Artificial Intelligence Technology and Corporate ESG Performance: Empirical Evidence from Chinese-Listed Firms

Abstract

In the era of artificial intelligence (AI), economic efficiency has an obvious role to play, but “non-economic benefits” have gradually become the focus of corporate attention; thus, environmental, social, and governance (ESG) has become a mainstream investment strategy. This paper empirically examines the impact of corporate application of AI technology on corporate ESG performance using a sample of 4858 listed companies in China from 2007 to 2022. The study finds that: (1) corporate application of AI technology can significantly enhance corporate ESG performance, and this conclusion still holds after a series of endogeneity treatments and robustness tests; (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; (3) Heterogeneity analysis shows that AI technology promotes ESG more significantly in more competitive industries and tech-nology-intensive firms, and more significantly in the eastern and central regions than in the western and northeastern regions, and that large- and medium-sized firms are similarly superior to small-sized firms, while medium-sized firms have more room for upward mobility than large-sized firms, which embody a higher promotion effect than large enterprises. This paper provides theoretical evidence that enterprises apply AI technology to improve ESG performance and empirical support around investing in ESG practices and promoting ESG development.

1. Introduction

With the continuous improvement of science and technology, the industrial revolution is also going on all the time. Following the rise of a series of artificial intelligence products such as the Large Language Model, China has stepped into a new round of artificial intelligence-led scientific and technological revolution, and in recent years, the Chinese government has introduced various policies related to guiding the development of industry and innovation.

A question that is closely related to the reality and policy context is: The economic benefits of promoting AI in enterprises, the digital transformation of enterprises, and the transformation of the entire technology industry are very obvious [1], but how do such initiatives affect the non-economic benefits in addition to the economic benefits? Economic development is the primary goal of every region and every enterprise, and indeed China has achieved rapid economic and social development in recent years, but because of this, climate change and severe environmental problems have also come to the fore. To address these issues, the Chinese government has also issued a series of policies that imply that Chinese companies are facing a green transition, and green finance is about to become a key focus for all types of companies, suggesting that “non-economic benefits” are becoming a decisive factor in long-term corporate development. ESG, as an emerging evaluation system in recent years, not only provides a measurement index for the degree of green finance of enterprises but also provides a general framework for enterprises to realize green transformation. ESG stands for Environmental, Social, and Governance, which can theoretically be a good measure of the “non-economic benefits” of enterprises. For example, it has been demonstrated in the literature that corporate ESG performance can promote innovation performance [2] and enhance green innovation [3], ESG ratings can inhibit corporate carbon emissions [4], and high-performance ESG engagement can improve corporate governance performance [5], etc. The above evidence suggests that ESG ratings have a significant impact on corporate governance.

The above evidence suggests that ESG ratings already have a certain foundation for evaluating the “non-economic benefits” of enterprises, and the subject of AI, as a new pursuit of enterprises in the direction of green finance, deserves to be pondered: Does the application of AI technology by enterprises promote their ESG performance? If there is a significant relationship, how positive or negative is this impact? Further, what are the differences in the mechanisms of action? What are the characteristics of the differences in the impact of AI technology adoption on ESG performance among firms in different regions? It is of great significance to systematically explore these questions and clarify the causal relationship as well as the internal mechanism to promote the reform of enterprises’ AI and put them into ESG practice. This paper will be based on the above issues. Specifically, this paper will be based on a research sample of Chinese-listed companies from 2007–2022, matched with Hutchison’s Corporate Social Responsibility (CSR) database and several authoritative ESG scoring databases: China Research Data Service Platform (CNRDS), Huazheng ESG Ratings, and Bloomberg ESG Ratings, to empirically test the impact of corporate application of AI technology on corporate ESG performance in multiple ways. Meanwhile, a series of robustness tests as well as endogeneity were conducted on top of the benchmark regressions, and the pro-cess of action was clarified with mechanism analyses and channel tests.

The rest of the paper is organized as follows: Section 2 compiles the existing relevant literature and puts forward the hypotheses of this paper accordingly; Section 3 introduces the data sources and variable settings of this paper; Section 4 reports the main empirical results on the impact of corporate application of AI technology on ESG performance; Section 5 provides further analysis; and Section 6 presents the conclusions and recommendations.

2. Literature Review and Theoretical Hypotheses

2.1. Artificial Intelligence Research

The scope of research on the application of artificial intelligence is relatively wide; For the application of AI in enterprises, its impact can be discussed from three aspects: employment, labor market and its development.

Different research levels have varying observational perspectives, but they mainly discuss it based on factors such as the automation efficiency improvements, decision-driven iterations, and risk management requirements brought about by advancements in artificial intelligence technology. They view it as a category of impact effects, assessing its impacts and mechanisms. Notably, existing literature measures this impact primarily in terms of external and internal aspects, indicating that the explanatory significance of their research conclusions also needs further differentiation.

Existing research can be divided into positive and negative attitudes towards the effects of AI on employment and the labor market. Giacomo Damioli et al. [6] showed in their study that AI technology has a significant positive impact on employment. Zhang Zhuo [7], in his study of the impact of AI-related industries on the number of jobs in the labor force, found that the importance of education and employment opportunities is increasing, and Georg Graetz and Guy Michaels [8] analyzed panel data from 17 countries and showed that although robots reduce the employment share of low-skilled workers, they do not significantly reduce total employment. In contrast, Daron Acemoglu and Pascual Restrepo [9] show that the impact of robots is different from that of other capital and technologies, with the addition of one robot per 1000 workers being able to reduce the employment-to-population ratio by 0.2% and salaries by 0.42%, and Daron Acemoglu and Pascual Restrepo [10] proposed a framework to analyze the impact of technological changes such as automation on labor demand and showed that automation always reduces the labor share of value-added and has the potential to reduce labor demand. Zhang Xinchun et al. [11] found that the substitution effect of the robotics industry is more prevalent in the middle-skill labor force based on 2000–2019 data of Chinese microenterprises. Sun Wenyuan [12] et al. examined the differences in the impact of robots on employment across sectors and showed a U-shaped relationship between overall employment and robot use. In addition, some literature has diversified into examining the impact of AI on employment. Wang Ting [13] used the instrumental variable “source of differences in robot use between firms” in his study, confirming the positive correlation between robot use and the increase in employment within firms. Zhang Qi-nan and others [14] used a matching method to compare the labor demand of robot adopters and non-adopters, and the results of the study showed that protecting employment by “restricting the adoption of robots” may fail due to the existence of external competition.

Compared with employment, the impact of AI on the development of enterprises is less studied, mainly to improve innovation performance and reduce pollution emissions, which is closely related to the ESG performance of enterprises. Shang Yuping [15] and others explored the impact of AI technology on enterprise pollution from the micro perspective, and the results show that AI technology can promote technological innovation, thus reducing enterprise emissions. Tian Hongna [16] and others empirically examined the impact of AI technology on the performance of green technology innovation of enterprises based on the theory of “stimulus-organism-response”, and Zhang Weike and Zeng Ming [17] indicated that the widespread adoption of AI can significantly reduce the energy intensity of enterprises.

Therefore, this paper proposes the hypothesis:

H1: 

Firms applying AI technology can enhance their ESG performance.

2.2. Corporate ESG Research

Since the establishment of the “dual-carbon” goal, the market requirements for green finance have become more stringent and the green constraints on corporate behavior have become more obvious. ESG is not only a passive evaluation criterion for enterprises but can also be used as a proactive framework for green transformation. In the context of the development of AI, the research on corporate ESG performance is still expanding based on the original richness.

The existing literature on ESG research is more extensive for the content of this paper, in which the study of ESG impact on the development of the enterprise itself is a prominent focus. Neha Malik and Smita Kashiramka [18], from regression panel data for the period of 2015–2021, found that the lender’s financial market in determining the credibility of the ESG disclosure will be taken into account, and the firms with better ESG performance also have better economic performance. Zhang Fan et al. [19] analyzed data from listed companies in China and found that ESG disclosure improves firms’ productivity and competitiveness. Yasser Eliwa et al. [20] used legitimacy theory and institutional theory to study a sample of 15 European Union countries, and the results showed that the stronger the ESG performance of the firms, the lower their cost of debt.

In addition to studies on the impact of ESG performance on firms’ direct earnings, some other studies have examined the effect of ESG on other natures. Scholars such as Yuanyuan Cheng and Mengjia Wang [21] also used data from listed companies for analysis. Their research found that when companies face higher external environmental pressure, ESG performance will have a greater negative impact on labor share. Wang Renxuan et al. [22] find that the stock market reaction to surplus announcements is more pronounced for firms with higher ESG ratings and that they are more attractive to long-term institutional investors. Empirical evidence from Wong Jin Boon and Zhang Qin [23] suggests that when ESG reputational risk is exacerbated, investors especially punish the stock price of firms with high cash share prices of firms with excessive levels, while Song Nie et al. [24] incorporate local government debt pressure, government behavior, and corporate ESG into the analytical framework and show that there is a negative correlation between debt pressure and firms’ ESG performance, which is particularly pronounced among firms with high political affiliation. Kong Xiangrong et al. [25] examined the nonlinear relationship between ESG and carbon emissions from the perspective of green credit, and they verified the negative moderating role of green credit in the relationship between ESG and carbon emissions by utilizing threshold regression and fixed effect models with panel data.

Therefore, this paper proposes the hypothesis:

H2: 

There is a positive moderating effect of digitalization degree on AI technology affecting corporate ESG performance.

2.3. Artificial Intelligence and Corporate ESG Performance

After combing through the existing studies, it is not difficult to find that while improving corporate ESG performance is in line with the requirements of China’s overall green transformation strategy, there are differences in the impact effects on the development of the companies themselves. Existing literature has delineated major categories and consolidated related crosscutting concepts for research on AI technologies in ESG as well as finance [26]. Wang Lang and Hou Sheng [27] use a combination of econometric modeling and qualitative comparative analysis to explore the relationship between digital transformation and firms’ ESG, and their study suggests that the uncertainty of digital transformation imposes some hidden costs on firms. Lan Lan and Zhifang Zhou [28] investigated the dual impact of digital orientation and corporate environmental, social, and governance (ESG) orientation on corporate innovation. Qi Yudong et al. [29] found that digital technology application and ESG performance have significant synergistic effects on firm performance with regional heterogeneity, while the study by Zhou Hailing and Liu Ji [30] shows a significant positive relationship between ICT and corporate ESG, and their research highlights the importance of energy efficiency in achieving ESG goals. This type of literature provides empirical evidence for the impact of enterprise digital transformation on ESG performance, but the mechanisms by which artificial intelligence affects ESG performance are rarely mentioned. It is only used as an explanatory factor in studies of employment and labor market changes. However, similar to the connection between enterprise ESG performance and sustainable development, empirical evidence for the relationship between artificial intelligence and improvements in enterprise green technology levels and production efficiency also exists. Therefore, reasonable speculation about the connection between artificial intelligence and enterprise ESG performance can be made.

Therefore, this paper proposes the hypothesis:

H3: 

The application of AI technology by enterprises mainly enhances the overall ESG performance by strengthening the environmental (E) and social (S) performance of enterprises.

3. Description of Data and Variables

3.1. Data Sources

To comprehensively reflect the effect of AI technology on corporate ESG performance, and to ensure the accuracy of the research results, this study adopts the corporate characteristics data from the Hexun Social Responsibility Report and merges it with the Chinese Research Data Services Platform (CNRDS), Huazheng, and Bloomberg ESG data, and selects a total of 5520 corporations in the years of 2007–2022 as the initial sample. Based on this, samples whose gearing ratio is greater than 1 or less than or equal to 0 are excluded, excluding ST and *ST enterprises, and finally obtaining panel data of 4858 enterprises for the years 2007–2022, with a total sample volume of 77,728. To exclude the interference of extreme values, this paper performs a 1% shrinkage before and after the non-comparison-type financial data variables involved in the paper before conducting the formal analysis.

3.2. Definition of Variables

3.2.1. Explained Variables

Corporate ESG scores (cnrds_esg). To comprehensively reflect the performance of corporate ESG, this paper focuses on China’s authoritative ESG evaluation system and adopts the ESG scores from the CNRDS database as the explanatory variables. Compared with other ESG sources in China, the design of the CNRDS ESG database is more complete and comprehensive. The measurement of ESG indicators not only traces the measurement characteristics of foreign ESG databases but also combines the actual situation of most enterprises in China, and through the statistics and summarization of the subdivided indicators, it can portray the specific performance of corporate ESG in many aspects. At the same time, the CNRDS database is also the most comprehensive ESG evaluation system of most of the existing ESG databases. ESG indicators are also the reference source of most existing ESG studies; therefore, the selection of CNRDS-ESG indicators in this paper can control the robustness of the results of this paper from the level of data sources, and this paper will standardize this indicator in the empirical model.

3.2.2. Explanatory Variables

Degree of Artificial Intelligence Adoption (lnAI). The extent to which an enterprise uses AI in its actual operations reflects the current level of AI in the enterprise and also determines, to a certain extent, the characteristics and structure of employment within the enterprise. Existing literature measures the degree of AI in enterprises mainly in terms of the number of industrial robots, the technological progress index, and patent authorization. However, examining the development prospects of China’s AI field, the application of AI technology by Chinese enterprises is still in the early stage; therefore, this paper adopts the ratio of the book value of robots to the number of employees as a measure of the degree of adoption of AI technology by enterprises, and logarithmic processing is carried out for specific applications.

3.2.3. Control Variables

Referring to the existing relevant literature related to the evaluation of corporate governance and social responsibility, this study selects the following variables in the empirical regression model to control, asset size (ZCGM): Total assets of the enterprise, profitability (YLNL): Return on net assets of the enterprise, enterprise age (CAGE): Years of establishment of the enterprise, investment value(TZJZ): Operating income compared to the previous year, capital structure (ZBJG): Cash ratio, financial risk (CWFX): Asset-liability ratio, decision-making governance (JCZL): Proportion of independent directors, equity concentration (GQJZ): The proportion of the largest shareholder, responsible supervision (ZRJD): Chairman andThe two positions of general manager are integrated. In specific applications, some variables are also logarithmized.

3.3. Descriptive Statistics

The descriptive statistics of the main variables in this paper are reported in

Table 1. Descriptive statistics for variables.

Variable	Mean	Standard Deviation	Minimum	Maximum
cnrds_esg	25.482	11.373	0.017	79.322
lnAI	12.982	1.137	3.785	22.195
lnZCGM	21.421	1.609	13.239	29.535
lnYLNL	−2.511	0.914	−7.824	0.735
lnCAGE	1.951	0.928	0	3.466
lnTZJZ	0.196	0.467	−4.102	7.955
lnZBJG	−0.534	1.131	−7.264	5.121
lnCWFX	−1.049	0.633	−4.948	1.257
lnJCZL	−0.985	0.164	−2.303	−0.223
lnGQJZ	3.445	0.477	0.888	4.497
ZRJD	0.288	0.453	0	1

. cnrds_esg has a mean of 25.482 and a variance of 11.373, and hz_esg has a mean of 4.24 and a variance of 0.985, which shows that the mean values of the ESG performance of firms, whether evaluated with the ESG indicator system of the CNRDS database or the CSI database, are at a lower level, which indicates that the overall ESG performance of Chinese-listed companies is on the low side, and there are also large differences in ESG performance between companies.

4. Empirical Analysis

4.1. Modeling

To examine the impact of the adoption of AI technology on the ESG performance of Chinese-listed firms, this paper constructs the following econometric model:

$${cnrds\_esg}_{i,t}={\alpha }_0+{\beta }_1{\ln AI}_{i,t}+{\displaystyle {\sum }_j{\beta }_j}Controls+Fixed\_Effect+{\epsilon }_{i,t}$$

(1)

The above model (1) is the model with the CNRDS database ESG as the explanatory variable, where $i$ represents different firms, $t$ represents different years, $Controls$ denotes the control variables in the model, and $\epsilon$ is the random error term. $\ln AI$ is the core explanatory variable in this paper. β₁ is the coefficient of interest in this paper, which measures the change in firms’ ESG performance with a 1% increase in Artificial Intelligence adoption. To reduce the endogeneity problem caused by the interaction between enterprises, this paper chooses to control for fixed effects in the following ways: ① control for industry + time fixed effects; ② control for industry × time fixed effects.

4.2. Benchmark Regression Results

When applying the econometric model to examine the impact relationship between variables, the type of fixed effects and the standard error control hierarchy will affect the regression coefficients as well as the standard errors themselves to varying degrees. Given this, this paper examines different combinations of the three factors of the control variables, fixed effects, and standard error clustering hierarchies, respectively, and seeks to report the baseline empirical results under the stricter fixed effects and standard error clustering hierarchies. Specific model details are reported in

Table 2. Benchmark regression results.

Variable	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
	cnrds_esg	cnrds_esg	cnrds_esg	cnrds_esg	cnrds_esg	cnrds_esg	cnrds_esg	cnrds_esg
	Controlling for Different Fixed Effects				Standard Error Clustering to Different Levels
lnAI	0.088 *** (0.005)	0.089 *** (0.005)	0.05 *** (0.008)	0.052 *** (0.008)	0.052 *** (0.008)	0.052 *** (0.008)	0.052 *** (0.008)	0.052 *** (0.008)
Controls	NO	NO	YES	YES	YES	YES	YES	YES
cons	YES	YES	YES	YES	YES	YES	YES	YES
Industry FE	YES	NO	YES	NO	NO	NO	NO	NO
Year FE	YES	NO	YES	NO	NO	NO	NO	NO
Industry × Year FE	NO	YES	NO	YES	YES	YES	YES	YES
Clustering to Enterprise	NO	NO	NO	NO	YES	NO	NO	NO
Clustering to Industry	NO	NO	NO	NO	NO	Yes	NO	NO
Clustering to Firm × Year	NO	NO	NO	NO	NO	NO	YES	NO
Clustering to Industry × Year	NO	NO	NO	NO	NO	NO	NO	YES
R2	0.385	0.425	0.448	0.484	0.484	0.484	0.484	0.484
obs	24,506	24,506	15,454	15,454	15,454	15,454	15,454	15,454

Note: *** represents regression coefficients that are significant at the 1% significance levels.

, with columns (1)–(4) focusing on examining the impact of different fixed effects. Where columns (1)–(2) examine the impact of different fixed effects on the findings without adding any other control variables, columns (3)–(4) examine the findings under the two types of fixed effects with all the above-mentioned control variables added. Columns (5)–(8) examine the impact of different standard error clustering levels, all adding the mentioned control variables, and the fixed effects are controlled for industry × time, whose standard errors are clustered to enterprise, industry, enterprise × year, industry × year, respectively. From the results in Table 2, it is easy to see that the degree of AI adoption presents a significant positive relationship with the ESG performance of enterprises, i.e., the application of AI technology by enterprises can significantly improve the comprehensive performance of enterprises in terms of environment, society, and governance. This conclusion is very robust, verifying that the research hypothesis H1 of this paper is valid.

4.3. Endogeneity Test

Endogeneity refers to the existence of a correlation between the explanatory variables and the error term in the model, which may further lead to the existence of bias in the model, thus reducing the explanatory power of the model. The benchmark regression shows a significant positive correlation between firms’ application of AI technology and firms’ ESG performance, but this result may have an endogeneity problem. There are three main sources of endogeneity: sample self-selection, omitted variables, and reverse causality, given which this paper will take different approaches to deal with each of these three cases.

4.3.1. Sample Self-Selection

The mechanism of self-selection bias lies in the non-random selection of explanatory variables. In this paper, it is reflected in the fact that enterprises with better ESG performance are more likely to apply artificial intelligence technology. This not only meets the original ESG performance requirements of the enterprise but also enables the enterprise to improve efficiency and gain innovation advantages through artificial intelligence technology. If this is the case, then it is not the application of artificial intelligence technology that has improved the enterprise’s ESG performance. The above analysis indicates that this paper may have a self-selection sample issue. Based on this, this paper uses the Heckman two-stage regression model to correct it.

Enterprises, as business organizations in the universal sense, have profitability as their main purpose, so they will decide whether to apply AI technology or not out of consideration for their development, and such considerations mainly involve their innovative concepts as well as their technological level. In this paper, we will characterize this with the following variables: R&D expenditure (YFZC), technology philosophy (JSLN), innovation score (CXDF), and number of R&D personnel (YFNum). Specifically, the explanatory variable in the first-stage regression is whether the firm applies artificial intelligence technology (AI_dummy), and the probit model is used to test the correlation between it and the above variables. Based on this, the corresponding Inverse Mills Ratio (IMR) is computed and the indicator is used to observe whether these variables affect the firm’s AI decisions. The ratio is subsequently used as an additional control variable to the benchmark regression as a way to correct the endogeneity problem caused by self-selection bias.

The specific results of Heckman’s two-stage regression are reported in

Table 3. Heckman’s two-stage regression.

Variable	(1) AI_dummy	(2) cnrds_esg
	Phase I	Phase II
lnAI	NO	0.068 *** (0.003)
YFZC	2.534 *** (0.275)	NO
JSLN	1.192 *** (0.046)	NO
CXDF	−1.023 *** (0.033)	NO
YFNum	0.001 *** (0)	NO
Controls	YES	YES
cons	YES	YES
IMR	NO	0.036 ** (0.019)
obs	19,093	19,093
R2	NO	0.086

Note: ** and *** represent regression coefficients that are significant at the 5% and 1% significance levels, respectively.

, where it can be found that the Inverse Mills Ratio (IMR) is significant in the regression at the 5% significance level, which suggests that the results of this paper’s benchmark regression do suffer from endogeneity due to sample self-selection; therefore, controlling for it through Heckman’s two-stage regression is necessary. It can be found that in the second stage of the regression, the coefficient of the degree of adoption of artificial intelligence is 0.068 and highly significant, which indicates that the results of this benchmark regression are still robust after controlling for self-selection bias.

4.3.2. Omitted Variables and Reverse Causation

Given the characteristics of the micro dataset used in this paper, there may be unobservable factors that can affect both firms’ innovation decisions and firms’ ESG performance, i.e., the omitted variable problem. The findings of this paper are intended to illustrate the enhancement of firms’ application of AI technology on ESG performance and the mechanism of its effect, but the firms with better ESG performance may also tend to apply AI technology, i.e., the reverse causation problem. However, firms with better ESG performance may also tend to apply AI technology, i.e., the improvement of ESG performance, which will, in turn, act on firms’ AI application, i.e., the reverse causality problem. This will also cause bias in the regression results. Considering that there is a certain time lag in the application of AI technology by enterprises to promote the improvement of their ESG performance, the instrumental variable method and the method of adding 1–3 period lagged terms are chosen to correct the omitted variable and reverse causation problems.

In the selection of instrumental variables, regarding existing research studies, the penetration of industrial robots in the United States (IV1) and the mean value of AI technology application in industries other than current enterprises (IV2) are selected as instrumental variables. First of all, in the United States, as the head country in the field of artificial intelligence, the overall development trend will have a guiding effect on the same industry in other countries, but the robot penetration degree of its local enterprises will not directly affect the ESG development of Chinese enterprises, so it meets the requirements of relevance and exclusivity; the degree of enterprise informatization and digitalization development is related to the characteristics of the industry in which it is located, and there are similarities between the application of AI technology and the digitalization of enterprises, which has been validated. There is a similarity between the application of artificial intelligence technology and enterprise digitalization, and it has been verified that the indicator of the degree of adoption of artificial intelligence is also in line with the reality that the information industry and high-end manufacturing industry are higher than the food and other industries, so the relevance requirement of this indicator is satisfied. By the same token, the average level of the application of artificial intelligence of the other enterprises in the industry will not directly affect the current ESG development of the enterprise, i.e., it meets the exclusivity requirement. In summary, the two variables of the penetration of industrial robots in the United States and the average application level of AI technology in the industry outside the current enterprise meet the requirements of instrumental variables.

The instrumental variable regression results are shown in

Table 4. Omitted variables and reverse causality tests (1).

Variable	(1)	(2)	(3)	(4)
	Phase I		Phase II
	lnAI		cnrds_esg
IV1	0.028 *** (0.006)	0.028 *** (0.007)
IV2	0.001 ** (0)	0.001 ** (0)
lnAI			0.121 ** (0.058)	0.114 ** (0.053)
Controls	YES	YES	YES	YES
Industry FE	YES	NO	YES	NO
Year FE	YES	NO	YES	NO
Industry × Year FE	NO	YES	NO	YES
cons	YES	YES	YES	YES
R2	0.602	0.625	0.407	0.472
Weak identification test: H0: Weakly identified
F-value			18.011	20.342
Overidentification test: H0: instruments are valid instruments
p-value			0.856	0.766
obs	12,463	12,463	13,729	13,729

Note: ** and *** represent regression coefficients that are significant at the 5% and 1% significance levels, respectively.

, and the regression results with lagged terms added are shown in

Table 5. Omitted variables and reverse causality tests (2).

Variable	(1)	(2)	(3)	(4)	(5)	(6)
	cnrds_esg	cnrds_esg	cnrds_esg	cnrds_esg	cnrds_esg	cnrds_esg
L1.lnAI	0.046 *** (0.008)	0.049 *** (0.008)
L2.lnAI			0.039 *** (0.008)	0.041 *** (0.008)
L3.lnAI					0.038 *** (0.009)	0.036 *** (0.009)
Controls	YES	YES	YES	YES	YES	YES
cons	YES	YES	YES	YES	YES	YES
Industry FE	YES	NO	YES	NO	YES	NO
Year FE	YES	NO	YES	NO	YES	NO
Industry × Year FE	NO	YES	NO	YES	NO	YES
R2	0.451	0.488	0.446	0.491	0.437	0.49
obs	14,032	14,032	12,261	12,261	10,502	10,502

Note: *** represents regression coefficients that are significant at the 1% significance levels.

. The first-stage results in Table 4 show that the instrumental variables are significant, at least at the 5% significance level. The F-value in the second-stage regression results is much larger than 10, so the hypothesis of weak instrumental variables can be rejected. The regression results also show that there is no over-identification problem, i.e., the statistical test of the validity of the instrumental variables passes in all cases. The estimated coefficients of the impact of firms’ application of AI technology on firms’ ESG performance in the second-stage regression results are still significantly positive, and the regression results in Table 5 are consistent with the benchmark regression, which suggests that the results of the benchmark regression are still reliable after this paper controls for the potential omitted variables and the endogeneity problem caused by reverse causation.

4.4. Robustness Tests

After the endogeneity test described above, this paper performs the following robustness test for the benchmark regression results.

4.4.1. Replacement of ESG Performance Measures

This paper uses several existing ESG indicators that are more commonly used and widely recognized by the academic community to replace the measures used in the main regressions of this paper for robustness testing, and the results are shown in

Table 6. Replacement of ESG indicator test results.

Variable	(1)	(2)	(3)	(4)
	hz_esg	hz_esg	pb_esg	pb_esg
lnAI	0.021 ** (0.01)	0.022 ** (0.01)	0.037 ** (0.017)	0.036 ** (0.018)
Controls	YES	YES	YES	YES
cons	YES	YES	YES	YES
Industry FE	YES	NO	YES	NO
Year FE	YES	NO	YES	NO
Industry × Year FE	NO	YES	NO	YES
R2	0.175	0.202	0.34	0.367
obs	15,454	15,454	5124	5124

Note: ** indicates that the coefficient is significant at the 5% significance level.

.

4.4.2. Consideration of the United Nations Sustainable Stock Exchanges Initiative and the Inclusion of Companies in MSCI Indices

The United Nations Sustainable Stock Exchanges Initiative, which the Shanghai and Shenzhen exchanges successively joined in 2017, may make the ESG performance of enterprises improve in 2018, and the MSCI indicators, which enterprises included in 2018, may likewise have an impact on the ESG performance of enterprises. These two types of initiatives may lead to the previous results only reflecting the characteristics of the changes in 2018 while masking the whole sample; therefore, to verify the robustness of the benchmark regression results, this paper excludes the samples of 2018 and later years and re-performs the benchmark regression, and the results are shown in

Table 7. Excluded sample-control firm nature test results.

Variable	(1)	(2)	(3)	(4)	(5)	(6)
	cnrds_esg	cnrds_esg	hz_esg	hz_esg	pb_esg	pb_esg
Consideration of the United Nations Sustainable Stock Exchanges Initiative and the inclusion of companies in the MSCI indices
lnAI	0.062 *** (0.012)	0.064 *** (0.012)	0.06 *** (0.014)	0.062 *** (0.014)	0.067 *** (0.024)	0.062 ** (0.025)
R2	0.404	0.438	0.132	0.161	0.315	0.344
obs	8175	8175	8175	8175	2871	2871
Controlling for the fixed effects of the nature of the firm
lnAI	0.05 *** (0.009)	0.046 *** (0.009)	0.36 *** (0.012)	0.035 *** (0.012)	0.072 *** (0.026)	0.067 ** (0.028)
R2	0.462	0.499	0.173	0.198	0.218	0.252
obs	11,064	11,064	11,064	11,064	2753	2753
Controls	YES	YES	YES	YES	YES	YES
cons	YES	YES	YES	YES	YES	YES
Industry FE	YES	NO	YES	NO	YES	NO
Year FE	YES	NO	YES	NO	YES	NO
Industry × Year FE	NO	YES	NO	YES	NO	YES

Note: ** and *** represent regression coefficients that are significant at the 5% and 1% significance levels, respectively.

. The results indicate that after excluding this effect, the promotion effect of enterprise application of AI technology on the ESG performance of enterprises is still significant, i.e., it indicates that the benchmark regression results are robust.

4.4.3. Fixed Effects Controlling for the Nature of the Enterprise

Different ownership structures of enterprises may also lead to differences in regression results. Compared to state-owned enterprises, private enterprises have more autonomy and flexibility in decision-making and production processes. Therefore, this article extracts a private enterprise sub-sample for regression analysis, and the regression results in Table 7 show that the conclusions of this article still hold.

4.5. Mechanism Testing

Previous research has shown that through benchmark regression, endogeneity treatment, and various robustness tests, this paper reaches a robust core conclusion: the application of artificial intelligence technology can significantly improve a com-pany’s ESG performance. This part provides evidence of a causal relationship between the two based on empirical testing. However, the internal mechanisms by which the application of artificial intelligence affects ESG performance under the context of corporate digital transformation are still unclear. What is the specific connection between the overall improvement in a company’s ESG level and its various components? This paper will address this issue in detail in this section.

4.5.1. Degree of Digitalization

When enterprises choose to apply artificial intelligence technology, due to policies being at different stages, enterprises may have already undergone a certain degree of digital transformation. Therefore, enterprises will be influenced by both the application of artificial intelligence technology and digital transformation in improving their ESG performance. Benchmark regressions have already shown that the application of artificial intelligence technology has a significant promoting effect on ESG performance. So, what is the mechanism of digital transformation in this promoting effect? This is worth further discussion.

In order to verify the mechanism role of digital transformation in the process of AI technology affecting firms’ ESG performance, this paper uses the word frequency data of firms’ digital transformation from the China Securities Market and Accounting Research (CSMAR) database to construct firms’ digital transformation indicators. It is constructed according to the following process: firstly, we collect and summarize the annual reports of all the enterprise samples involved in this paper’s benchmark regression through Python crawler technology and collate all the characteristic words about ‘digital transformation’ in these annual reports through the text extraction function. In order to ensure the accuracy of the analysis, this paper excludes the feature words with negative meanings such as ‘no’, ‘non’, etc. Finally, the word frequencies of the enterprise digital transformation feature words are statistically summed to get the total word frequencies, and then the total word frequencies are logarithmically processed to get the enterprise digital transformation degree index used in the analysis of this paper.

Table 8. Characteristic terms for the degree of digital transformation of an enterprise.

Classification	Characteristic Words
Big Data Technology	Big Data, Data Mining, Text Mining, Data Visualization, Heterogeneous Data, Augmented Reality, Mixed Reality, Virtual Reality
Cloud Computing Technology	Cloud Computing, Streaming Computing, Graph Computing, In-Memory Computing, Multi-Party Secure Computing, Brain-Like Computing, Green Computing, Cognitive Computing, Converged Architecture, Billion-Class Concurrency, EB-Level Storage, Internet of Things, Information Physical Systems
Blockchain Technology	Blockchain, Digital Currency, Distributed Computing, Differential Privacy Technology, Smart Financial Contracts

below shows all the feature words. After obtaining the digital transformation degree index, this paper adds the digital transformation variables and their interaction terms with AI technology variables on the basis of the baseline regression model to obtain the model as shown in (2), where the moderator represents the degree of enterprise digitalization. Specific regression results are shown in

Table 9. Mechanism analysis test results.

Variable	(1)	(2)
	cnrds_esg	cnrds_esg
lnAI	0.049 *** (0.009)	0.049 *** (0.009)
Moderator	−0.033 ** (0.017)	−0.037 ** (0.017)
lnAI × Moderator	0.003 ** (0.001)	0.003 ** (0.001)
R2	0.419	0.453
obs	17,723	17,723
Controls	YES	YES
cons	YES	YES
Industry FE	YES	NO
Year FE	YES	NO
Industry × Year FE	NO	YES

Note: ** and *** represent regression coefficients that are significant at the 5% and 1% significance levels, respectively.

, which shows that the coefficient of the interaction term between enterprise digital transformation variables and AI technology variables is significantly positive, i.e., in the process of AI influencing enterprise ESG, the degree of enterprise digitalization plays a positive moderating role, and the hypothesis H2 of this paper is established.

$$\begin{array}{c}{cnrds\_esg}_{i,t}={\alpha }_0+{\beta }_1\ln {AI}_{i,t}+{\displaystyle {\sum }_j{\beta }_j}Controls+\\ {\beta }_2{Moderator}_{i,t}+{\beta }_3{Moderator}_{i,t}\times \ln {AI}_{i,t}+Fixed\_Effect+{\epsilon }_{i,t}\end{array}$$

(2)

4.5.2. Impact on ESG Sub-Component

The benchmark regression in this paper has shown that the application of AI technology can improve the overall ESG performance of enterprises, and this paper will further explore how this application process can affect the overall ESG evaluation of enterprises by influencing their performance in the environment (E), society (S), and governance (G). In this paper, we also refer to the evaluation indexes for corporate ESG sub-items in the CNRDS database and replace the total ESG indexes in the benchmark regression with the three sub-item indexes of E, S, and G, respectively. The results of applying AI technology to influence each sub-item are shown in columns (1), (3), and (5) in

Table 10. Channel analysis test results.

Variable	(1)	(2)	(3)	(4)	(3)	(6)
	cnrds_E	GTI	cnrds_S	CP	cnrds_G	EIC
lnAI	1.289 *** (0.117)	0.022 ** (0.009)	0.361 *** (0.105)	0.077 *** (0.01)	−0.394 *** (0.08)	−5.281 *** (1.166)
R2	0.267	0.309	0.211	0.281	0.391	0.117
obs	17,723	17,723	17,723	17,626	17,723	17,723
Controls	YES	YES	YES	YES	YES	YES
cons	YES	YES	YES	YES	YES	YES
Industry FE	YES	YES	YES	YES	YES	YES
Year FE	YES	YES	YES	YES	YES	YES

Note: ** and *** represent regression coefficients that are significant at the 5% and 1% significance levels, respectively. Limited to space reasons, only the fixed enterprise + year test results are shown; enterprise × year test results do not change the conclusions of this paper.

. The results show that the environmental (E) and social (S) indicators are the main bias for enterprises to improve ESG performance, and the application of AI by enterprises is to improve the overall ESG performance by improving the performance of both the environment and society, i.e., the research hypothesis H4 of this paper is valid. The negative coefficient in column (5) indicates that there is a threat to corporate governance from corporate AI technology applications.

Further, the channels through which AI technology affects the performance of each sub-component are examined. For the environmental (E) subitem, some studies indicate that technological progress will have an impact on corporate pollution, so this paper selects the green technology innovation indicator (GTI) as a proxy variable to examine whether AI technology can promote corporate green technology innovation and thus enhance its environmental (E) performance; similarly, for the social (S) sub-item, corporate social responsibility is closely related to its philanthropic responsibility, so this paper takes the logarithmic value of the donation amount (CP) as a proxy variable; for the governance (G) sub-component, the introduction of an AI chatbot by an enterprise that autonomously generated erroneous information leading to inappropriate behavior on the part of the user and thus increased the risk of a data breach had a knock-on effect on the effectiveness of the enterprise’s internal controls. Furthermore, in addition to real-life cases, there are studies that show that machine learning-led AI technologies play a crucial role in advancing the healthcare and biomedical aspects of healthcare, which also means that applications designed using these technologies require a high level of security, as some machine learning algorithms are able to amplify the corrupting effect that occurs in their training datasets through the use of malicious data, which can lead to security breaches that can hinder diagnosis and thus endanger the lives of patients [31,32]. This situation will undoubtedly create conflicts within the companies involved in the healthcare industry, leading to a decrease in the effectiveness of their internal controls, while other studies have shown that the effectiveness of internal controls (EIC) affects the governance behaviors of firms by, for example, influencing the credit ratings of corporate bonds. Therefore, this paper adopts the Dibble internal control index to examine whether AI technology can influence governance (G) scores by affecting the effectiveness of corporate internal control. The three types of channels analyzed above are tested. The test results are shown in columns (2), (4), and (6) in Table 10, in which the coefficients of columns (2) and (4) are significantly positive, indicating that AI technology can enhance the performance of enterprises in the environmental (E) sub-component by strengthening the level of corporate green technology innovation, and can enhance the performance of enterprises in the social (S) sub-component by improving corporate philanthropic responsibility. The coefficient of column (6) is significantly negative, which verifies the analysis results in column (5) and indicates that AI technology weakens the effectiveness of corporate internal control, which threatens and negatively affects corporate governance and ultimately leads to a decrease in the score of the governance (G) sub-component.

5. Further Analysis

After the previous empirical evidence and tests, this paper has obtained the robust and reliable core conclusion that the application of AI technology has a significant contribution to the ESG performance of Chinese firms, and there is a mediating role between technological progress and philanthropic responsibility. In this subsection, this paper focuses on the heterogeneity of the application of AI technology on the ESG performance of various types of firms.

5.1. Heterogeneity of Industry Competition

The effect of artificial intelligence technology applications may vary depending on the competitive environment in which the enterprise is located. Market competition can stimulate the motivation of enterprises to apply technology, and AI technology itself has innovative attributes that can develop enterprises’ information technology and big data application capabilities to a certain extent to enhance their productivity and optimize their ESG performance. For example, manufacturing companies usually use AI-driven intelligent production systems to predict equipment downtime and improve production efficiency, which makes it possible to reduce production costs while reducing the negative impact on the environment; at the same time, highly competitive industries are usually at the forefront of technological innovation, and companies will be more active in researching and developing innovative technologies such as AI in order to stand out in the market; in contrast, companies in industries with a low degree of competition face less market pressure and lack the incentive to apply AI technologies on a large scale, which also makes the performance of the two types of companies in the three dimensions of E, S, and G differ. Therefore, it can be expected that, compared with industries with low competition levels, AI technology will have a more significant role in promoting the ESG performance of enterprises in highly competitive industries. Accordingly, this paper measures the level of competition in the industry by the inverse value of the standard deviation of the profitability of the main business margin of all sample industries within the secondary industries classified by the SEC and defines the industry with a level of competition greater than the median as highly competitive; otherwise, it is a low-competition industry and carries out a group regression accordingly. The test results are shown in columns (1)–(2) of

Table 11. Industry competition/firm-intensive heterogeneity.

Variable	(1)	(2)	(3)	(4)	(5)
	High Competitiveness	Low Competitiveness	Labor-Intensive	Capital-Intensive	Technology-Intensive
lnAI	0.077 *** (0.021)	0.042 *** (0.018)	0.068 ** (0.031)	−0.039 * (0.022)	0.083 ** (0.036)
R2	0.474	0.413	0.44	0.492	0.342
obs	6896	9280	5843	4165	6168
Controls	YES	YES	YES	YES	YES
cons	YES	YES	YES	YES	YES
Industry FE	YES	YES	YES	YES	YES
Year FE	YES	YES	YES	YES	YES

Note: *, **, and *** represent regression coefficients that are significant at the 10%, 5%, and 1% significance levels, respectively. Limited to space reasons, only the fixed enterprise + year test results are shown; enterprise × year test results do not change the conclusions of this paper.

, in which the coefficient of the AI technology variable in the sample of high-competition industries is 0.07, while the coefficient in the sample of low-competition industries is only 0.042 so that the facilitating effect of AI technology within the high-competition industries is more significant.

5.2. Intensive Enterprise Heterogeneity

Is there heterogeneity in the impact of AI technology on firms’ ESG across different intensive firms? Certain industries have a large number of employees but relatively low technological requirements, while certain industries have a greater need for capital investment and a higher proportion of machinery and equipment. Do these differences lead to differences in the impact of AI technology? This paper divides the industry in which the enterprise is located into labor-intensive, capital-intensive, and technology-intensive (Labor-intensive industries include 19 industries, such as warehousing and animal husbandry; capital-intensive industries include 18 industries, such as electrical machinery and real estate; and technology-intensive industries include 12 industries, such as public facilities management, automobile manufacturing, and special equipment manufacturing.), and conducts the regression respectivelyThe results are shown in columns (3)–(5) of Table 11, which show that the coefficient of AI indicators in technology-intensive enterprises is the most significant, i.e., the promotion effect of AI technology is more significant in technology-intensive enterprises. This is because technology-intensive firms usually rely on advanced technological equipment and more complex production processes and also have a higher proportion of knowledge-based employees who are able to embed AI technology more deeply into the production system and are more capable of applying AI technology. In contrast, labor-intensive enterprises have simpler production processes and rely more on human resources, and therefore cannot maximize the use of AI technology. This makes it more significant for technology-intensive firms to use AI technology to improve ESG performance.

5.3. Regional Heterogeneity

China has a relatively vast land area; thus, the economic development, industrial structure, population size, and other conditions in various regions are quite different. Different regional characteristics make the level of application of AI technology also have a certain degree of heterogeneity. The Central Leading Group for Finance and Economics considered and studied the Beijing-Tianjin-Hebei Synergistic Development Plan Outline in February 2015, and the Political Bureau of the CPC Central Committee considered and passed the Yangtze River Economic Belt Development Plan Outline in March 2016, which promoted the economic development of China’s eastern and central regions while encouraging enterprises to adopt advanced technologies to enhance competitiveness and sustainability, creating a favorable environment for enterprises to apply AI to improve their ESG performance, while the western and northeastern regions have corresponding supportive policies. However, the effect of policy incentives in promoting the integration of AI and corporate ESG is relatively weak. Therefore, to further examine the impact of the application of AI technology on the ESG performance of enterprises, this paper will be based on the regional division of enterprises into the eastern, central, western, and northeastern four areas (The eastern region includes 10 provinces and municipalities, including Beijing, Tianjin, Hebei, Shanghai, Jiangsu, Zhejiang, Fujian, Shandong, Guangdong, and Hainan; the central region includes 6 provinces, including Shanxi, Anhui, Jiangxi, Henan, Hubei, and Hunan; the western region includes 11 provinces, including Inner Mongolia, Guangxi, Chongqing, Sichuan, Guizhou, Yunnan, Shaanxi, Gansu, Qinghai, Ningxia, and Xinjiang; and the northeastern region includes 3 provinces, including Liaoning, Jilin, and Heilongjiang. The northeastern region includes 3 provinces: Liaoning, Jilin, and Heilongjiang.), conducting group regression, and the regression results are shown in (1)–(4) of

Table 12. Regional/scale heterogeneity.

Variable	(1)	(2)	(3)	(4)	(5)	(6)	(7)
	East	Central	Western	North-Eastern	Large	Medium	Small
lnAI	0.048 *** (0.009)	0.104 *** (0.026)	0.03 (0.024)	−0.004 (0.056)	0.038 ** (0.015)	0.061 *** (0.015)	0.001 (0.023)
R2	0.453	0.516	0.491	0.551	0.398	0.484	0.51
obs	10,273	6586	5843	1804	8168	8169	8169
Controls	YES	YES	YES	YES	YES	YES	YES
cons	YES	YES	YES	YES	YES	YES	YES
Industry FE	YES	YES	YES	YES	YES	YES	YES
Year FE	YES	YES	YES	YES	YES	YES	YES

Note: ** and *** represent regression coefficients that are significant at the 5% and 1% significance levels, respectively. Limited to space reasons, only the fixed enterprise + year test results are shown; enterprise × year test results do not change the conclusions of this paper.

.

As can be seen from Table 12, the study’s sample mainly focuses on the eastern and western regions, where the promotion of artificial intelligence technology to corporate ESG performance is notably significant, but for the western and northeastern regions, the effect is not significant, showing obvious heterogeneity. The reason for this phenomenon may be that the eastern region is more economically developed, its technology intensity and capital intensity are far from being comparable to other regions, its competition is more intense, and enterprises have more urgent needs for the application of AI technology and therefore must invest more in the development of AI, which leads to the enhancement of the ESG performance of enterprises. The western and northeastern regions, on the other hand, are more slowly developed, and their demand for the application of AI technology is also more urgent. In contrast, the West and Northeast are slower to develop and have a lower demand for AI technology, so they are less responsive to the impact on corporate ESG performance.

5.4. Size Heterogeneity

For enterprises with different scales, their resource acquisition ability, information advantage, and technological strength also have a large gap, resulting in heterogeneity in the degree of their application of AI technology. To study whether the impact of the application of AI technology on the ESG of the enterprise will be due to the differences in the size of the enterprise, this paper categorizes them into large, medium, and small enterprises. The regression results are shown in columns (5)–(7) of Table 12.

From the results in Table 12, it is easy to see that the promotion effect of the application of AI technology on the ESG performance of enterprises is very significant in both large- and medium-sized enterprises, but not in small-sized enterprises, which also shows heterogeneity, which may be due to the fact that larger enterprises have a stronger competitive advantage as well as financial support, and the effectiveness of their application of AI technology is more effective compared to that of small enterprises. At the same time, it can also be found that the effect of medium-sized enterprises applying artificial intelligence technology to improve their ESG performance is more significant than that of large enterprises, this deserves a separate discussion. Specifically, medium-sized firms have a simpler organizational structure compared to large firms, which makes them more flexible in their use of AI technology. At the same time, medium-sized enterprises have a certain scale, but are not yet bloated, so their resource allocation is more focused, and they do not need to formulate a very detailed resource allocation plan, which large-scale enterprises do, which makes it easier for medium-sized enterprises to focus on investing their limited resources in AI projects that are critical to the development of the enterprise to achieve precision and speed in the decision-making process. For example, the smaller size of a medium-sized technology company allowed it to spend most of its funds on developing an AI-based customer service system, which, through intelligent analysis of customer needs and rapid response, has improved customer satisfaction and enhanced its image performance at the social (S) level. Another aspect that is easily overlooked is that medium-sized firms also tend to face greater competitive pressures in the marketplace, which, as analyzed earlier, prompts them to make more active use of AI technologies to improve their competitiveness and sustainability, and in the process of doing so, they tend to improve their ESG performance as well.

6. Conclusions and Recommendations

With the increasingly stringent requirements of sustainable development at the enterprise level, and against the backdrop of the current deepening of artificial intelligence in China, the application of AI technology by enterprises has gradually become a large-scale trend. Digital technology is constantly innovating, and the integration of this trend with the strategic development of enterprises will become more and more profound. The fact that AI technology enhances the economic benefits of enterprises has become a consensus in the academic world. To explore the “non-economic benefits” brought about by the application of AI technology by enterprises, this paper matches the ESG indexes of enterprises from CNRDS, Huazheng, and Bloomberg databases, and finally selects the listed enterprises in China from 2007–2022 as the sample of the study. In this paper, we consider constructing a theoretical model of the corporate application of AI technology and ESG performance, empirically test the research hypotheses involved in the theoretical model, and provide substantial evidence that the corporate application of AI technology affects ESG performance.

The findings show that the application of AI technology has a positive impact on corporate ESG performance, and the findings are still reliable after a series of endogeneity and robustness tests. The channel test finds that the improvement of corporate ESG performance is mainly due to the improvement of corporate environmental (E) and social (S) performance in the process of applying AI technology and that green technological innovation and corporate responsibility are the main factors for the improvement of environmental (E) and social (S) performance, respectively. In addition, AI technology also weakens the effectiveness of corporate internal control, which threatens corporate governance and ultimately leads to a reduction in corporate governance (G) performance. The mechanism analysis shows that the degree of enterprise digitalization can play a positive moderating role in the process of enterprises applying AI technology to enhance ESG performance. Subsequent in-depth analysis shows that the facilitation effect of AI technology is more significant in more competitive industries and technology-intensive firms; firms in the east and center of the country have a more significant facilitation effect of AI in promoting ESG performance relative to firms in the west and northeast of the country, and large firms and medium-sized firms are similarly better than small firms. Further, it can be noted that the facilitation effect of medium-sized firms is greater than that of large firms since medium-sized enterprises have more room for self-improvement.

Based on the above findings, this paper draws the following insights:

Currently, we are in the critical period of realizing the goal of “dual-carbon”, enterprises and governments at all levels should pay enough attention to “green” and “sustainability”, and the application of AI technology can optimize the internal structure and improve energy efficiency, thus enhancing the ESG performance of enterprises. On the one hand, enterprises should be encouraged to apply AI technology to improve their “green finance” level, and on the other hand, the core concept of the ESG evaluation system should be advocated to guide enterprises to utilize AI to enhance their ESG performance and increase their investment in ESG practice. At the same time, the current relevant research mostly provides evidence from the macro level, and there is a large gap for micro data of a smaller scale. Open research studies on the internal operation or management level of various enterprises using questionnaires or other effective ways can not only help enterprises to check and fill in their own development deficiencies but also provide relevant scholars with real and reliable research samples, making important contributions to the development of AI applications and ESG integration.