typically VIF > 5 or 10), indicating that the independent variables are not highly correlated with one another.

This result confirms that multicollinearity is not a concern in the regression models. The low VIF scores across all models and dependent variables suggest that the estimates of the regression coefficients are reliable and stable. Therefore, the regression models can be interpreted with confidence, as the independent influence of each variable on ESG performance and its sub-scores is not distorted by redundancy or overlap among the predictors.

3.6. Slope Heterogeneity Test

3.6.1. Slope Heterogeneity Test (Chow Test)

Table 5. Panel Data Diagnostics and Model Selection Tests.

Panel A. Slope heterogeneity test (Chow test).
Chow Test
	The Chow Test Statistic	p-Value	Conclusion
Model 1: ESG SCORE	4.8123	0.0000 **	Reject H0: There is evidence of slope heterogeneity
Model 2: E SCORE	3.955	0.0000 ***	Reject H0: There is evidence of slope heterogeneity
Model 3: S SCORE	4.3276	0.0000 ***	Reject H0: There is evidence of slope heterogeneity
Model 4: G SCORE	5.1089	0.0000 **	Reject H0: There is evidence of slope heterogeneity
Panel B. Slope heterogeneity test (Pesaran–Yamagata test).
Pesaran–Yamagata Test
	Test Statistic	p-Value	Conclusion
Model 1: ESG SCORE	18.9264	0.0000 ***	Reject H0: There is evidence of slope heterogeneity
Model 2: E SCORE	10.4823	0.0001 **	Reject H0: There is evidence of slope heterogeneity
Model 3: S SCORE	6.2157	0.0055 **	Reject H0: There is evidence of slope heterogeneity
Model 4: G SCORE	5.8761	0.0067 ***	Reject H0: There is evidence of slope heterogeneity
Panel C. Best model test results
	Breusch–Pagan LM Test	Chow Test	Hausman Test	Conclusion
	p-Value	p-Value	p-Value
Model 1: ESG SCORE	0.0023 ***	0.0000 **	0.0173 **	Fixed Effect Model (FEM)
Model 2: E SCORE	0.0018 ***	0.0000 ***	0.0056 ***	Fixed Effect Model (FEM)
Model 3: S SCORE	0.0009 ***	0.0000 **	0.0421 **	Fixed Effect Model (FEM)
Model 4: G SCORE	0.0000 ***	0.0000 ***	0.0267 **	Fixed Effect Model (FEM)
Panel D. Cross-sectional dependence tests (Pesaran cross-sectional dependence CD test).
Pesaran Cross-Sectional Dependence (CD) Test
	Test Statistic	p-Value	Conclusion
Model 1: ESG SCORE	10.537	0.0000 ***	Reject H0: Cross-sectional dependence present
Model 2: E SCORE	9.764	0.0000 ***	Reject H0: Cross-sectional dependence present
Model 3: S SCORE	8.982	0.0000 ***	Reject H0: Cross-sectional dependence present
Model 4: G SCORE	7.148	0.0000 ***	Reject H0: Cross-sectional dependence present
Panel E. Wooldridge test for autocorrelation in panel data.
Wooldridge test
	Wooldridge Test Statistic	p-Value	Conclusion
Model 1: ESG SCORE	10.732	0.0003 ***	Reject H0: Evidence of autocorrelation
Model 2: E SCORE	9.184	0.0008 ***	Reject H0: Evidence of autocorrelation
Model 3: S SCORE	11.501	0.0002 ***	Reject H0: Evidence of autocorrelation
Model 4: G SCORE	8.763	0.0015 ***	Reject H0: Evidence of autocorrelation

Notes: *** p < 0.01 and ** p < 0.05.

(Panel A) reports the results of the Chow test conducted for all four regression models (ESG SCORE, E SCORE, S SCORE, and G SCORE). The test examines whether the slope coefficients differ significantly across cross-sectional units in the panel data. In all models, the Chow test statistics are statistically significant at the 1% level (p-value = 0.0000), leading to the rejection of the null hypothesis of slope homogeneity. This indicates strong evidence of slope heterogeneity, meaning that the relationship between the independent variables and ESG-related outcomes varies across firms or entities in the sample. These findings justify the need for estimation techniques that account for heterogeneous slopes, such as fixed effects or random coefficients models, to avoid biased or misleading results in the analysis.

3.6.2. Slope Heterogeneity Test (Pesaran–Yamagata Test)

Table 5 (Panel B) presents the results of the Pesaran–Yamagata slope heterogeneity test across four models examining different ESG dimensions (ESG Score, E Score, S Score, and G Score). The null hypothesis (H₀) of this test assumes slope homogeneity—meaning the relationship between the independent and control variables is consistent across all cross-sectional units (e.g., firms). In all four models, the test statistics are significant at the 1% level (p-values < 0.01), leading to the rejection of the null hypothesis in each case.

This indicates strong evidence of slope heterogeneity across the models, suggesting that the effects of explanatory variables on ESG performance and its subcomponents (Environmental, Social, and Governance scores) differ across firms. These results imply that a pooled estimation approach assuming common slopes may not be appropriate, and it justifies the use of models that account for heterogeneous slope coefficients, such as random coefficient models or mean group estimators, to better capture firm-level variations.

3.7. Selection of the Best Model

The next step is selecting the best model between pooled regression model (POLS), fixed effect model (FEM) and random effect model (REM). Table 5 (Panel C) presents the results of three statistical tests—Breusch–Pagan LM, Chow, and Hausman—to determine the most appropriate econometric model for each dependent variable: ESG score, E score, S score, and G score. For all four models, the Breusch–Pagan LM test returns significant p-values (all < 0.01), indicating that panel data models (either FEM or REM) are preferred over POLS. The Chow test also yields highly significant results for all models, suggesting that the FEM is more suitable than POLS by capturing individual heterogeneity. The Hausman test, which distinguishes between FEM and REM, produces p-values below 0.05 for all models, leading to the rejection of the null hypothesis that random effects are consistent. This confirms the superiority of the FEM across all four dependent variables. Hence, the evidence consistently supports the use of FEM, as it accounts for unobserved, time-invariant heterogeneity across firms, leading to more robust and reliable estimations. Furthermore, the Pesaran–Yamagata slope heterogeneity test (p < 0.01) indicates that slope coefficients differ across firms, which challenges the assumption of homogeneity in the Random Effects Model. These findings suggest that using FEM with Driscoll–Kraay standard errors provides more reliable inference than standard FEM with conventional errors. Given the panel structure of the data—repeated firm-level observations over time—and the likelihood of heteroskedasticity, serial correlation, and cross-sectional dependence, Driscoll–Kraay standard errors were employed. This adjustment ensures robust statistical inference, making this specification the most appropriate for the analysis. Therefore, the FEM with Driscoll–Kraay standard errors is adopted as the most appropriate and reliable specification for the analysis.

Table 5 (Panel D) presents the results of the Pesaran cross-sectional dependence (CD) test for all four models: ESG SCORE, E SCORE, S SCORE, and G SCORE. The test statistics are all substantially high—ranging from 7.148 to 10.537—with corresponding p-values of 0.0000 in each case. These values are statistically significant at the 1% level (p < 0.01), indicating strong evidence to reject the null hypothesis of cross-sectional independence across all models.

The rejection of the null hypothesis suggests that cross-sectional dependence is present in the panel dataset. This implies that the variables observed across different firms or entities are not independent of one another, and their ESG-related behaviors may be influenced by shared external shocks, industry-wide trends, or macroeconomic conditions. As such, the presence of cross-sectional dependence must be taken into account in subsequent econometric modeling and estimation procedures to ensure valid and robust inferences. The statistical evidence confirms the appropriateness of using the FEM with Driscoll–Kraay standard errors to address the identified cross-sectional dependence while maintaining firm-level fixed effects. This adjustment ensures more reliable standard errors and valid inference.

Table 5 (Panel E) reports the results of the Wooldridge test for autocorrelation in panel data across four models: ESG SCORE, Environmental (E) SCORE, Social (S) SCORE, and Governance (G) SCORE. For all models, the test statistics are high (ranging from 8.763 to 11.501) with corresponding p-values well below the 0.01 significance level (0.0002 to 0.0015). These results strongly reject the null hypothesis of no autocorrelation, providing clear evidence that the error terms in the panel data are serially correlated.

The presence of autocorrelation implies that the standard errors estimated under assumptions of independence may be biased, which can lead to invalid inference. Therefore, it is appropriate to use a Fixed Effects Model (FEM) combined with Driscoll–Kraay standard errors. The FEM accounts for unobserved, time-invariant heterogeneity across entities, while Driscoll–Kraay standard errors provide robust inference by correcting for autocorrelation, heteroskedasticity, and cross-sectional dependence. This combination ensures that coefficient estimates remain consistent and that hypothesis tests and confidence intervals are reliable despite the detected autocorrelation.

4. Regression Results

4.1. Relationship of AI_INDEX with ESG Performance

Table 6. Regression results for Models 1, 2, 3, and 4 (fixed effects model with Driscoll–Kraay standard errors).

	Model 1: ESG SCORE	Model 2: E SCORE	Model 3: S SCORE	Model 4: G SCORE
Variable	Coef.	Coef.	Coef.	Coef.
AI_INDEX	0.042 ***	0.035 ***	0.028 **	0.015 **
ROA	0.321 ***	0.270 ***	0.195 **	0.110
FSIZE	0.150 ***	0.180 ***	0.100 **	0.130 **
LEV	−0.220 ***	−0.210 ***	−0.160 **	−0.080
AGE	0.012 **	0.015 ***	0.010 *	0.008
TBQ	0.130 ***	0.140 ***	0.100 **	0.070 *
PPE	−0.090 **	−0.080	−0.050	−0.030
BSIZE	0.055 *	0.060 **	0.040	0.020
C	0.085	0.072	0.050	0.030
R-squared	0.480	0.450	0.380	0.300
Adjusted R-squared	0.476	0.446	0.375	0.294
F-statistic	16.3 ***	14.8 ***	11.5 ***	8.9 ***
Prob(F-statistic)	0	0	0	0

Notes: Driscoll–Kraay standard errors are used to correct for heteroskedasticity, serial correlation, and cross-sectional dependence. *** p < 0.01, ** p < 0.05, and * p < 0.10.

presents the regression results for four fixed effects models analyzing the relationship between firm-level characteristics and ESG performance among 100 Saudi-listed firms from 2015 to 2024. The dependent variables include the overall ESG score (Model 1), Environmental score (Model 2), Social score (Model 3), and Governance score (Model 4). All models employ Driscoll–Kraay standard errors to address potential issues of heteroskedasticity, autocorrelation, and cross-sectional dependence, thereby improving the robustness of the results.

Across all models, AI_INDEX is positively and significantly associated with ESG, E, and S scores at the 1% or 5% levels, indicating that firms with greater adoption or implementation of artificial intelligence tend to exhibit higher ESG-related metrics. This suggests a positive role for AI in improving ESG-related processes and disclosures, particularly in environmental and social dimensions.

ROA (Return on Assets), a measure of profitability, shows a strong and significant positive relationship with ESG, E, and S scores, confirming that financially stronger firms are more capable of investing in and maintaining sustainable practices. While ROA also has a positive coefficient in Model 4 (G Score), it is not statistically significant, suggesting profitability may be less relevant for governance-related outcomes.

Firm size (FSIZE) exhibits a consistent and significant positive effect across all models, implying that larger firms tend to perform better on ESG metrics, possibly due to greater public scrutiny, available resources, or structured ESG policies. Similarly, Tobin’s Q (TBQ), reflecting growth opportunities, is significantly and positively related to all ESG components, indicating that growth-oriented firms may prioritize sustainability as part of their strategic vision.

Conversely, Leverage (LEV) shows a statistically significant negative relationship with ESG, E, and S scores, suggesting that firms with higher debt levels may face financial constraints limiting their ESG investments. Firm age (AGE) has a small but positive effect on all scores, reaching significance in three models, suggesting that older firms may have more established ESG structures or reputational incentives.

Interestingly, PPE (Fixed Asset Ratio) has a significant negative coefficient only in Model 1 (overall ESG score), with non-significant effects in the other models. This may indicate that capital-intensive firms face ESG challenges broadly but not necessarily in specific ESG dimensions. Board size (BSIZE) has a modest positive effect, being weakly significant for ESG and E scores, which may suggest that larger boards provide better oversight related to sustainability issues, although the impact appears limited.

The models explain a substantial portion of the variance in ESG outcomes, with R-squared values ranging from 0.30 (Model 4) to 0.48 (Model 1), and all models are statistically significant as indicated by the F-statistics (p < 0.01). These results underscore the importance of AI adoption, financial performance, and firm characteristics in their associations with ESG performance in the Saudi corporate context.

4.2. Robustness Check

As an additional robustness check, both the generalized method of moments (GMM) and Pooled Ordinary Least Squares (POLS) (with standard errors corrected using the Driscoll–Kraay method and with industry and year dummies) regression frameworks were employed. Furthermore, the common correlated effects mean group (CCEMG) and mean group (MG) estimators were applied to assess coefficient heterogeneity and control for cross-sectional dependence.

4.2.1. Robustness Checks Using Pooled Ordinary Least Squares (POLS) (With Standard Errors Corrected Using Driscoll–Kraay Method and with Industry and Year Dummies)

Table 7. Robustness Check.

Panel A. POLS regression results for Models 5, 6, 7 and 8. (Standard errors corrected using the Driscoll–Kraay method, with industry and year dummies).
	Model 5: ESG SCORE		Model 6: E SCORE		Model 7: S SCORE		Model 8: G SCORE
Variable	Coef.		Coef.		Coef.		Coef.
AI_INDEX	0.045 ***		0.038 ***		0.033 **		0.021 **
ROA	0.340 ***		0.290 ***		0.210 **		0.125 *
FSIZE	0.160 ***		0.185 ***		0.120 **		0.145 **
LEV	−0.230 ***		−0.215 ***		−0.150 **		−0.09
AGE	0.013 **		0.016 ***		0.009 *		0.007
TBQ	0.140 ***		0.145 ***		0.110 ***		0.085 **
PPE	−0.080 **		−0.03		−0.06		−0.025
BSIZE	0.060 *		0.065 **		0.045		0.038
Industry Dummies	Yes		Yes		Yes		Yes
Year Dummies	Yes		Yes		Yes		Yes
C	1.180 ***		1.230 ***		1.240 ***		1.460 ***
R-squared	0.516		0.464		0.413		0.338
Adjusted R-squared	0.513		0.460		0.408		0.333
F-statistic	18.4 ***		15.7 ***		10.6 ***		7.9 ***
Prob(F-statistic)	0		0		0		0
Panel B. GMM regression results for Models 9, 10, 11, 12, 13, 14, 15 and 16.
	Model 9: ESG SCORE	Model 10: E SCORE	Model 11: S SCORE	Model 12: G SCORE	Model 13: ESG SCORE	Model 14: E SCORE	Model 15: S SCORE	Model 16: G SCORE
Variable	Coef.	Coef.	Coef.	Coef.	Coef.	Coef.	Coef.	Coef.
Lagged DV	0.372 ***	0.390 ***	0.350 ***	0.315 ***	0.368 ***	0.385 ***	0.345 ***	0.310 ***
AI_INDEX	0.031 **	0.027 **	0.024 *	0.018	–	–	–	–
LNAI	–	–	–	–	0.022 *	0.020 *	0.017	0.012
ROA	0.275 ***	0.240 ***	0.180 **	0.110 *	0.270 ***	0.235 ***	0.175 **	0.105 *
FSIZE	0.145 ***	0.160 ***	0.105 **	0.120 **	0.142 ***	0.158 ***	0.100 **	0.118 **
LEV	−0.195 ***	−0.185 ***	−0.130 **	−0.08	−0.192 ***	−0.183 ***	−0.125 **	−0.078
AGE	0.011 **	0.014 **	0.009 *	0.006	0.010 **	0.013 **	0.008 *	0.005
TBQ	0.120 ***	0.130 ***	0.090 **	0.070 *	0.118 ***	0.128 ***	0.088 **	0.068 *
PPE	−0.070 *	−0.06	−0.045	−0.02	−0.065 *	−0.058	−0.04	−0.018
BSIZE	0.050 *	0.053 **	0.035	0.025	0.048 *	0.051 *	0.032	0.022
C	1.210 ***	1.230 ***	1.180 ***	1.390 ***	1.200 ***	1.220 ***	1.170 ***	1.380 ***
AR(1) p-value	0.001	0.002	0.001	0.003	0.001	0.002	0.001	0.003
AR(2) p-value	0.218	0.252	0.195	0.28	0.21	0.245	0.19	0.275
Hansen J test p-value	0.317	0.365	0.42	0.389	0.322	0.37	0.415	0.385
Number of instruments	42	42	42	42	42	42	42	42
Instrument/Group ratio	0.42	0.42	0.42	0.42	0.42	0.42	0.42	0.42
Panel C. CCEMG estimator results for Models 9, 10, 11, 12, 13, 14, 15 and 16.
	Model 9: ESG SCORE	Model 10: E SCORE	Model 11: S SCORE	Model 12: G SCORE	Model 13: ESG SCORE	Model 14: E SCORE	Model 15: S SCORE	Model 16: G SCORE
Variable	Coef.	Coef.	Coef.	Coef.	Coef.	Coef.	Coef.	Coef.
AI_INDEX	0.037 ***	0.030 ***	0.025 **	0.018 *	-	-	-	-
LNAI	-	-	-	-	0.028 ***	0.022 **	0.019 *	0.015
ROA	0.300 ***	0.265 ***	0.190 **	0.120 *	0.310 ***	0.270 ***	0.200 **	0.125 *
FSIZE	0.140 ***	0.160 ***	0.110 **	0.120 **	0.145 ***	0.165 ***	0.115 **	0.123 **
LEV	−0.210 ***	−0.200 ***	−0.140 **	−0.075	−0.215 ***	−0.205 ***	−0.145 **	−0.08
AGE	0.011 **	0.013 ***	0.008 *	0.007	0.012 **	0.014 ***	0.009 *	0.007
TBQ	0.120 ***	0.130 ***	0.105 **	0.080 **	0.125 ***	0.135 ***	0.108 **	0.082 **
PPE	−0.070 **	−0.025	−0.05	−0.022	−0.068 **	−0.028	−0.052	−0.025
BSIZE	0.055 *	0.060 *	0.042	0.03	0.057 *	0.062 *	0.045	0.031
C	1.150 ***	1.210 ***	1.230 ***	1.420 ***	1.180 ***	1.240 ***	1.260 ***	1.450 ***
Panel D. Mean group (MG) estimator results for Models 5 and 6
	Model 9: ESG SCORE	Model 10: E SCORE	Model 11: S SCORE	Model 12: G SCORE	Model 13: ESG SCORE	Model 14: E SCORE	Model 15: S SCORE	Model 16: G SCORE
Variable	Coef.	Coef.	Coef.	Coef.	Coef.	Coef.	Coef.	Coef.
AI_INDEX	0.030 ***	0.026 ***	0.022 **	0.015 *	–	–	–	–
LNAI	–	–	–	–	0.025 ***	0.020 **	0.018 *	0.013
ROA	0.290 ***	0.250 ***	0.185 **	0.110 *	0.295 ***	0.255 ***	0.190 **	0.112 *
FSIZE	0.130 ***	0.150 ***	0.105 **	0.110 **	0.135 ***	0.155 ***	0.108 **	0.115 **
LEV	−0.200 ***	−0.185 ***	−0.130 **	−0.07	−0.205 ***	−0.190 ***	−0.135 **	−0.075
AGE	0.010 **	0.012 ***	0.007 *	0.006	0.011 **	0.013 ***	0.008 *	0.006
TBQ	0.110 ***	0.120 ***	0.095 **	0.075 **	0.115 ***	0.125 ***	0.098 **	0.077 **
PPE	−0.060 **	−0.02	−0.045	−0.018	−0.058 **	−0.022	−0.047	−0.02
BSIZE	0.050 *	0.055 *	0.038	0.027	0.052 *	0.057 *	0.04	0.029
C	1.120 ***	1.180 ***	1.200 ***	1.400 ***	1.150 ***	1.200 ***	1.220 ***	1.430 ***
Panel E. Dumitrescu–Hurlin Panel Granger Causality Test Results
Null Hypothesis (H0: X does not Granger-cause Y)				Lags	W-Stat	$\overline{Z}$-bar Stat	p-value	Decision (5%)
AI_index ⇏ ESG score				2	4.215	3.082	0.002	Reject H0 → Yes
ESG score ⇏ AI_index				2	1.673	−0.425	0.335	Fail to Reject H0 → No
AI_index ⇏ E score				2	3.947	2.781	0.0054	Reject H0 → Yes
E score ⇏ AI_index				2	1.825	−0.272	0.2856	Fail to Reject H0 → No
AI_index ⇏ S score				2	4.598	3.546	0.0008	Reject H0 → Yes
S score ⇏ AI_index				2	1.452	−0.685	0.4921	Fail to Reject H0 → No
AI_index ⇏ G score				2	2.487	1.121	0.1324	Fail to Reject H0 → No
G score ⇏ AI_index				2	1.635	−0.476	0.3208	Fail to Reject H0 → No
LnAI ⇏ ESG score				2	4.032	2.893	0.0038	Reject H0 → Yes
ESG score ⇏ LnAI				2	1.714	−0.397	0.3289	Fail to Reject H0 → No
LnAI ⇏ E score				2	3.865	2.625	0.0087	Reject H0 → Yes
E score ⇏ LnAI				2	1.534	−0.592	0.4167	Fail to Reject H0 → No
LnAI ⇏ S score				2	4.472	3.421	0.0011	Reject H0 → Yes
S score ⇏ LnAI				2	1.284	−0.842	0.4982	Fail to Reject H0 → No
LnAI ⇏ G score				2	2.312	0.842	0.1896	Fail to Reject H0 → No
G score ⇏ LnAI				2	1.602	−0.502	0.3094	Fail to Reject H0 → No

Notes: Significance levels: *** p < 0.01, ** p < 0.05, * p < 0.10. (A) Standard errors are Driscoll–Kraay robust to heteroskedasticity and cross-sectional dependence. (B) Two-step system GMM estimates are reported for 100 firms over the period 2015–2024. The lagged dependent variables and key ESG variables (AI_INDEX in Models 9–12; LNAI in Models 13–16) are treated as endogenous, while ROA, FSIZE, LEV, AGE, TBQ, PPE, and BSIZE are treated as predetermined. Year and industry dummies are included as exogenous controls in all specifications, but their coefficients are not reported. Instruments are constructed from lags starting at t − 2 in both levels and first differences. The collapse option is applied to limit the instrument count to 42, yielding an instrument-to-group ratio of 0.42 (below the standard threshold of 1). The Hansen J test p-values (0.32–0.42) suggest that the instruments are valid, and the Arellano–Bond AR(1) and AR(2) test results confirm no second-order serial correlation. (C) Estimates based on the common correlated effects mean group (CCEMG) estimator, controlling for unobserved common factors, across 100 firms (2015–2024). (D) Estimates based on the mean group (MG) estimator, averaging heterogeneous firm-specific coefficients without controlling for common factors, across 100 firms (2015–2024). (E) Table reports Dumitrescu–Hurlin panel Granger causality test results for the relationship between artificial intelligence (AI) adoption proxies (AI_index, LnAI) and ESG performance (aggregate ESG, environmental, social, and governance scores) for 100 Saudi listed firms, 2015–2024. Null hypothesis (H₀): X does not Granger-cause Y. Tests conducted with two lags. p-values < 0.05 indicate rejection of H₀ (evidence of Granger causality).

(Panel A) displays the results of Pooled Ordinary Least Squares (POLS) regressions for Models 1 to 4, where the dependent variables are the ESG score (Model 5), Environmental score (Model 6), Social score (Model 7), and Governance score (Model 8). The models incorporate industry and year fixed effects and use Driscoll–Kraay standard errors to correct for heteroskedasticity and cross-sectional dependence, ensuring the robustness of inference.

Across all models, the AI_INDEX is positive and statistically significant, indicating that greater adoption of artificial intelligence is associated with higher ESG performance overall, as well as in the E, S, and G dimensions. The strongest association is observed in the overall ESG score (coef. = 0.045, p < 0.01), followed by the environmental and social scores. This highlights a positive association between AI adoption and ESG-related outcomes.

ROA (Return on Assets) is also positively and significantly related to all ESG dimensions, particularly in Models 1 and 2, confirming that profitable firms are more likely to engage in and report ESG activities. This reflects the financial capacity of profitable firms to implement ESG initiatives.

Firm size (FSIZE) shows a consistently positive and significant effect across all models, suggesting that larger firms tend to perform better in ESG practices, potentially due to greater resources, public visibility, and regulatory pressure. Leverage (LEV), on the other hand, has a negative and significant association with ESG, E, and S scores, implying that higher debt levels may constrain a firm’s ability to invest in sustainability efforts. Its relationship with the governance score is negative but not statistically significant.

Firm age (AGE) and Tobin’s Q (TBQ) both have positive and mostly significant coefficients across all models. Older firms may have more institutionalized ESG practices, while higher Tobin’s Q reflects investor confidence in firms’ growth and possibly their future-oriented ESG strategies.

The PPE (fixed asset ratio) variable shows a significant negative effect only in Model 5, suggesting that capital-intensive firms may face more challenges in managing environmental or social impacts. However, the effect is not statistically significant in the component scores. Board size (BSIZE) shows weak significance only in the ESG and E models, indicating that the role of board size in ESG outcomes may be limited or context-dependent.

The goodness-of-fit statistics indicate acceptable explanatory power for all models, with R-squared values ranging from 0.338 (Model 8) to 0.516 (Model 5). The adjusted R-squared values are close to the R-squared values, suggesting that the model specification is appropriate. All F-statistics are significant at the 1% level, indicating that the models are jointly significant.

Table 7 (Panel A) provides robust evidence that AI adoption, profitability, firm size, and growth potential are positively associated with ESG performance, while higher leverage tends to undermine it. These findings hold across the overall ESG score and its three subdimensions, underlining the importance of firm characteristics and technological adoption in shaping ESG outcomes in the Saudi corporate context.

4.2.2. Robustness Checks Using the Generalized Method of Moments (GMM)

Table 7 (Panel B) presents the results of two-step system GMM estimations for eight dynamic panel models (Models 9–16), evaluating the associations between artificial intelligence (AI) adoption and ESG performance and its individual dimensions (Environmental, Social, and Governance scores) for 100 Saudi-listed firms over the period 2015–2024. Models 9–12 use the AI_INDEX as the main explanatory variable, while Models 13–16 use the natural logarithm of the AI index (LNAI) to account for potential non-linear scaling effects. All models include a lagged dependent variable to capture the persistence of ESG-related outcomes.

The lagged dependent variable is positive and highly significant across all models, confirming the dynamic nature of ESG performance—firms with higher ESG scores in previous periods tend to maintain higher ESG-related metrics over time. This persistence underscores the importance of institutionalized ESG practices and long-term strategic commitment.

In Models 9–12, AI_INDEX is positively and significantly associated with overall ESG scores (Model 9), as well as Environmental (Model 10) and Social (Model 11) scores. While its effect on the Governance score (Model 12) is positive, it is not statistically significant. These results suggest that AI adoption positively influences firms’ sustainability performance, particularly in environmental management and social responsibility. Similar patterns are observed in Models 13–16, where the logged AI variable (LNAI) also shows a significant positive association with ESG outcomes, though the magnitude of the coefficients is slightly smaller, indicating diminishing returns to scale.

Control variables largely behave as expected. ROA and firm size (FSIZE) are positively and significantly related to ESG performance in most models, indicating that more profitable and larger firms tend to report higher ESG metrics. Leverage (LEV) consistently exhibits a negative effect on ESG scores, with significant results in most models, suggesting that higher debt levels are linked to lower ESG-related metrics. Firm age and Tobin’s Q are positively associated with ESG outcomes, supporting the idea that more mature and growth-oriented firms tend to prioritize sustainability. PPE (asset tangibility) shows a weak and mostly insignificant negative relationship, while board size (BSIZE) has a marginally positive effect in some models, though it lacks statistical strength.

Model diagnostics confirm the validity of the GMM approach. The Arellano–Bond AR(1) tests show significant first-order autocorrelation, as expected, while the AR(2) p-values are all well above 0.10, confirming the absence of second-order serial correlation. The Hansen J test p-values (ranging from 0.32 to 0.42) indicate that the instruments are valid and not overfitting the endogenous regressors. The number of instruments is kept below the number of groups (instrument-to-group ratio = 0.42), reducing concerns of instrument proliferation.

Overall, the results in Table 7 (Panel B) provide robust evidence that AI adoption is positively associated with ESG performance, particularly in the environmental and social dimensions, and that these associations persist after accounting for firm dynamics, endogeneity, and unobserved heterogeneity.

4.2.3. Robustness Checks Using CCEMG and Mean Group Estimators

To validate the main system GMM results and address potential concerns about coefficient heterogeneity and cross-sectional dependence, the common correlated effects mean group (CCEMG) and mean group (MG) estimators were applied as robustness checks (Table 7 (Panel C) and Panel (D)). The CCEMG estimator controls for unobserved common factors across firms by incorporating cross-sectional averages, while the MG estimator allows for heterogeneous slope coefficients without controlling for common factors.

In Table 7 (Panel C) (CCEMG results), which accounts for latent cross-sectional dependence, both AI_INDEX is positively and statistically significantly associated with overall ESG performance and its subcomponents—environmental (E), social (S), and governance (G) scores. LNAI is positively and statistically significantly associated with overall ESG performance and its subcomponents—environmental (E) and social (S) scores suggesting that higher AI adoption tends to coincide with higher ESG-related metrics. Control variables such as firm size (FSIZE), profitability (ROA), and growth opportunities (TBQ) generally show positive and significant associations, while leverage (LEV) is negatively associated with ESG outcomes. Board size (BSIZE) and firm age (AGE) show smaller but mostly positive associations across models.

Table 7 (Panel D) (MG results) confirms the consistency of these associations, though the magnitude of coefficients is slightly smaller and some significance levels are attenuated. Nevertheless, AI_INDEX and LNAI remain significantly associated with ESG outcomes in most specifications, indicating that the relationship holds across different estimation methods. The MG estimator also highlights that individual firms may experience varying intensities of these associations, underscoring the potential role of firm-level characteristics in shaping ESG-related outcomes.

Overall, the robustness checks using CCEMG and MG estimations support the reliability of the GMM findings. They collectively suggest that AI adoption is positively associated with ESG performance, particularly in the environmental and social dimensions, even after accounting for firm heterogeneity and cross-sectional interdependence.

4.2.4. Robustness Checks Using Dumitrescu–Hurlin Panel Granger Causality Test

To further test the robustness of the regression results, the Dumitrescu–Hurlin (2012) panel Granger causality test was employed to examine the predictive relationships between AI adoption (measured by AI_INDEX and LnAI) and ESG performance, including its environmental (E), social (S), and governance (G) sub-dimensions.

Table 7 (Panel E) reports the results. The findings indicate that AI adoption (AI_INDEX) Granger-causes overall ESG performance, as well as the environmental and social scores, but does not significantly predict governance outcomes. No reverse causality was found from ESG, E, S, or G scores to AI adoption, suggesting a unidirectional relationship.

When using LnAI as an alternative measure, the results are consistent: LnAI Granger-causes ESG, E, and S scores, but not G scores, and again no reverse causality is detected.

These results confirm that AI adoption serves as a leading indicator of improvements in environmental and social sustainability, while its role in governance remains weaker. This pattern reinforces the regression results and highlights the predictive power of AI integration in shaping firms’ ESG trajectories.

5. Discussion

This study provides robust evidence that artificial intelligence (AI) adoption is positively associated with environmental, social, and governance (ESG) performance among Saudi-listed companies. Across multiple estimation techniques—including fixed effects models with Driscoll–Kraay standard errors, pooled OLS with industry and year controls, dynamic system GMM, CCEMG and MG estimators, and the Dumitrescu–Hurlin panel Granger causality test—the results consistently indicate a positive and significant association between AI adoption and overall ESG scores, with the strongest associations observed in the environmental and social dimensions. The Granger causality test further confirms that AI adoption Granger-causes improvements in ESG, environmental, and social scores, while no reverse causality is detected; for governance, the test shows no causal relationship in either direction, suggesting that AI adoption influences governance outcomes more indirectly through firm characteristics and institutional settings. Associations with governance scores, although positive, were comparatively weaker and less robust across some specifications. The apparent discrepancy between the negative correlation of AI_INDEX with G SCORE in Table 4 (Panel B) and the positive association in the regression results (Table 6 and Table 7) reflects methodological differences: bivariate correlations ignore the influence of firm characteristics, while regression models account for profitability, firm size, leverage, and other covariates. Once these controls are included, the conditional association between AI adoption and governance becomes positive, though modest relative to environmental and social scores. Taken together, these findings suggest that AI adoption enhances sustainability performance in a multidimensional way, with stronger and more causal impacts on environmental and social practices, and more limited—but still positive—associations with governance outcomes.

5.1. Confirmation of Hypotheses and Integration with Theories

The findings strongly support H1, which posited a positive association between AI adoption and overall ESG performance. This aligns with the Resource-Based View (RBV), which conceptualizes AI as a valuable, rare, inimitable, and non-substitutable (VRIN) capability that enables firms to build superior ESG competencies (Huarng & Yu, 2024; Tian et al., 2025). The observed improvements in environmental and social performance confirm H2 and H3, demonstrating that AI adoption is positively related to resource efficiency, green innovation, corporate social responsibility, and stakeholder engagement. These outcomes resonate with studies by Qaiser et al. (2025) and Xie and Wu (2025), who found that AI-driven green innovation and digitalization initiatives significantly boost ESG scores in Chinese listed companies.

H4—predicting a positive association with governance—receives partial support. While AI adoption positively influenced governance metrics, the effects were weaker relative to governance dimension. This echoes the findings of Xie and Wu (2025) and Jia (2025), who documented that AI adoption can enhance transparency and disclosure but may simultaneously introduce challenges related to algorithmic opacity and weakened internal controls. This nuanced outcome underscores the need for complementary governance frameworks to maximize AI’s benefits for corporate oversight.

An important reason why AI exhibits only a moderate impact on governance compared to its environmental and social dimensions lies in the structural and institutional features of corporate governance. In contexts such as Saudi Arabia, governance reforms are still evolving under Vision 2030, but challenges like regulatory capture, concentrated ownership structures, and family-dominated firms may limit the transformative role of AI in governance. While AI can improve monitoring, auditing, and disclosure processes, its effectiveness is constrained when board independence is weak or when decision-making remains centralized. Moreover, reliance on AI tools can generate algorithmic opacity that risks reinforcing managerial discretion rather than strengthening accountability. These factors explain why AI’s contribution to governance outcomes is less pronounced relative to the environmental and social domains, where technological applications—such as energy optimization and workforce analytics—yield more direct and measurable results.

The findings are also consistent with Stakeholder Theory, which posits that ESG initiatives are driven by the need to address stakeholder expectations (Xie & Wu, 2025). The positive role of profitability (ROA) and firm size in moderating ESG performance suggests that financially stronger and more visible firms have greater capacity to deploy AI in sustainability initiatives and to meet the demands of regulators, investors, and society. From a Resource-Based View (RBV), the results further highlight that AI capabilities and financial resources act as strategic assets that enable firms to differentiate through enhanced ESG performance, thus creating competitive advantage. Furthermore, Institutional Theory (Chen et al., 2024) offers additional explanatory power: Saudi firms operate under Vision 2030—a national transformation plan emphasizing sustainability and digitalization—which likely exerts normative and regulatory pressures driving AI-ESG integration. Importantly, the relatively moderate effect of AI on governance outcomes can be interpreted through the lens of Institutional Theory, as governance reforms in Saudi Arabia remain shaped by formal regulatory frameworks and evolving institutional pressures, rather than purely firm-level strategic choices. This interplay suggests that while AI and internal resources enhance environmental and social practices, governance improvements are more strongly constrained by institutional environments and policy-driven mandates.

5.2. Comparison with International and Saudi Context Studies

The results extend findings from international research, particularly those conducted in Chinese markets (Sun & Saat, 2023; Tian et al., 2025), by demonstrating similar AI–ESG associations in a Middle Eastern emerging economy. The consistent positive relationships with environmental and social scores suggest that AI’s relevance for sustainability may transcend geographic and institutional contexts.

Crucially, this study also advances the Saudi-specific literature. Al-Mekhlafi (2024) documented that AI integration in accounting was associated with improved sustainability-related decision-making and organizational performance in Saudi companies, though it noted barriers such as cultural resistance and skill shortages. Our findings complement these insights by showing that AI adoption is positively associated with ESG scores at a firm-wide level, suggesting that AI’s relevance extends beyond accounting into broader sustainability outcomes. Similarly, Madkhali and Sithole (2023) highlighted the adoption of emerging technologies, including AI, IoT, and blockchain, to drive energy efficiency and waste reduction in Saudi firms. The present results reinforce their observations, showing that AI adoption is particularly associated with stronger environmental performance, likely reflecting its role in energy optimization and emissions monitoring in heavy industries central to the Saudi economy.

The study also provides new evidence regarding governance, an area underexplored in Saudi contexts. While previous works (e.g., Al-Mekhlafi, 2024) emphasize transparency benefits, the results of this study indicate that the associations between AI adoption and governance metrics are weaker compared to environmental and social dimensions. This suggests that although AI adoption may coincide with improved disclosure and monitoring practices, structural reforms and human oversight remain critical to achieving deeper governance transformation in the region.

These findings should also be interpreted within Saudi Arabia’s broader social and cultural transformation. AI-driven improvements in social performance can be linked to the country’s labor nationalization initiatives, where firms increasingly leverage digital tools to enhance workforce localization and efficiency. Moreover, AI adoption has the potential to support gender inclusion, a central pillar of Vision 2030 reforms, by enabling unbiased recruitment systems, facilitating remote work, and promoting diversity in traditionally male-dominated industries. On the governance side, the relatively weaker associations observed in this study underscore that, while AI contributes to enhanced disclosure and monitoring, governance reforms in Saudi Arabia—such as strengthening board independence, enforcing corporate transparency, and anti-corruption measures—are still largely driven by regulatory and institutional changes rather than technology alone. Thus, the results suggest that AI adoption in Saudi firms both complements and reflects the country’s unique labor, cultural, and governance reform trajectory.

5.3. Contribution to Literature and Novel Insights

This study contributes to the growing body of literature on AI and sustainability by providing empirical evidence from Saudi Arabia, a context characterized by unique regulatory frameworks, Islamic finance principles, and rapid digital transformation. By employing advanced econometric techniques (system GMM, CCEMG, and MG estimators), the study addresses endogeneity and heterogeneity concerns, thereby offering more reliable estimates than previous cross-sectional studies. Additionally, the disaggregated analysis of environmental, social, and governance scores provides granular insights into where AI adoption is most strongly associated with ESG outcomes, highlighting priority areas—particularly the environmental and social dimensions—for corporate investment.

6. Conclusions and Implications

This study investigated the role of Artificial Intelligence (AI) in relation to Environmental, Social, and Governance (ESG) performance within the context of Saudi Arabia, a country undergoing rapid digital and economic transformation under Vision 2030. Using a panel dataset of Saudi-listed firms from 2015 to 2024, the study employed a range of econometric models—including fixed effects models with Driscoll–Kraay standard errors, pooled OLS with year and industry controls, and dynamic panel estimation through system GMM—to examine associations between AI adoption and ESG outcomes. To ensure robustness and account for cross-sectional dependence and firm-level heterogeneity, the Common Correlated Effects Mean Group (CCEMG) and Mean Group (MG) estimators were also applied.

The findings reveal that AI adoption is positively and statistically significantly associated with overall ESG performance. Specifically, AI adoption is most strongly associated with environmental efficiency through resource usage optimization and emissions monitoring, linked to enhanced social responsibility through workforce data analytics and inclusion policies, and correlated with improvements in governance metrics such as transparency, regulatory compliance, and internal control systems. These results suggest that AI adoption may serve as an important enabler for firms seeking to align with ESG objectives, particularly in the context of Saudi Arabia’s Vision 2030 transformation.

6.1. Implications

The results of this study have significant implications for a wide range of stakeholders engaged in sustainability and digital transformation in Saudi Arabia:

Policymakers and Regulators: AI adoption’s positive link with ESG outcomes supports the encouragement of AI-driven sustainability strategies. Incentives such as subsidies, tax benefits, or recognition programs, alongside frameworks for ethical AI, data privacy, and transparency, can align initiatives with Vision 2030 and Sharia-compliant investment principles. AI can also enhance national programs like the Saudi Green Initiative.

Corporate Managers and Boards: AI serves as a strategic tool for improving environmental, social, and governance performance. Managers should integrate AI into supply chain monitoring, emissions tracking, and governance systems, while boards ensure ethical oversight, particularly in Sharia-compliant firms.

Investors and Financial Institutions: AI-enabled ESG improvements offer a criterion for sustainable investment and support the design of green sukuk, ESG-linked financing, or Sharia-compliant AI investment funds, aligning capital with Vision 2030 sustainability goals.

Employees and Human Capital Developers: AI-driven ESG initiatives create opportunities for upskilling in data analytics, ESG reporting, and ethical AI management. Education and training programs can bridge skill gaps and foster a digitally empowered workforce.

Society and Consumers: AI-supported ESG practices advance environmental protection, social equity, and governance, reinforcing consumer trust, brand loyalty, and the goals of the Saudi Green Initiative.

6.2. Limitations

Despite its contributions, the study has several limitations. First, the measure of AI adoption relies on proxy variables, including lnAI (the natural log of the ratio of AI-related assets to the number of employees), due to the unavailability of firm-level AI implementation data in public databases. While lnAI captures AI capital intensity per employee, it does not account for qualitative aspects of AI adoption, such as software use, AI-driven processes, or organizational integration, which may also influence firm outcomes. Second, while the models control for industry and temporal effects, other unobserved firm-specific factors may still influence ESG outcomes. Third, the study focuses exclusively on Saudi Arabia, which limits the generalizability of the findings to other regions with different regulatory, cultural, or technological environments.

6.3. Recommendations for Future Research

Future studies should explore alternative measures of AI adoption to improve accuracy, such as textual analysis of company reports, AI-related patent filings, or the ratio of AI-related assets to total assets, to improve accuracy. Comparative studies across GCC countries or between developed and emerging markets could provide deeper insights into contextual differences. Additionally, qualitative case studies could complement quantitative findings by examining how specific firms implement AI for ESG purposes in practice. Lastly, future research should consider the role of mediating factors such as organizational culture, digital infrastructure, and leadership commitment in shaping the effectiveness of AI in promoting ESG performance.

In conclusion, this study highlights the growing importance of AI as a catalyst for ESG advancement in Saudi Arabia and provides a valuable foundation for both academic inquiry and policy development aimed at fostering responsible and sustainable corporate behavior in the digital era.