Artificial Intelligence and Enterprise Green Innovation: Evidence from a Quasi-Natural Experiment in China

Abstract

Against the backdrop of addressing global climate change, whether the new generation of information technology, centered on artificial intelligence (AI), can promote comprehensive green transformation and achieve the “dual carbon” goal has become an important issue in China’s national development strategy. The research objective of this paper is to explore the causal relationship between AI and green innovation (GI). In this study, we conduct a quasi-natural experiment using the National New Generation Artificial Intelligence Innovation and Development Pilot Zone (NAIPZ). On the basis of data from A-share-listed companies from 2013 to 2022, we use a staggered difference-in-difference model to study the impact and mechanism of AI on corporate GI. Research results show that AI can improve the GI of enterprises. Mechanism analysis results show that AI promotes GI in enterprises by improving internal governance and optimizing human capital, while industry competition can increase the promotion effect of AI on GI. Heterogeneity analysis results indicate that the promotion effect of AI on GI is particularly prominent in the eastern region, high-tech industries, and non-state-owned enterprises. This study addresses the important question of whether the NAIPZ can promote GI in enterprises, thereby providing empirical evidence and policy references for accelerating the integration and development of AI and GI in China.

1. Introduction

Faced with the challenge of climate change, the global community is seeking to achieve carbon neutrality goals, but a gap remains between current actions and global targets. To achieve carbon neutrality, countries worldwide must closely integrate innovation and green development concepts and actively promote GI. AI technology, with its advanced data processing and analysis capabilities, is widely regarded as having the potential to help businesses achieve green development goals [1]. As an important technological revolution, AI has become a crucial driving force in leading technological revolutions, promoting industrial structure upgrading, and advancing the development of new quality productivity [2]. Many countries, including the United States and EU member states, have included AI in their national strategic plans. In recent years, China has issued a series of important policies supporting the development of AI technology. In 2019, the Chinese government issued the policy for establishing the NAIPZ. Shanghai and Beijing were the first to establish AI pilot zones, followed by Shenzhen, Hangzhou, Tianjin, and other places. As of 2022, a total of 18 cities and regions in China had established AI pilot zones. These zones are tailored to local conditions, utilizing regional resource advantages and industrial agglomeration effects to advance AI research and technology applications. In 2023, the scale of China’s AI core industry expanded considerably, reaching CNY 578.4 billion, and it continued to rise at a 14% growth rate. As promoters of green productivity, practitioners of green transformation, and leaders of environmental regulations, enterprises play an indispensable role in promoting the green transformation of economic development and enhancing the level of GI. In this context, AI technology, as an important support for the development of GI in enterprises, is demonstrating its enormous potential and value.

The application of AI can help save costs and alleviate insufficient funding for GI in manufacturing enterprises [3]. Introducing intelligent machines into the manufacturing production process can considerably reduce enterprise costs, not only in terms of replacing low-skilled labor engaged in repetitive and procedural work but also in terms of managing the full cycle of visualization, automation, and intelligent production. In addition, AI can help enterprises eliminate the information asymmetry in GI, increase creative exchange, collision, and cooperation, discover the connections and intersections between green R&D fields, and promote R&D personnel to obtain key information of GI quickly. However, questions regarding the impact of AI applications on GI in manufacturing industries and the mechanisms behind this remain unclear, making it difficult to provide specific guidance for the real economy. Existing research has failed to provide answers to these questions. On the one hand, substantial literature focuses on the implementation barriers, decision making, operations, business models, and broad goals of AI [4,5]. In the field of innovation, the existing literature mainly focuses on how to empower and accelerate digital innovation, promote product and service innovation, and promote responsible innovation practices [6]; however, the connection between AI technology and GI has not been explored to a large extent. In addition, many scholars have studied how AI affects regional economic growth from a macro perspective. Existing research on the influencing factors of GI mainly focuses on environmental pollution issues, such as the demand for haze pollution control, the effectiveness of environmental regulatory policies, and the improvement of environmental information transparency [7,8]. These studies emphasize the dual driving effects of external environmental pressure and internal resource management on GI behavior without considering the enhancement of GI by intelligent technology. In addition, the existing literature has begun to explore the potential connection between AI and GI, revealing the enormous potential of AI technology in improving production efficiency, optimizing resource allocation, and reducing environmental pollution [9,10]. However, these studies generally measure AI through industrial robot data, AI patents, or AI word frequency at the enterprise level, potentially leading to endogeneity issues and inaccurate results.

This study takes the establishment of China’s NAIPZ as a quasi-natural experiment and systematically analyzes the impact and mechanisms of AI on GI in enterprises. The contribution of this study is threefold. (1) In terms of theoretical research, this study provides innovative insights into the effect of AI on enterprises’ GI. It not only addresses the critical question of whether establishing the NAIPZ can promote GI in enterprises but also offers empirical support and policy recommendations for the steady development of AI pilot zones. This study fills the gap in the literature on AI in the micro-GI field. (2) From a research perspective, we explore the micro-mechanisms through which AI promotes GI in enterprises, focusing on internal governance and human capital optimization. We clarify the theoretical framework and practical implications of intelligent driving on GI in enterprises, thereby enriching the body of research on AI. (3) In terms of research content, we conduct a heterogeneity analysis across three dimensions: geographical location, industry technological level, and the differences in enterprise property rights. Our analysis reveals that the impact of AI on enterprise GI varies according to these factors.

2. Literature Review and Theoretical Analysis

2.1. Literature Review

Two literature sources are highly relevant to this study: one on the economic and environmental effects of AI technology applications and the other on the factors influencing GI in enterprises. In the evaluation of AI’s impact, Graetz and Michaels (2018) empirically demonstrated that industrial robots exhibit higher production efficiency than human labor, significantly enhancing labor productivity and promoting economic growth [11]. Cockburn et al. (2018) and Frey and Osborne (2017) pointed out that AI and automation technologies have had a revolutionary impact on the economic activities and decision making behaviors of enterprises, considerably altering their operating models and competitive dynamics, enabling them to seek more optimized decision making paths in more efficient production [12,13]. Aghion et al. (2022) proposed that industrial automation impacts the efficiency improvement, skill complementarity, technology selection, and industry competition of enterprises. These four effects collectively influence enterprise innovation; therefore, the effects on enterprise innovation are ambiguous [14]. Fang (2021) argued that industrial robots can monitor energy consumption and emissions in real time during the production process, thereby regulating excessive emissions caused by improper energy use in enterprises [15]. Li and Li (2023) constructed a local equilibrium model and demonstrated that incorporating industrial robots into enterprise production can achieve clean production, energy conservation, and emission reduction. Improving total factor productivity and energy utilization efficiency is the primary method for enterprises to engage in green production [16]. Nie et al. (2022) empirically found that the widespread use of industrial robots helps enterprises implement clean production practices [17].

In research on the influencing factors of corporate green innovation (GI), most existing literature focuses on the impact of environmental regulation (ER) and digital transformation (DT) on corporate green innovation (GI). ER includes policies that charge companies for their polluting behavior and provide subsidies for energy-saving and emission-reduction practices. Peng et al. (2021) argued that well-designed ER tools can grant necessary autonomy to individual businesses, thereby incentivizing them to actively produce GI results and reduce costs caused by environmental pollution effectively [18]. Qi et al. (2021) [19] explored the role of the introduction of the carbon emission trading pilot policy in promoting GI among enterprises. Empirical analysis revealed that the GI level of enterprises in the policy pilot areas was improved [19]. Zhou (2024) found that China’s command-and-control ER can improve the quality of GI in enterprises [20]. In contrast, some scholars have come to completely opposite conclusions. Brunnermeier and Cohen (2003) analyzed manufacturing data in the United States and found that ER did not provide the expected additional incentives for GI among companies [21]. Leeuwen and Mohnen (2017) found that ER can inhibit the efficiency of GI in certain situations [22].

Lv et al. (2023) used data at the enterprise and city levels to study the role of DT in the overall quantity and quality of GI in enterprises. The results showed that DT promotes enterprises to engage in more and higher-quality GI activities by influencing environmental attention, and there are spatial spillover effects [23]. Xu et al. (2024) found that DT promotes GI in enterprises by increasing R&D investment and information disclosure, and this effect is particularly notable in high-tech enterprises [24]. With the implementation of the NAIPZ in China, scholars have used this event as a quasi-natural experiment and evaluated the micro effects of the NAIPZ using the difference-in-difference method. Most research results have found that the NAIPZ can effectively improve the ESG level and carbon performance of enterprises [25,26]. However, there has not been in-depth research on GI in enterprises.

2.2. Theoretical Analysis

2.2.1. Direct Impact of AI on GI in Enterprises

First, AI enhances the depth of R&D for enterprises, thereby promoting GI. AI is an extension and deepening of information technology. AI can penetrate and analyze large amounts of data in the R&D process, discover operational patterns in seemingly unrelated data, and help enterprises improve R&D depth while uncovering more valuable information resources [27]. Second, the application of AI promotes the integration of advanced and traditional technologies [28]. The application of AI in manufacturing is not merely about replacing human labor but involves using the introduction and transformation of intelligent machines as a bridge to deeply integrate advanced information technology with traditional manufacturing technology, thus influencing enterprise R&D of intellectual property through human–machine collaboration. The human–machine integrated intelligent system can fully utilize current intelligent information technology, rely on big data, achieve precise control over the entire process of R&D and management, reduce environmental pollution caused by overproduction, and promote the comprehensive improvement of enterprise GI capabilities. Finally, the application of AI can effectively reduce the R&D costs of enterprises [29]. Compared to previous information technologies, AI has the unique feature of deep learning. The development and application of deep learning can considerably reduce the marginal search cost of each R&D effort, enabling enterprises to innovate continuously, conduct trials, and improve green technologies at a lower cost. Overall, the application of AI can promote further integration and upgrading of R&D processes, technologies, costs, and subsequent management, thereby increasing the frequency and success rate of GI in enterprises. Therefore, the R&D efficiency and management capability of enterprise intellectual property will be improved, the traditional enterprise innovation system will be strengthened, and the GI level of the enterprise will be further enhanced. Thus, we propose hypothesis H1: AI can improve enterprises’ GI.

2.2.2. Indirect Impact of AI on GI in Enterprises

(1)

Optimization of internal governance

Intelligent transformation has changed the traditional development and operation modes of enterprises, reshaping new paths and mechanisms of corporate governance. The application of AI technology is beneficial for improving corporate governance by making enterprises more focused on long-term interests [30]. AI technology can improve the information transparency of manufacturing enterprises, reduce irrational managerial behavior, and thus enhance their corporate governance. Specifically, AI reduces the cost of information disclosure for manufacturing enterprises, improves their ability and willingness to disclose information, and contributes to enhanced corporate governance. The use of AI technology can improve the efficiency of information transmission within enterprises, enhance the transparency of production, R&D, internal management, and financial management, increase the supervision of institutional investors and external markets, reduce agency problems to a considerable extent, improve the internal governance environment, and enhance corporate governance [31]. In addition, AI technology can reduce information asymmetry between small and medium-sized shareholders and controlling shareholders, incentivizing small and medium-sized shareholders to engage in corporate governance. AI technology also enables small and medium-sized shareholders to participate more conveniently in corporate governance, thereby improving the level of shareholder governance to some extent [32].

The improvement in corporate governance can motivate managers to actively engage in GI activities [33]. GI activities are an important strategic behavior for enterprises, which are closely linked to corporate governance decision making. However, due to the high risks and uncertainties associated with GI, some managers exhibit short-sighted behavior and prioritize investments in daily business activities over GI activities. The improvement in corporate governance facilitates the operation of relevant supervisory and incentive mechanisms within enterprises, reducing the likelihood of managers neglecting GI activities due to short-term thinking [34]. Furthermore, an improvement in corporate governance enhances the oversight of company managers by various stakeholders, mitigating managerial irrationality and enabling managers to focus more on the long-term development of the enterprise when making innovative decisions. In conclusion, the improvement in corporate governance can reduce or avoid irrational managerial behavior to a certain extent, effectively enhancing the level of GI within enterprises. Thus, we propose hypothesis H2: AI can drive enterprise GI by improving internal governance.

(2)

Upgrading human capital

The application of AI technology can enhance the initiative of enterprise employees to acquire knowledge, foster knowledge exchange across departments, strengthen the innovation capabilities of enterprises, and increase GI output. First, the application of AI technology allows employees across various departments to easily collect valuable information, overcoming time and location constraints, which significantly improves knowledge acquisition efficiency for enterprise employees. Therefore, the use of AI technology helps increase employees’ initiative and enthusiasm for acquiring knowledge, leading to human capital enhancement through knowledge accumulation. Second, improvements in human capital within enterprises enable employees to better understand and master knowledge, facilitating communication and interaction that generate knowledge spillover effects, which in turn influence the enterprise’s GI activities. For R&D and design departments, AI technology enables personnel to easily access valuable information, which contributes to enhancing human capital levels. Finally, improvements in human capital within enterprises foster the exchange of innovative ideas among R&D personnel, breaking through existing technological bottlenecks and enhancing the GI capabilities of manufacturing enterprises.

The intelligent transformation of enterprises requires a high-quality labor force capable of leveraging AI technology to enhance innovation capabilities and improve the GI level of enterprises. The application of AI technology compels manufacturing enterprises to introduce skilled labor to effectively address the challenges encountered during the process of intelligent transformation. Enterprises must also provide specialized AI training to some of their internal employees, turning them into high-quality laborers proficient in applying intelligent technology. Employees with relevant skills can leverage AI technology to drive innovation activities, thereby enhancing the GI level of enterprises [35]. High-quality managers, in particular, can use AI technology to develop more accurate GI strategies, mitigate risks, increase targeted innovation investment, improve R&D success rates, and, ultimately, enhance the GI level of enterprises [36]. Thus, we propose hypothesis H3: AI can drive enterprise GI by upgrading human capital.

2.2.3. Regulatory Effect of Industry Competition

Industry competition is a common phenomenon where companies compete for market share using strategies like pricing, service, innovation, and other means. With technological advancements, the competitive landscape of industries has undergone significant changes. AI has endowed enterprises with unprecedented capabilities, including data analysis, intelligent decision making, and other functions. These new capabilities have improved the efficiency and productivity of enterprises, thereby enhancing their competitiveness [37]. However, this has also resulted in a widening technological gap among companies within the industry, intensifying competition. The level of industry competition exerts external governance effects, reflecting the competitive landscape of enterprises within their industry, which can filter out more competitive firms and influence their GI activities [38].

Increased industry competition can encourage companies to actively utilize AI technology. Faced with intense industry competition, enterprises experience greater competitive pressure and encounter more uncertain risks. To avoid being eliminated from the industry, enterprises will seek to develop new business lines through the introduction of AI and digital technologies, thereby promoting GI, improving green competitiveness, and gaining a competitive advantage [39,40]. Therefore, enterprises can leverage AI technology to access more heterogeneous resources, stand out in the industry, reduce production costs, improve production efficiency, lower pollution emissions, and gain a larger market share. Moreover, the advantages brought by AI technology provide favorable conditions for enterprise innovation. GI in enterprises heavily relies on human, material, and financial resources. Enterprises can use AI and information technology to acquire information, disseminate knowledge, and make R&D decisions, thereby optimizing resource allocation and facilitating intelligent transformation. In competitive industries, external pressures heighten companies’ awareness of the impact of AI technology, prompting them to improve their innovation environments to promote GI activities. Thus, the greater the industry competition, the stronger the role of AI in promoting GI in enterprises. Thus, we propose hypothesis H4: industry competition can increase the promotion effect of AI on GI.

3. Research Design

3.1. Model Settings

DID is a statistical method used to evaluate the effectiveness of policy or project implementation. The basic idea is to estimate the net effect of policy or project implementation by comparing the changes before and after policy implementation between the experimental and control groups. Staggered DID requires the data to be multi-period panel data, with individuals in the control group not receiving policy intervention during any period, while individuals in the treatment group do not receive the intervention at exactly the same time. Additionally, policy withdrawal is not allowed. Therefore, staggered DID is particularly suitable for pilot policies launched in batches across different regions or industries. Given that AI pilot zones were established in batches in 2019, 2020, and 2021, this paper uses staggered DID to examine the impact of AI on GI. The specific model settings are as follows:

$${\mathrm{G}\mathrm{I}}_{it}={\alpha }_0+{\alpha }_1{\mathrm{A}\mathrm{I}}_{it}+{\alpha }_2{\mathrm{C}\mathrm{V}}_{it}+{\mu }_i+{\gamma }_t+{\epsilon }_{it},$$

(1)

where i and t denote the enterprise and the year, respectively, GI denotes the GI level, and AI denotes the NAIPZ policy. In particular, AI_it is a dummy variable (if the city where enterprise i is located established a AI pilot zone in year t, the value is 1; otherwise, it is 0). CV is the control variable matrix. μ denotes the individual fixed effects (FE), and γ denotes the year FE. The coefficient of ${\alpha }_1$ reflects the effect of AI on GI. If the coefficient of ${\alpha }_1$ is significantly positive, it can verify hypothesis H1 in this paper.

3.2. Data Source and Variable Description

3.2.1. Data Sources

Manufacturing is the primary industry for energy consumption and a key sector for the application of AI technology. The year 2013 marked an important turning point in the development of AI technology, with breakthroughs in deep learning demonstrating its tremendous potential across multiple fields. Subsequently, countries around the world, especially developed nations, have increasingly focused on the disruptive impact that AI may have on society and the economy, elevating AI development to a strategic priority. Therefore, we select Chinese A-share-listed manufacturing enterprises from 2013 to 2022 as our sample to test the effect of the NAIPZ on enterprise GT. Before conducting the regression analysis, this paper excludes companies marked as special treatment by the stock exchange, including ST, PT, and *ST companies, and removes samples with missing data for the main variables. Additionally, continuous variables are truncated at the 1% upper and lower tails. Company characteristic data are obtained from the CSMAR database.

3.2.2. Variable Description

This paper uses the logarithm of the number of green patents authorized by enterprises as a measure of the level of GI for listed companies. The number of patent authorizations not only directly reflects technological innovation achievements but also serves as an important criterion for evaluating the substantive innovation level of enterprises [41]. Compared to green patent applications, authorized green patents are often of higher quality and reliability. Therefore, we select green patent authorizations as the indicator for measuring the GI of enterprises, following the approach of Xu et al. (2023) [42].

AI: We use the NAIPZ policy as the proxy indicator for AI. Huang et al. (2024) and Feng et al. (2025) have used the establishment of the NAIPZ as a proxy variable for AI to study the impact of AI on corporate ESG and carbon performance [25,26]. This provides literature support for us to use the NAIPZ, a quasi-natural experiment, to measure AI. The NAIPZ is the dummy variable in this paper, used to indicate whether the city where the enterprise i is located established an AI pilot zone in year t.

Based on the existing literature [19,20,43], we select some enterprise-level control variables that will affect enterprise GI, as shown in

Table 1. Variable description and descriptive statistics.

Variable	Symbol	Definition	Mean	SD
Green innovation	GI	The logarithm of the number of green patents authorized	1.004	1.291
Enterprise size	Size	Ln (enterprise total assets)	22.19	1.193
Asset liability ratio	Lev	Ratio of total liabilities to total assets	0.392	0.181
Return on assets	Roa	Ratio of net profit to total assets	0.048	0.063
Board size	Board	Ln (the number of board members)	2.112	0.19
Ownership concentration	Top	Proportion of shares held by the largest shareholder	0.332	0.138
Tobin’s Q value	TobinQ	Ratio of market value to asset replacement value	2.176	1.34
Enterprise age	Age	Ln (enterprise age + 1)	2.91	0.296
Fixed asset ratio	Fixed	Ratio of net fixed assets to total assets	0.222	0.127
Enterprise R&D investment	Rd	Ratio of R&D investment to operating revenue	0.052	0.045

below.

4. Results Analysis

4.1. Benchmark Regression

Table 2. Overall impact of AI on GT.

	(1)	(2)	(3)
	GI	GI	GI
AI	0.207 ***	0.384 ***	0.197 ***
	(0.0350)	(0.0387)	(0.0351)
Size		0.388 ***	0.129 ***
		(0.0165)	(0.0359)
Lev		1.036 ***	−0.0830
		(0.0880)	(0.122)
Roa		0.785 ***	0.261
		(0.212)	(0.200)
Board		−0.0230	−0.0331
		(0.0705)	(0.0972)
Top		−0.251 ***	0.0831
		(0.0936)	(0.207)
TobinQ		−0.0145	0.00506
		(0.00959)	(0.00951)
Age		−0.140 ***	0.122
		(0.0478)	(0.234)
Fixed		−0.993 ***	−0.103
		(0.107)	(0.160)
Rd		0.405 ***	0.0996 **
		(0.0419)	(0.0436)
Firm FE	YES	NO	YES
Year FE	YES	YES	YES
Observations	9055	9055	9055
R-squared	0.731	0.212	0.732

Note: Robust standard errors are in parentheses; * p < 0.1, ** p < 0.05, and *** p < 0.01.

reports the impact of AI on GI in enterprises. When control variables and FE are not included, the regression coefficient of AI is 0.207, which is statistically significant, indicating that AI has a significant impact on GI in enterprises. When control variables and FE are included, the regression coefficient of AI on GI in enterprises is 0.197, indicating that the higher the level of AI usage in enterprises, the more beneficial it is for initiating GI. From the estimated coefficient size of the model, the implementation of the NAIPZ policy has increased the GI level of pilot enterprises by 0.19% compared to non-pilot enterprises. Our conclusion aligns with the findings of existing scholars. Lee et al. (2022) used robot data to demonstrate that the application of AI technology can improve the GI level in enterprises [44]. At the same time, AI has significant environmental governance effects and can effectively reduce corporate pollution emissions [45,46]. This study also reveals the significant positive impact of AI on corporate GI by utilizing the exogenous impact of the NAIPZ, enriching the literature on the environmental governance effects of AI.

4.2. Robustness Tests

4.2.1. Parallel Trend

The application of AI in various enterprises demonstrates that the timing of AI adoption varies across manufacturing enterprises. Therefore, the event study method is used for parallel trend testing. The specific approach involves adding the interaction term between the treatment group dummy variable and the time dummy variable in model (1). If the coefficients of the interaction term prior to the AI application point are insignificant, then the parallel trend test is considered passed. We present the parallel trend results in a graph, as shown in Figure 1. It can be observed that when enterprises adopt AI, significant changes in GI occur between pilot and non-pilot enterprises.

4.2.2. Placebo Test

To further investigate the extent to which the impact of AI pilot zones on GI is influenced by random factors, we conduct a placebo test to mitigate this potential bias. We follow the approach of La Ferrara et al. (2012) [47] and randomly sample 500 times to construct a “virtual policy dummy variable” based on the distribution of AI variables in the benchmark regression. We then re-estimate the DID coefficients and p-value distribution using model (1), as shown in Figure 2. The mean regression coefficients for enterprise GI are close to 0, and the p-values are mostly greater than 0.1. Therefore, the impact of AI on corporate GI is not driven by random factors, and the conclusion drawn in the previous section is reliable.

4.2.3. Replace Core Variable

Increasingly, studies in the literature use text analysis to measure the level of AI in enterprises and subsequently use it as a proxy variable for AI. This paper selects the frequency of AI-related terms in the annual reports of manufacturing companies to characterize the AI indicator and applies the logarithmic transformation.

Table 3. Replace explanatory variable.

	(1)	(2)	(3)
	GI	GI	GI
AI	0.166 ***	0.162 ***	0.102 ***
	(0.0437)	(0.0388)	(0.0335)
CV	NO	YES	YES
Firm FE	YES	NO	YES
Year FE	YES	YES	YES
Observations	9055	9055	9055
R-squared	0.73	0.206	0.731

Note: Robust standard errors are in parentheses; *** p < 0.01.

presents the results after replacing the explanatory variable, indicating that AI improves the GI of enterprises.

4.2.4. Exclusion of the Effects of Environmental Policies

GI is influenced by various environmental policies issued at different levels in China, such as carbon trading policies, low-carbon city policies, and green credit policies. Because it is not feasible to eliminate the effects of all environmental policies, we primarily control the policy impacts at the provincial, city, and industry levels. Specifically, we introduce interaction terms for province FE and year FE, city FE and year FE, and industry FE and year FE in model (1). The results are presented in columns one to three of

Table 4. Exclusion of impact of environmental policies.

	(1)	(2)	(3)
	GI	GI	GI
AI	0.267 ***	0.102 *	0.212 ***
	(0.0519)	(0.057)	(0.0370)
CV	YES	YES	YES
Firm FE	YES	YES	YES
Year FE	YES	YES	YES
Observations	9040	8505	9016
R-squared	0.742	0.769	0.743

Note: Robust standard errors are in parentheses; * p < 0.1 and *** p < 0.01.

, and the coefficients for AI remain significant.

4.2.5. Propensity Score Matching (PSM)-DID

This paper uses PSM-DID to verify the robustness of the core conclusion. Firstly, we use a Logit regression model to estimate the propensity score for each enterprise. Secondly, based on the PSM method, we select non-pilot enterprises whose propensity scores most closely match those of the pilot enterprises. This method helps reduce systematic differences in GI capabilities among different enterprises, thereby enabling a more accurate assessment of the impact of AI pilot zones. Finally, nearest neighbor, radius, and kernel density matching estimation methods are used to verify the reliability of the results. As shown in

Table 5. PSM-DID.

	(1)	(2)	(3)
	GI	GI	GI
AI	0.159 *	0.199 ***	0.197 ***
	(0.08)	(0.0352)	(0.0351)
CV	YES	YES	YES
Firm FE	YES	YES	YES
Year FE	YES	YES	YES
Observations	2305	9010	9002
R-squared	0.808	0.733	0.732

Note: Robust standard errors are in parentheses; * p < 0.1 and *** p < 0.01.

, the estimated coefficients for the core explanatory variable, AI, are significantly positive at the 10% level, further confirming the robustness of the above research conclusions.

4.2.6. Other Robustness Tests

Considering that companies within the same industry or city may exhibit similarities in GI and that differences in GI may exist among companies across different industries or cities, this paper employs clustering robust standard errors at the industry and city levels for regression analysis. The regression results are presented in columns (1) and (2) of

Table 6. Other robustness tests.

	(1)	(2)	(3)	(4)
	GI	GI	GI	GI
AI	0.197 ***	0.197 ***	0.204 ***	0.216 ***
	(0.0424)	(0.0504)	(0.043)	(0.062)
CV	YES	YES	YES	YES
Firm FE	YES	YES	YES	YES
Year FE	YES	YES	YES	YES
Observations	9055	9055	6353	9055
R-squared	0.732	0.732	0.762	0.743

Note: Robust standard errors are in parentheses; *** p < 0.01.

, indicating that even with changes in clustering hierarchy, the coefficients of AI remain positive. The longer the sample interval, the more susceptible the research results are to potential interference. Therefore, we further narrow the sample interval to 2016–2021, as shown in column (3). After adjusting the sample range, the conclusions of this paper remain robust. The selection of NAIPZ demonstration cities is closely related to factors like urban economic development, industrial structure, and location characteristics, all of which significantly impact corporate GI. To investigate the impact of urban characteristic variables on GI over time, we include interaction terms between key variables, such as per capita GDP, industrial structure level, and whether the city is at or above the sub-provincial level, along with linear trends over time in the benchmark model to alleviate the non-random selection problem of NAIPZ demonstration cities. Column (4) presents the regression results, which remain robust even after controlling for non-random factors, such as urban characteristic variables.

4.3. Mechanism Verification

4.3.1. Analysis of Intermediary Effect

We construct model (2) to estimate how AI affects GI through internal governance and human capital.

$${Med}_{it}={\alpha }_0+{\alpha }_1\mathrm{A}\mathrm{I}+{\alpha }_2{\mathrm{C}\mathrm{V}}_{it}+{\mu }_i+{\gamma }_t+{\epsilon }_{it},$$

(2)

In model (2), Med refers to enterprise internal governance (IG) and human capital (HC). This paper uses the internal control index from the DIB database to measure the level of IG in enterprises. A higher value indicates a higher level of IG in an enterprise. We use the proportion of employees with a bachelor’s degree or above in a company to measure its HC level.

Table 7. Mechanism test.

	(1)	(2)
	IC	HC
AI	0.0323 *	0.125 *
	(0.0179)	(0.0672)
CV	YES	YES
Firm FE	YES	YES
Year FE	YES	YES
Observations	9055	9055
R-squared	0.449	0.81

Note: Robust standard errors are in parentheses; * p < 0.1.

demonstrates the impact of AI on IG and HC. The coefficients for AI are positive, indicating that AI enhances both IG and HC in enterprises. The existing literature supports these findings, confirming the significant correlation between AI and IG and HC [30]. Existing research has identified IG and HC as key factors in promoting improvements in GI levels within enterprises. Enhancing corporate governance can also mitigate or prevent irrational managerial behavior to some extent, thereby effectively improving the GI level in enterprises [33]. Additionally, related studies suggest that high-quality employees with relevant skills can apply AI technology to innovation activities, thereby enhancing the GI level of enterprises [35]. Therefore, AI can promote corporate GI by improving IG and HC.

4.3.2. The Regulatory Effect of Industry Competition

We analyze the moderating effect of industry competition on the relationship between AI and GI in enterprises. This paper uses the Herfindahl–Hirschman Index (HHI) to measure the level of industry competition. A higher HHI value indicates lower industry competition.

Table 8. Regulatory analysis.

	(1)	(2)
	GI	GI
AI*HHI	−0.104 *	−0.075 *
	(0.057)	(0.04)
AI	0.225 ***	0.212 ***
	(0.0545)	(0.055)
HHI	−0.176 *	−0.233 *
	(0.095)	(0.122)
CV	NO	YES
Firm FE	YES	YES
Year FE	YES	YES
Observations	9055	9055
R-squared	0.731	0.733

Note: Robust standard errors are in parentheses; * p < 0.1, *** p < 0.01.

presents the regression results for the interaction term between AI and HHI. The coefficients of AI*HHI are negative, suggesting that greater industry competition strengthens the positive impact of AI on GI. Therefore, industry competition enhances the positive effects of AI on GI in enterprises.

4.4. Heterogeneity Analysis

4.4.1. Geographical Position

There are differences in the overall development strength across regions, and the spillover effects of technology introduction and innovation opportunities also vary, which affect the GI behavior of enterprises. We divide the sample enterprises into two groups based on their regions, the eastern region and the mid-west region, and perform grouped regression tests. The test results are shown in

Table 9. Heterogeneity analysis.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)
	Eastern	Mid-West	Labor -Intensive	Capital -Intensive	Technology -Intensive	SOEs	Non-SOEs
	GI	GI	GI	GI	GI	GI	GI
AI	0.234 ***	0.181 *	0.217 ***	0.0249	0.266 **	0.2 ***	0.229 ***
	(0.066)	(0.095)	(0.0442)	(0.0708)	(0.110)	(0.042)	(0.071)
CV	YES	YES	YES	YES	YES	YES	YES
Firm FE	YES	YES	YES	YES	YES	YES	YES
Year FE	YES	YES	YES	YES	YES	YES	YES
Observations	6790	2265	761	2556	5738	2432	6623
R-squared	0.711	0.673	0.564	0.666	0.753	0.775	0.703

Note: Robust standard errors are in parentheses; * p < 0.1, ** p < 0.05, and *** p < 0.01.

. The regression results indicate that AI has a significant positive impact on corporate GI behavior in both the eastern and mid-west regions of China. However, in terms of the magnitude of the coefficient, the absolute value in the eastern region is higher than that in the mid-west region. This phenomenon may be attributed to regional differences in innovation resources. The eastern region benefits from abundant innovation resources, well-developed innovation networks, and more government policies encouraging GI behavior, which reduce innovation risks, increase R&D investment, and ultimately improve GI levels.

4.4.2. Industry Heterogeneity

AI technology exhibits specific applicability in different application scenarios, and its impact on GI in enterprises may vary across industries. Following the classification method of Huang et al. (2023), we categorize listed companies into three industry types: labor-intensive, capital-intensive, and technology-intensive [48]. The specific results are presented in Table 9, confirming that the promotion effect of AI technology on enterprise GI is primarily concentrated in technology-intensive industries. This suggests that technology-intensive industries, due to their complexity and reliance on advanced technology, generally have high data accessibility and processing demands, allowing them to fully leverage the intelligent advantages of AI technology. The deep integration of AI technology can replace certain repetitive and low-creativity R&D activities, increasing firms’ reliance on AI and thereby enhancing their innovation motivation, demonstrating AI’s significant role in promoting GI behavior in enterprises. Meanwhile, R&D personnel constitute a large proportion of employees in high-tech enterprises, and the optimization effect of human capital is more pronounced. Thus, the promoting effect of AI on GI behavior is more evident in these industries.

4.4.3. Property Rights Nature

Enterprises are key microeconomic entities in the application of AI technology, and different types of enterprises may exhibit heterogeneous effects on GI behavior. Given the importance of enterprise type, this paper examines the heterogeneous effects of AI on GI from the perspective of ownership structure. Table 9 presents the impact of AI on GI in state-owned enterprises (SOEs) and non-SOEs. The results indicate that the coefficient of AI in non-SOEs is higher than in SOEs, suggesting that AI has a stronger positive effect on GI in non-SOEs. This discrepancy may be attributed to differences in factor adjustment flexibility across ownership structures. Compared to non-SOEs, SOEs face greater rigidity in factor adjustments, lower sensitivity to market dynamics, and limited flexibility in resource allocation. The rigid allocation of human capital and physical assets in SOEs constrains their ability to fully harness the innovation-enhancing potential of AI. Additionally, AI tends to have a greater impact on high-productivity enterprises. Given that non-SOEs generally exhibit higher productivity, the AI-induced enhancement in GI behavior is more pronounced.

5. Conclusions

Promoting enterprise GI and fostering social sustainable development has become a critical issue. This study examines the impact of AI on GI in enterprises from the perspective of the NAIPZ policy. Our findings suggest that the establishment of AI pilot zones significantly enhances GI achievements among enterprises in the pilot areas, indicating that AI facilitates GI. The mechanism analysis reveals that AI promotes GI in enterprises by enhancing internal governance and optimizing human capital, while industry competition further strengthens this effect. The heterogeneity analysis results indicate that the effect of AI on GI is particularly pronounced in the eastern region, high-tech industries, and non-SOEs. Previous studies have utilized industrial robot data to explore the relationship between AI and GI at both the provincial and enterprise levels, revealing a significant improvement effect of AI on GI [3,9,44]. Our conclusions align with these findings to some extent. We make a novel contribution by identifying and testing the mediating roles of internal governance and human capital, enriching research on the channels through which AI affects GI. Furthermore, we employ a quasi-natural experimental approach to investigate the relationship between AI and GI, which helps mitigate endogeneity issues. Our study also expands the policy evaluation framework of the NAIPZ policy. Based on these findings, we propose the following recommendations.

Given that AI pilot zones can enhance GI, policymakers can improve GI in enterprises through policy optimization in AI pilot zones. The government should steadily promote the expansion of AI pilot zones and comprehensively enhance GI in enterprises. Particularly in cities with well-established AI infrastructure and a favorable industrial environment, even though they have not yet been officially designated as AI pilot zones, their development potential in AI is enormous. Enterprises in non-pilot areas can gradually develop an efficient and feasible GI model by drawing on the valuable experiences of AI pilot zones, providing key references for the future expansion and quality improvement of AI pilot zones. To mitigate the adverse effects of frequent policy or technological changes, it is crucial to ensure a steady increase in GI quantity while conducting in-depth research and pursuing higher GI quality. This ensures that AI pilot zones can be effectively and continuously promoted. Additionally, other developing countries can learn from China’s experience in developing AI pilot zones and enhance their green competitiveness by investing in AI development and formulating supportive policies.

The mechanism results indicate that optimizing human capital serves as a crucial channel for AI to enhance GI in enterprises. Therefore, enterprises should focus on acquiring and applying new knowledge and technologies while attracting high-level talent. Enterprises must also increase their R&D investment and recruit skilled R&D personnel. The participation of R&D personnel enhances the efficiency of utilizing new knowledge, facilitating result transformation and high-level innovation output. Enterprises should strive to enhance their human capital and optimize their labor structure. Human capital is a critical resource for innovation, as high-level talent possesses strong learning and knowledge application abilities, allowing them to rapidly grasp cutting-edge knowledge and technology. Moreover, high-level talent can engage in both complementary and autonomous innovation with AI technology.

Industry competition enhances the role of AI in promoting GI in enterprises. Therefore, in the current context of green development, enterprises should recognize the strategic importance of improving their GI level for sustainable growth and cultivate a strong sense of social responsibility. Enterprises should actively integrate the concept of green development into both green process innovation and green product innovation from a long-term perspective, continuously optimizing production processes and enhancing product quality to better meet market demand. Moreover, enterprises should embed environmental responsibility and GI into their corporate culture, aligning with the national green development strategy in the new era. By doing so, enterprises can enhance their corporate image and reputation, gain greater trust from social stakeholders, mitigate negative externalities, and address legal and ethical concerns associated with GI. This approach fosters stronger connections with key stakeholders, including the government and consumers. Furthermore, AI promotes enterprise GI by improving internal governance. Thus, heavily polluting enterprises should optimize their internal management systems, improve production processes, rethink their development strategies, and align their growth with environmental and social sustainability.

This study has certain limitations. Our research on the impact pathways of AI on GI in enterprises is relatively conventional, and we have not explored other potential pathways in depth. Different AI pilot cities have implemented varying policy measures to foster AI development, which may have heterogeneous impacts on corporate GI. However, we have not conducted a comprehensive evaluation of the impact of different policy designs in these pilot cities. Moreover, due to data limitations, we were unable to identify specific cases to support our research hypothesis regarding the impact of AI on corporate GI. In future research, we will conduct an in-depth analysis of the impact mechanisms and policy heterogeneity effects of AI on corporate GI.