Harnessing AI for Green Innovation: The Role of Executive Cognition

Abstract

While AI is widely recognized as an industrial transformation catalyst, how AI translates into green innovation remains insufficiently understood. Drawing on socio-technical systems theory and upper echelons theory, this study investigates how AI adoption influences green innovation and how managerial cognition shapes this relationship. Using data from Chinese A-share listed firms spanning 2012 to 2024, we reveal that AI significantly promotes green innovation by serving as an endogenous technological force. Managerial cognition (green cognition, innovation cognition, long-termism) serves as a critical boundary condition: all three dimensions positively moderate the AI–green innovation nexus, indicating equivalent technological inputs yield divergent outputs depending on executive interpretation frameworks. Mechanism analyses demonstrate AI operates through three channels: information transparency (improving carbon data quality), compliance internalization (embedding requirements into digital systems), and value creation (transforming environmental data into profit sources). Heterogeneity tests show AI’s effect is more pronounced in high-tech industries and under intense market competition. This study reveals the moderating role of managerial cognition—and its multidimensional construct—in the relationship between AI and green innovation. Practically, it provides actionable insights for cultivating managerial cognition to bridge the gap between AI potential and green innovation realization.

1. Introduction

Adopting green practices remains a significant consideration for today’s firms [1,2]. Incorporating environmental sustainability into corporate strategy has become a strategic imperative and a key priority for firms [3,4]. Prior research on green innovation mainly adopted an output-oriented perspective, emphasizing the value, with a particular emphasis on corporate performance [5,6], derived from firms’ green activities, without fully examining their operational modes and motivators, thereby leaving cross-firm differences in green innovation engagement largely unexplained. Given that green innovation is characterized by lengthy cycles, substantial investment, high risks, and dual externality issues [7,8], corporate motivation to pursue it may often be insufficient [9]. Therefore, stimulating its vitality necessitates a deeper exploration of its intrinsic drivers.

Currently, the extensive application of artificial intelligence has progressively expanded from production stages to R&D activities, driving the development of industrial intelligence while simultaneously reshaping corporate innovation behavior [10]. The deep integration of artificial intelligence and green innovation is crucial for energy transition, promising to shift AI’s role from mere “instrumental application” to “systemic reshaping”. As a key driving force in the new round of scientific and technological revolution and industrial transformation, artificial intelligence holds significant potential in enhancing corporate productivity and fostering economic growth [11]. It assists human employees in boosting creativity, particularly benefiting experts with stronger job skills [12], and provides pathway support for overcoming development barriers during technological transformation by improving production operations and enhancing labor efficiency [13]. However, the lack of an underpinning theoretical framework and difficulty accessing data are cited as barriers to further understanding the link between artificial intelligence and green innovation, and the pathways through which AI drives corporate green innovation have also rarely been systematically explored.

Notably, a profound academic paradox presents itself: why do enterprises exhibit significant heterogeneity in both the pathways adopted and the outcomes achieved in green innovation despite operating under comparable technological and market conditions? A critical yet often overlooked moderating factor may lie in the role of the manager. Corporate actions are fundamentally influenced by managerial cognition, a dimension insufficiently accounted for in conventional analytical frameworks. Prevailing theories, such as technological determinism or institutional theory, reveal explanatory gaps in addressing this observed variation. Technological determinism sees technology as an autonomous force. It predicts that AI adoption should yield uniform green innovation outcomes. This contradicts the evident differences across firms. Institutional theory emphasizes external pressures. These include regulatory, competitive, and social forces. Yet, it cannot explain why firms facing identical pressures choose divergent green innovation paths. These gaps call for a cognition-centered perspective. Such a view highlights how managers interpret AI’s potential. It also underscores their strategic agency in converting technological possibilities into innovation outcomes. In contrast, the lens of managerial cognition offers a pivotal theoretical perspective to bridge this gap. A firm’s strategic response is, in essence, a reflection of its managers’ strategic intent [14]. In the context of corporate greening, managerial cognition filters through the organizational hierarchy to instigate specific initiatives [15]. Substantial research corroborates that top management, by deliberately adjusting organizational structures, can catalyze the emergence of strategic responses to new innovation imperatives and hasten their implementation [16]. Managerial cognition, a multidimensional construct, shapes how executives interpret and act within their strategic environment. As a general-purpose technology, the application of AI as a green enabler is fundamentally filtered through managerial cognition.

Based on the integration of socio-technical systems theory and upper echelons theory, we propose a unified analytical framework. Socio-technical systems theory offers the foundational lens of “technology-social dual subsystems.” It posits that organizational outcomes emerge from the dynamic interaction between technical elements (AI) and social elements (managerial agency). Upper echelons theory then identifies managerial cognition as the core element within the social subsystem. This cognition filters and shapes strategic responses to technological opportunities. From this integration, we derive three cognitive dimensions: managerial green cognition (motivation—whether executives prioritize environmental sustainability), managerial innovation cognition (capability—whether executives maintain open, exploratory orientations toward new technologies), and managerial long-termism (commitment—whether executives sustain strategic patience for long-cycle investments). Together, these dimensions capture the complete cognitive infrastructure required to translate AI potential into green innovation outcomes.

Grounded in this theoretical synthesis, this study delineates the critical role of managerial cognition in guiding how artificial intelligence fosters green innovation. Using panel data from Chinese A-share listed firms from 2012 to 2024, we examine how AI adoption influences corporate green innovation. Specifically, we address two questions: whether AI technology promotes green innovation, and how managerial cognition acts as a boundary condition in this relationship. Our findings offer dual contributions: theoretically, providing a systematic basis for enhancing the quality and effectiveness of corporate green innovation; and practically, furnishing theoretical insights and empirical evidence to support the transformation and upgrading of related industries through AI, thereby advancing the high-end, intelligent, and sustainable development of enterprise.

While prior literature has predominantly explored green innovation through the lenses of institutional pressure or resource-based view, and has largely treated artificial intelligence as an exogenous technological shock or a productivity-enhancing tool, this study shifts the analytical focus toward the endogenous role of AI and its interaction with managerial cognition. By integrating socio-technical systems theory and upper echelons theory, we propose a cognition-centered contingency framework that explains not only whether AI promotes green innovation but also how and under what conditions this translation occurs. This perspective moves beyond technological determinism and offers a more nuanced understanding of firm-level heterogeneity in green innovation outcomes.

This paper contributes to the literature in the following ways. First, we establish a positive relationship between AI adoption and corporate green innovation. This shifts scholarly attention from the consequences of green innovation to its antecedents. Prior work has focused on the outcomes of green innovation, such as its positive impact on financial and ESG performance. Research has also explored external enablers like environmental policy. However, the internal drivers of green innovation remain underexamined. By showing that a firm’s strategic engagement with AI acts as a significant internal motivator, our work addresses this gap. It enriches the theoretical understanding of what drives green innovation within firms. Second, we propose managerial cognition as a moderating framework. This explains how executives’ interpretations shape the translation of AI capability into green outcomes. While the overall positive effect of AI is recognized, substantial heterogeneity in firm-level outcomes exists. Why similar technological inputs lead to different innovative outputs has lacked a clear explanation. We argue that this heterogeneity is fundamentally shaped by managerial cognition. By integrating the lenses of managerial green cognition, innovation cognition, and long-termism, we provide a focused framework. It captures the motivational, absorptive, and commitment-related dimensions of management. This moves the discourse beyond simple “technology-push” or “institution-pull” narratives. It offers a more nuanced, mechanism-based account of strategic choice and reveals the key boundary conditions for translating AI into green innovation.

The rest of this paper is structured as follows: Section 2 proposes hypotheses based on existing research. Section 3 describes the research design and methodology. Section 4 presents the results and discussions. Section 5 displays further research. Section 6 concludes this study.

2. Theoretical Analysis and Research Hypotheses

2.1. Artificial Intelligence and Green Innovation

2.1.1. The Concept and Characteristics of Artificial Intelligence

AI has been suggested as the “Fourth Industrial Revolution” [17], which is characterized by a shift in decision-making from humans to machines [18]. In essence, the advancement of AI applications is a process of dynamic adaptation between technology and organizational structures. Existing research primarily defines artificial intelligence from technological and functional perspectives. The former emphasizes AI’s ability to perform complex tasks traditionally reserved for humans by employing technologies such as logical rules and machine learning [19], while the latter highlights AI’s technical characteristics of enhancing human decision-making and assuming traditional “human” roles [20]. This paper defines artificial intelligence as a suite of complex technologies, including deep learning and computer vision, that can be programmed through algorithms to mimic human capabilities. It embodies three core functions of assisting humans, replacing humans, and surpassing humans.

While artificial intelligence shares common features with digital technologies, it also possesses the following distinctive attributes. First, self-learning capability [21]: through algorithms such as deep learning and machine learning, artificial intelligence can develop human-like reasoning and inductive abilities by interacting with external information environments. Second, strong generality [22]: Artificial intelligence exhibits cross-domain and cross-task adaptability. It can deeply integrate into specific professional fields while also enabling collaborative innovation across multiple domains through knowledge transfer, demonstrating powerful adaptability and scalability. This positions AI not merely as a tool for compliance, but as a strategic driver that enables firms to actively redesign operations, offerings, and business models to advance green transformation, thereby shifting the locus of green innovation from external pressures to internal technological agency.

2.1.2. Baseline Hypothesis

Socio-technical systems theory posits that organizations are composite systems. These systems form through dynamic interactions between social and technical subsystems [23,24]. The social system consists of organizational members. It also includes their interactive networks. This system encompasses behaviors of individuals or groups. These behaviors occur in developing and applying technical elements. The technical system comprises emerging technologies. It also includes tools, equipment, and related elements [25]. The social system influences the technical system’s effectiveness. Changes in the technical system also trigger adjustments in the social system. Both jointly determine organizational outcomes [26].

In the era of artificial intelligence, the technical system is undergoing a paradigm shift driven by algorithmic technologies like machine learning and deep learning. This technical system, characterized by autonomous decision-making, continuous evolution, and data dependency, fundamentally differs from traditional rule-based technical architectures [27]. Building on socio-technical systems theory, artificial intelligence, as a general-purpose technology, not only embeds itself as a technical element but also acts as a restructuring force, collectively driving green innovation within the multi-level socio-technical system (landscape, regime, and niche levels).

AI promotes green innovation in three ways. First, AI serves as an “innovation process accelerator” and “solution incubator.” It processes multi-source data. These sources include environmental, production, and supply chain data. This enhances the ability to identify opportunities for green technologies. It also enables rapid experimentation and it optimizes efficiency. These effects increase the diversity and feasibility of green innovation. Second, AI acts as a “system architecture reconfigurer.” It permeates the regime level. It promotes evolution of deep-seated rules. Consequently, it shapes institutionalization of green innovation concepts. This occurs from both organizational and participant dimensions. Third, AI-enabled green practices emerge in institutional and market spheres. They give rise to new standards. They also create new business models. These trigger adjustments in policies. They also trigger adjustments in incentive systems. These effects broaden market and institutional space for green innovation. Simultaneously, AI platforms foster development of green supply chains. These supply chains are more transparent. They are also more collaborative. AI platforms also foster industrial ecosystems. These collectively contribute to continuous refinement of green innovation.

Based on the above analysis, we propose a hypothesis:

H1. 

The application of artificial intelligence within firms is more likely to foster green innovation.

2.2. The Moderating Effect of Executive Cognition

Socio-technical systems theory provides the analytical lens of “technology-social dual subsystems,” yet it remains silent on which specific elements within the social subsystem drive technological outcomes. Upper echelons theory, which posits that an organization’s strategic choices are a reflection of the cognitive structures of its top management team [28], fills this gap by specifying managerial cognition as the core element within the social subsystem that filters and directs technological potential. Thus, the two theories are organically linked: while socio-technical systems theory establishes that both technical and social subsystems co-determine green innovation outcomes, upper echelons theory identifies managerial cognition as the critical social subsystem component that explains why firms with equivalent AI capabilities exhibit divergent green innovation performance.

As a general-purpose technology, the potential of AI to serve as a “green enabler” must be strategically realized. While artificial intelligence propels the advancement of the technical subsystem, the effective performance of this integrated socio-technical system depends on the underlying support of the social system [25]. Whether it is deployed to optimize processes for efficiency or to redesign business models for sustainable value creation hinges critically on how managers frame its strategic role. Therefore, unpacking how AI adoption shapes green innovation requires a close examination of managerial cognition.

Managerial cognition is multidimensional, encompassing how executives perceive, interpret, and enact their strategic environment [29,30]. Integrating upper echelons theory with a socio-technical systems perspective, this study advances a cognition-centered framework. We argue that three cognitive dimensions—managerial green awareness, innovation orientation, and long-termism—form a coherent theoretical triad.

Although the three cognitive dimensions—green cognition, innovation cognition, and long-termism—are conceptually related, they capture distinct aspects of managerial interpretation. Managerial green cognition reflects the content of strategic attention [31], i.e., whether environmental sustainability is perceived as a core strategic issue. Managerial innovation cognition reflects the process orientation toward novelty and uncertainty [32], shaping how managers approach technological change and absorb new knowledge. Managerial long-termism reflects the temporal horizon of decision-making [33], influencing the willingness to commit resources to initiatives with delayed payoffs. Together, they form a tripartite cognitive lens—motivation (green), capability (innovation), and commitment (time)—through which managers interpret and act upon AI’s potential for green innovation.

2.2.1. Managerial Green Cognition

Managerial green cognition captures whether executives regard “green” as a central strategic agenda [34]. It determines how managers prioritize environmental sustainability in corporate strategy [35], thereby shaping whether and how AI is leveraged for green innovation. This cognitive lens determines how executives interpret the strategic relevance of external signals (such as environmental regulations and green market demands) and internal technological capabilities (such as the algorithmic advantages of AI) for advancing green innovation. For the firm, managerial green cognition influences both the intensity of willingness to pursue green innovation and the priority accorded to resource allocation, explaining why some companies proactively engage in sustainability initiatives that go beyond mere compliance.

Managers with higher green cognition identify green innovation opportunities in complex data [36]. They may assign greater weight to environmental benefits in AI project evaluations, commit sustained resources to long-term green innovation initiatives, and view AI-driven green innovation as a pathway to competitive advantage, not merely efficiency gains. This cognitive framework steers AI investment toward environmental challenges. AI’s technical characteristics strengthen this process. Visual analytics, simulation, and scenario modeling help managers translate abstract goals into concrete business models. Managers observe feasible pathways and potential returns more clearly. This reinforces their green awareness. It guides consistent commitment to green innovation.

When managerial green cognition is high, AI becomes a dynamic “engine” for green innovation. Firms facing similar pressures diverge in strategy. Some limit AI to compliance-oriented tasks. Others achieve synergy between cognition and technology. They leverage AI for low-carbon technologies, net-zero supply chains, and novel green business models.

Based on the above analysis, we propose a hypothesis:

H2a. 

Managerial green cognition plays a positive moderating role in the relationship between artificial intelligence and green innovation.

2.2.2. Managerial Innovation Cognition

Managerial innovation cognition reflects open or conservative attitudes toward innovation. It drives AI-enabled green innovation. This open cognitive framework [37] guides managers in interpreting emerging technologies. It shapes innovation willingness, performance, and capability [38]. This explains why some firms seize AI opportunities, while others lag.

High innovation cognition managers approach AI as “explorers” [39]. They conceptualize AI not merely as a tool for efficiency gains but as a strategic enabler of green value creation. In navigating the uncertainties inherent in early-stage initiatives, they are willing to shoulder technological and market ambiguities. By fostering absorptive capacity through cross-functional teams and agile governance mechanisms, they accelerate the integration and application of AI’s underlying technical logic within environmental innovation processes.

AI’s predictive accuracy and automation capabilities reinforce this cognition. They stimulate continued AI deployment for green advancement. Under equivalent AI conditions, higher innovation cognition accelerates green innovation penetration. Firms with weak cognition confine AI to automation. Those with strong cognition drive systemic reconstructions of green processes and business models. They transition from “possessing AI” to “innovating with AI.”

Based on the above analysis, we propose a hypothesis:

H2b. 

Managerial innovation cognition plays a positive moderating role in the relationship between artificial intelligence and green innovation.

2.2.3. Managerial Long-Termism

Managerial long-termism reflects managers’ temporal orientation toward technology investments and sustainable competitive advantage [33]. It determines whether firms sustain commitment to long-cycle, high-uncertainty AI-enabled green innovation. This cognition represents strategic patience at the organizational level. It guides managers to view AI green innovation as a vehicle for future core competencies. It provides patience to endure long cycles, high uncertainty, and delayed returns. It resists short-term performance pressures.

Higher long-termism managers adopt “intertemporal value trade-offs” [40]. They characterize AI-driven green innovation as a strategic investment rather than a financial burden, exhibiting unwavering strategic commitment amidst uncertainty and volatility. By continuing to allocate resources to AI-oriented environmental projects, retaining specialized talent, and preserving organizational spaces for learning, they create temporal windows that facilitate the maturation and commercialization of emerging technologies.

AI’s evolutionary logic reinforces long-termism. AI capabilities grow exponentially through data accumulation and algorithmic iteration. Examples include optimizing supply chain carbon footprints. These outputs provide observable evidence of staged value creation. Under equivalent AI conditions, long-termism guides organizations to co-evolve with AI technology. Some firms abandon AI green projects under short-term pressures. Others persist through iteration and achieve systemic transformation.

Based on the above analysis, we propose a hypothesis:

H2c. 

Managerial long-termism plays a positive moderating role in the relationship between artificial intelligence and green innovation.

The theoretical model is shown in Figure 1. The ultimate goal of green innovation, the long-term value creation, will be further analyzed and studied.

3. Research Design

3.1. Sample Selection and Data Sources

This article selects a sample of A-share listed companies from 2012 to 2024, and data related to artificial intelligence and managerial cognition comes from the annual reports of listed companies. The green innovation data is sourced from the Chinese Research Data Services (CNRDS) database, while other data is sourced from the Chinese Research Data Services (CNRDS) database and the China Stock Market and Accounting Research Database (CSMAR). Process the data as follows: (1) Remove the sample of financial industry companies. (2) Exclude the ST company sample for the current year. (3) Manually fill in missing data by searching annual reports, and remove samples that cannot be filled in with missing values. After the above processing, integrate data from various sources into a unified research database for subsequent analysis and modeling. A total of 20,383 observation values are obtained, and Stata 17 is used for data processing and analysis. To avoid the impact of outliers or extreme values on the test results, the continuous variables are truncated at the 1% level. Through these detailed data collection and processing steps, this study ensures the high quality and reliability of the data, laying a relatively solid foundation for exploring the relationship between artificial intelligence and green innovation.

3.2. Variable Definition and Measurement

The definitions and measurements of each variable in this study are as detailed below.

3.2.1. Artificial Intelligence

According to the Organisation for Economic Co-operation and Development (2019), artificial intelligence is defined as a suite of technologies that utilize computer systems to simulate human intelligence [41]. This encompasses data analysis, pattern recognition, machine learning, and automated decision-making, aimed at automating or enhancing business processes, optimizing resource allocation, aiding strategic decision-making, and ultimately improving organizational efficiency and innovation capabilities. To accurately measure the degree of artificial intelligence in a large sample and overcome the problems of the “0/1” dummy variable not being able to reflect the intensity and progress of artificial intelligence of listed companies [42], as well as the challenges associated with the small sample size and limited generalizability of questionnaire surveys [43], this paper refers to existing research [44,45] and chooses to use text analysis to measure the degree of artificial intelligence, thereby improving data accuracy and result relevance.

Specifically, the initial artificial intelligence lexicon was developed based on terms drawn from the China Artificial Intelligence Industry Analysis Report 2020, which supplied foundational terminology. To broaden its coverage, the lexicon was expanded through synonym suggestions from deep learning recommendations. Following existing research [46,47], the lexicon was further supplemented. A total of 73 keywords for artificial intelligence were determined. Based on the natural language processing, sentences that do not describe the listed companies were excluded from the text mining. Word frequency count was conducted to construct the artificial intelligence variables. Meanwhile, as annual reports are an important means for listed companies to disclose public information to the public, and the annual reports of listed companies are audited by auditors and disclosed, they can accurately and comprehensively reflect the actual situation of the enterprise in the year. The use of this method to measure the level of artificial intelligence has strong objectivity and high credibility. In addition, due to the right-skewed distribution characteristics of such data, this article performs logarithmic processing on word frequency to measure artificial intelligence.

3.2.2. Green Innovation

This paper measures green innovation by the proportion of green patent applications to total patent applications. A high ratio suggests that firms integrate green innovation into their core strategy rather than merely engaging in superficial compliance to meet external environmental pressures. Compared to the absolute number of green patent applications, this ratio better captures the greening intensity of the innovation structure and more clearly reflects the green orientation of a firm’s innovative activities. By focusing on the optimization of innovation structure, this measure avoids biases stemming from firm size or overall R&D scale, helps control for unobserved confounding factors [48], and thus mitigates potential endogeneity concerns in empirical analysis. In the subsequent analysis, two alternative measures are used for robustness checks: the ratio of green invention patent applications to total invention patent applications (GI_inv), and the ratio of green utility model patent applications to total utility model patent applications (GI_uti).

3.2.3. Executive Cognition

(1) Managerial green cognition

The Management’s Discussion and Analysis (MD&A) is a core section of the annual report of listed companies, authored by management to explain the company’s financial position, operating results, future risks, and strategic outlook. The language choices, thematic focus, and narrative framework directly reflect management’s key areas of attention, value judgments, and strategic intent [44]. Therefore, the frequency of related terms such as “green,” “sustainable,” “environmental protection,” “low-carbon,” “sustainable,” “emission reduction,” “ecological,” “clean production,” “eco-friendly,” “carbon neutrality” and “low-carbon” can effectively serve as an indirect yet objective proxy for whether management prioritizes environmental issues strategically. By calculating the ratio of “the number of green-related words/the total word count of the MD&A,” the resulting proportion standardizes the measurement, enabling comparability of MD&A texts across different companies and years. This metric essentially measures the “share of attention” allocated by management to environmental issues within the limited space of strategic communication. A higher proportion indicates that environmental issues hold a more important position in management’s overall strategic thinking. Therefore, managerial green cognition (MGC) is measured in this paper as the ratio of managerial green cognition word frequency to the total word count in the MD&A section.

(2) Managerial innovation cognition

Following the same rationale used to construct the variable for managerial green cognition, managerial cognitive structure influences their linguistic framework. When management holds a strong innovation-oriented mindset, managers are more inclined to use innovation-related concepts, narratives, and vocabulary in their communication to shape internal and external strategic expectations. Therefore, the frequency of innovation-related terms in the MD&A, such as “innovation,” “R&D,” “technology development,” “new product,” “patent,” “technological breakthrough,” “originality,” “indigenous innovation,” and “technology iteration”, can serve as an effective textual proxy for their intrinsic innovation cognition. Managerial innovation cognition (MIC) is measured in this paper as the ratio of managerial innovation cognition word frequency to the total word count in the MD&A section.

(3) Managerial long-termism

Following earlier studies [33], managerial long-termism (MLT) is measured in this paper as the ratio of long-termism-related word frequency to the total word count in the MD&A section. The keywords representing managerial long-termism include “long-term,” “long-run,” “sustainable development,” “enduring,” and “strategic patience,” among others.

3.2.4. Control Variables

In terms of selecting control variables, this article controlled for enterprise size (SIZ), financial leverage (LEV), growth rate (GRO), return on assets (ROA), R&D spending (R&D), whether the board chair also serves as CEO (DUA), number of executives (MS), equity concentration (TOP), cash holdings (CAS), and company age (AGE). In addition, the model further controls for year fixed effects (YEA) and industry fixed effects (IND).

The definition and calculation of specific variables are shown in

Table 1. Measurement methods.

Classification	Variables	Symbols	Measurement Methods
Dependent variable	Green Innovation	GI	The proportion of green patent applications to total patent applications
Independent variable	Artificial Intelligence	AI	In the annual reports of listed companies, add 1 to the number of artificial intelligence keywords and take the natural logarithm
Interaction variables	Managerial Green Cognition	MGC	In the MD&A section of the annual report of listed companies, add 1 to the number of green cognition keywords for managers and take the natural logarithm
	Managerial Innovation Cognition	MIC	In the MD&A section of the annual report of listed companies, add 1 to the number of innovation cognition keywords for managers and take the natural logarithm
	Managerial Long-termism	MLT	In the MD&A section of the annual report of listed companies, add 1 to the number of long-termist keywords for managers and take the natural logarithm
Control variables	Firm Size	SIZ	Natural logarithm of the logarithm of the company’s total assets at the end of the year
	Financial Leverage	LEV	Total liabilities over total assets
	Company Growth	GRO	Revenue growth rate
	Return on Assets	ROA	Ratio of company’s net income to total assets
	Dual Employment	DUA	If the CEO is also the board chair, it is assigned a value of 1; otherwise, it is 0
	Management Size	MS	The number of top managers
	Top Equity Concentration	TOP	Shareholding ratio of the largest shareholder
	Cash Holdings	CAS	(Cash + net short-term investment)/total assets
	Company Age	AGE	Subtract the year the company went public from the current year
	Year	YEA	Set time dummy variables by year
	Industry	IND	Set industry dummy variables according to the industry classification standard of the China Securities Regulatory Commission (2012)

.

3.3. Research Model

This article constructs an OLS multiple linear regression model to examine the impact of artificial intelligence on green innovation and conducts a mechanism test. Firstly, the benchmark model is constructed as follows:

$$\mathrm{GIi},\mathrm{t}=\mathrm{\beta }0+\mathrm{\beta }1\mathrm{AIi},\mathrm{t}+\mathrm{\gamma }\mathrm{jCONi},\mathrm{t}+\mathrm{IND}+\mathrm{YEA}+\mathrm{\epsilon }\mathrm{i},\mathrm{t}$$

(1)

where i represents the enterprise, and t represents the year; CON is the control variable defined above; IND is an industry dummy variable; YEA is the annual dummy variable, which is a constant term, and it is the regression coefficient; and ε for perturbation terms.

This article built Model (2) to test the interaction effects of managerial cognition:

$$\mathrm{GIi},\mathrm{t}=\mathrm{\beta }0+\mathrm{\beta }1\mathrm{AIi},\mathrm{t}+\mathrm{\beta }2\mathrm{interactioni},\mathrm{t}+\mathrm{\beta }3\mathrm{AI}\ast \mathrm{interactioni},\mathrm{t}+\mathrm{\gamma }\mathrm{jCONi},\mathrm{t}+\mathrm{IND}+\mathrm{YEA}+\mathrm{\epsilon }\mathrm{i},\mathrm{t}$$

(2)

where interaction represents managerial green cognition, managerial innovation cognition, and managerial long-termism, and CON represents the control variable.

4. Empirical Results and Analysis

4.1. Descriptive Statistics and Correlation Analysis

The results of the main variables’ descriptive statistics are shown in

Table 2. Descriptive statistics of variables.

Variables	Observations	Mean	Std. Dev.	Min	Max
GI	20,383	0.052	0.142	0.000	1
AI	20,383	0.873	1.191	0	6.277
MGC	20,383	1.053	0.903	0.000	5.994
MIC	20,383	4.698	0.875	0.000	7.644
MLT	20,383	3.730	0.535	0.000	7.114
SIZ	20,383	22.222	1.340	14.941	28.791
LEV	20,383	0.425	0.207	0.0320	1.411
GRO	20,383	0.175	0.410	−0.657	3.588
ROA	20,383	0.038	0.067	−0.533	0.231
DUA	20,383	0.283	0.450	0.000	1.000
MS	20,383	6.410	2.327	2.000	16.000
TOP	20,383	0.342	0.147	0.080	0.758
CAS	20,383	0.181	0.128	0.008	0.765
AGE	20,383	17.650	5.711	3.000	35.000

. The minimum value of AI is 0, and the maximum value is 6.2777, indicating that there is a significant difference in the current AI intensity among Chinese listed companies. The standard deviation exceeding the mean further implies a right-skewed distribution, where a subset of firms has achieved relatively sophisticated AI integration, while the majority remain at early adoption stages. For green innovation, its minimum and maximum green innovation are 0 and 1, indicating differences in different enterprises; the average value is 0.052, indicating a low probability of enterprises to persist with green innovation. The managerial cognition variables similarly exhibit meaningful variation. Managerial green cognition (MGC) ranges from 0 to 5.994, indicating heterogeneous attention to environmental issues among top executives. Managerial innovation cognition (MIC) shows the highest mean value at 4.698, suggesting that innovation-oriented rhetoric is relatively prevalent in managerial discourse, though the maximum of 7.644 reveals considerable room for variation. Managerial long-termism (MLT) ranges from 0 to 7.114, with the relatively lower standard deviation indicating more uniform temporal orientations compared to other cognitive dimensions.

The correlation analysis of the main variables is shown in

Table 3. Correlation analysis of main variables.

Variables	GI	AI	MGC	MIC	MLT
GI	1.000
AI	0.071 ***	1.000
MGC	0.123 *	0.127 **	1.000
MIC	0.083 ***	0.329 ***	0.147 ***	1.000
MLT	0.076 ***	0.281 ***	0.137 ***	0.487 ***	1.000

Note: The correlation coefficient is tested by Pearson’s test, where * indicates significance in the two-tailed test at the 10% level. ** Indicates significance in the two-tailed test at the 5% level. *** Indicates significance in the two-tailed test at the 1% level.

. The correlation coefficients between AI and GI are significantly positive, which preliminarily proves Hypothesis 1. The correlation coefficients between the main variables are small, indicating a lower likelihood of multicollinearity.

4.2. The Impact of Artificial Intelligence on Green Innovation

4.2.1. The Main Effect Test of Artificial Intelligence on Green Innovation

The regression results of artificial intelligence on green innovation are shown in Model 1 in

Table 4. Regression results of main effect tests.

Variables	GI	GI
	Model 1	Model 2
AI	0.009 ***	0.009 ***
	(15.74)	(15.5)
SIZ		0.003 ***
		(4.34)
LEV		0.035 ***
		(8.22)
GRO		−0.002
		(−0.32)
ROA		0.046 ***
		(4.29)
DUA		−0.004 **
		(−2.44)
MS		0.002 ***
		(6.44)
TOP		−0.015 ***
		(−3.26)
CAS		−0.047 ***
		(−4.5)
AGE		−0.026 ***
		(−10.4)
YEA	Yes	Yes
IND	Yes	Yes
Cons	0.0439 ***	0.041 ***
	(52.7)	(2.82)
N	20,838	21,607
AdjR2	0.0055	0.012

Note: **, and ***, respectively, indicate significance at the 5%, and 1% levels under the t-test; T-values is in parentheses; the same below.

. Model 1 tests the relationship between AI and green innovation. After controlling for the influence of other variables, the coefficient of AI is also significantly positive, which confirms Hypothesis 1. At the same time, multicollinearity tests were conducted on Models 1 and 2, and the average VIF of the models was 4.24. The VIF of the main variables was lower than 10, indicating that there is no multicollinearity problem in the model.

4.2.2. Interaction Effect Analysis of Managerial Cognition

Models 1 to 3 in

Table 5. Regression results of interaction effect tests.

Variables	GI
	(1)	(2)	(3)	(4)	(5)	(6)
AI	0.011 ***	0.013 ***	0.007 ***	0.010 **	0.007 ***	0.023 ***
	(17.97)	(15.24)	(11.68)	(2.83)	(10.67)	(4.93)
MGC	0.021 ***	0.023 ***
	(26.32)	(23.84)
AI*MCG		0.003 ***
		(3.51)
MIC			0.010 ***	0.010 ***
			(10.95)	(9.95)
AI*MIC				0.005 ***
				(3.85)
MLT					0.0144 ***	0.017 ***
					(10.08)	(10.42)
AI*MLT						0.004 ***
						(3.49)
YEA	Yes	Yes	Yes	Yes	Yes	Yes
IND	Yes	Yes	Yes	Yes	Yes	Yes
Cons	0.089 ***	0.085 ***	0.012	0.011	−0.003	−0.013
	(5.86)	(5.74)	(0.83)	(0.73)	(−0.19)	(−0.81)
N	20,383	20,383	20,383	20,383	20,383	20,383
AdjR2	0.029	0.029	0.015	0.015	0.015	0.015

Note: ** and *** denote significance at the 5% and 1% levels under the t-test, respectively. To save space, the regression results of the control variables are omitted in the table, the same as below.

report the interaction effect testing results of managerial cognition. Columns (1)–(2), (3)–(4), and (5)–(6) report the interaction effects of managerial green cognition, managerial innovation cognition, and managerial long-termism on the relationship between AI and green innovation, respectively. As shown in the table, Hypothesis 2a, Hypothesis 2b, and Hypothesis 2b are supported.

4.3. Endogeneity and Robustness Testing

4.3.1. Instrumental Variable Method (IV)

The model in this article may have endogeneity issues caused by omitted variables. In view of this, for AI, the mean AI level in the same industry in the region where the enterprise is located (MAI) is selected as the instrumental variable. Two-stage regression is performed on the original model by setting instrumental variables. As shown in

Table 6. Endogeneity testing.

Variables	GI	GI
	IV	PSM
AI	0.087 *	0.160 ***
	(2.04)	(3.39)
CON	Yes	Yes
YEA	Yes	Yes
IND	Yes	Yes
Cons	−2.224 ***	−2.126 ***
	(−5.69)	(−6.04)
N	21,607	21,607
AdjR2/Wald chi2	0.362	0.344

Note: * and *** denote significance at the 10% and 1% levels under the t-test, respectively.

, Hypothesis 1 is further supported.

4.3.2. Propensity Score Matching Method (PSM)

To minimize endogeneity issues in the model, this article continues to apply the propensity score matching method to test the main model. Firstly, a PSM test was conducted on AI. The samples were divided into two groups based on AI, and 1:1 nearest neighbor matching was performed. After matching, the standardized deviation of all covariates was of less than 10%, indicating good matching quality. Regression analysis was conducted using the matched samples, and the regression results are shown in Table 6. It was found that the coefficient is significantly positive. Hypothesis 1 is supported.

4.3.3. Lag Explanatory Variable

To further verify the robustness of the conclusion, this article takes the T + 1 period value (FGI) for green innovation, and the explanatory variable and control variable take the T period value for multiple regression. The results show that the coefficients are significantly positive, confirming Hypothesis 1.

4.3.4. Replace Explanatory Variables

Furthermore, following the alternative measurement approach detailed in Section 3.2.2, we conduct robustness checks by employing disaggregated indicators of green innovation to verify that our findings are not sensitive to specific measurement choices. Specifically, we distinguish between green invention patents (GI_inv)—which represent high-quality, substantive technological breakthroughs subject to strict substantive examination—and green utility models (GI_uti)—which capture incremental improvements with lower inventive steps. By replacing the baseline aggregate measure, the regression results across both specifications consistently yield significantly positive coefficients on AI (see

Table 7. Robustness test.

Variables	FGI	GI_inv	GI_uti
AI	0.064 ***	0.103 **	0.105 **
	(16.32)	(2.96)	(2.67)
CON	Yes	Yes	Yes
YEA	Yes	Yes	Yes
IND	Yes	Yes	Yes
Cons	−2.073 ***	−6.537 ***	−6.496 ***
	(−4.23)	(−3.99)	(−3.97)
N	18,063	21,607	21,607
AdjR2	0.394	0.384	0.369

Note: ** and *** denote significance at the 5% and 1% levels under the t-test, respectively.

), thereby validating Hypothesis 1.

5. Further Analysis

5.1. Analysis of the Mechanisms Through Which Artificial Intelligence Affects Green Innovation

This paper attempts to explore the intermediate transmission process through which artificial intelligence empowers green innovation, aiming to reveal the critical pathways through which artificial intelligence translates into green innovation outcomes. Existing research generally overlooks the specific mechanisms through which AI reshapes corporate green innovation. In this section, we explore the potential mechanisms through which AI may influence green innovation. Rather than conducting formal mediation analysis—which would require establishing causality from AI to mechanisms and from mechanisms to green innovation—we examine whether AI is associated with three theoretically motivated channel variables. This approach aligns with our broader positioning of this study as an analysis of associational relationships and mechanism exploration.

This paper posits the existence of three mechanisms: information transparency, compliance internalization, and value creation. First, the information transparency mechanism corresponds to how enterprises respond to external pressure transmission. AI-enabled transparency of carbon data across the entire value chain transforms ambiguous environmental impacts into quantifiable, traceable, and attributable management indicators, thereby compressing the space for “greenwashing” and compelling resources to tilt toward genuine and effective green technology solutions. Second, the compliance internalization mechanism: Through dynamic compliance monitoring and simulation, AI preemptively embeds environmentally friendly solutions into daily operations and supply chain configurations, enabling enterprises to proactively identify compliance risks and reduce violations. This transforms green innovation from a passive cost of regulatory compliance into a proactive strategic investment. Third, the value creation mechanism. By mining multi-source data, AI enhances the ability to realize data value that traditional methods find difficult to break through. This shifts green innovation from “responsibility-oriented” to “competitiveness-oriented,” fundamentally solving the problem of insufficient motivation for green innovation and making it an endogenous choice for enterprises pursuing competitive advantage. The above three mechanisms clarify that AI is not merely a tool for substituting labor but a driving architecture that reconstructs corporate environmental strategy and innovation logic.

5.1.1. Information Transparency Mechanism

AI can enhance the measurability, reportability, and verifiability of environmental data (particularly carbon emission information), reduce information asymmetry, and strengthen internal and external supervision and incentive mechanisms, thereby driving enterprises to internalize environmental externalities and stimulate green technology innovation and management innovation. This paper constructs the indicator WFreCy using the word frequency of carbon information disclosure in corporate social responsibility reports to test the information transparency mechanism of AI-enabled green innovation. The estimation results in Column (1) of

Table 8. Mechanism testing of the impact of AI on green innovation.

Variables	WFreCy	VioAct	Mvd
AI	0.006 **	−0.568 ***	0.012 ***
	(2.76)	(−5.97)	(4.69)
YEA	Yes	Yes	Yes
IND	Yes	Yes	Yes
Cons	−2.235 ***	−2.207 ***	−9.086 ***
	(−5.06)	(−8.03)	(−6.91)
N	21,607	21,607	21,607
AdjR2	0.399	0.399	0.415

Note: ** and *** denote significance at the 5% and 1% levels under the t-test, respectively.

verify that AI effectively improves the quality of corporate carbon emission data to a certain extent. Thus, it verifies that AI can meet increasingly stringent regulatory disclosure requirements by enhancing information transparency, forcing internal process optimization, and driving green innovation.

5.1.2. Compliance Internalization Mechanism

AI can strengthen enterprises’ intelligent early warning and decision-support capabilities for compliance risks by learning from massive regulatory texts, environmental standards, and historical penalty cases, thereby dynamically identifying environmental compliance risk points in production and operations and guiding enterprises to avoid violations, thus enhancing the internal compliance level of green innovation. This paper constructs the indicator VioAct using violation records of listed companies to test the compliance internalization mechanism of AI-enabled green innovation. The estimation results in Column (2) of

Table 9. AI, green innovation, and corporate value.

	TBQ_1	TBQ_2	TBQ_1	TBQ_2
AI	0.112 ***	0.114 ***	0.066 ***	0.068 ***
	(9.06)	(10.23)	(9.97)	(9.64)
GI		0.050 ***		0.031 ***
		(0.010)		(0.011)
CON	Yes	Yes	Yes	Yes
YEA	Yes	Yes	Yes	Yes
IND	Yes	Yes	Yes	Yes
Cons	−8.716 ***	−5.965 ***	9.538 ***	−9.653 ***
	(−8.52)	(−11.74)	(6.27)	(−11.14)
N	17,319	17,319	14,580	14,580
AdjR2	0.321	0.322	0.326	0.327

Note: *** denote significance at the 1% levels under the t-test.

verify that AI effectively reduces corporate violations to a certain extent. Thus, it verifies that AI can transform environmental regulatory requirements into executable and monitorable digital compliance systems by reducing corporate violations, prompting enterprises to systematically reduce violation risks through green process innovation and workflow reengineering, and achieving sustainable operations.

5.1.3. Value Creation Mechanism

AI can identify potential green efficiency improvement points and emerging market opportunities by mining and integrating multi-source environmental data, and promote corporate green innovation by enhancing the ability to realize data value, transforming environmental data from management costs into innovation elements and profit sources, thereby incentivizing enterprises to develop green business model innovations. This paper constructs the indicator Mvd using the word frequency of data value realization in annual reports to test the value creation mechanism of AI-enabled green innovation. The estimation results in Column (3) of Table 9 verify that AI effectively enhances corporate data value realization capabilities to a certain extent. Thus, it verifies that AI, through deep mining and intelligent analysis of environmental data, transforms environmental performance into economic value, thereby becoming an intrinsic driving force for the development of corporate green innovation.

5.2. Artificial Intelligence, Green Innovation, and Enterprise Value

Building upon the preceding evidence, we further examine whether AI-driven green innovation translates into tangible long-term value appreciation, thereby completing the causal chain from technological deployment to sustainable competitive advantage. Given that green innovation entails substantial upfront investments, extended gestation periods, and uncertain short-term returns, verifying its ultimate economic consequences is essential for justifying resource commitment. To address this inquiry, this study investigates whether AI enhances enterprise value through the mediating channel of green innovation, effectively bridging the gap between environmental strategy and long-term financial performance. This article uses Tobin’Q (TBQ) as the proxy variable for enterprise value, which is widely used to measure the market value of enterprises. Due to the time required for green innovation to affect the realization of enterprise value, the dependent variable is subjected to lag one period (TBQ_1) and lag two periods (TBQ_2) treatment, and the test results are shown in Table 9. As can be seen, AI can promote green innovation, thereby enhancing the long-term value.

5.3. Heterogeneity Testing of AI in Promoting Green Innovation

5.3.1. Heterogeneity Test Based on Different Industries

This article verifies whether a listed company is a high-tech enterprise through the qualification recognition of high-tech enterprises, and tests the impact of artificial intelligence on green innovation in groups. The grouping test results are shown in

Table 10. Heterogeneity test results.

	GI
	(1)		(2)
	High Tech Enterprises	Non-High Tech Enterprises	High Market Competitiveness	Low Market Competitiveness
AI	0.057 ***	0.063	0.077 ***	0.058
	(3.99)	(1.78)	(4.69)	(1.73)
CON	Yes	Yes	Yes	Yes
YEA	Yes	Yes	Yes	Yes
IND	Yes	Yes	Yes	Yes
Cons	−2.576 ***	−1.746 ***	−2.583 ***	−1.993 ***
	(−9.87)	(−10.23)	(−9.47)	(−9.79)
N	8976	9087	8872	9191
AdjR2	0.408	0.344	0.491	0.294
p-value of group difference	0.001 ***		0.055 **

Note: ** and *** denote significance at the 5% and 1% levels under the t-test, respectively.

. The green innovation of high-tech enterprises is significantly higher than that of non-high-tech enterprises. Moreover, artificial intelligence also has a more significant impact on non-high-tech enterprises.

5.3.2. Heterogeneity Testing Based on Different Market Competition

Intense market competition brings operational risks such as unstable cash flow and reduced overall profit margins to enterprises. Enterprises therefore place greater emphasis on change to improve the difficulties they face under high levels of competition. Successful artificial intelligence can enhance the competitive advantage of enterprises and is a necessary path for them to stand invincible in a dynamic competitive environment. In view of this, this article uses group testing to determine the situational role of market competition intensity on the impact of artificial intelligence on green innovation. The Herfindahl Index reflects the intensity of industry competition. This paper categorizes the sample by industry and divides it into two groups based on the high and low levels of market competition. The regression results are shown in Table 10. The results showed that artificial intelligence had a positive impact on green innovation in both high- and low-market-competition groups. However, in the high-market-competition group, artificial intelligence had a more significant positive impact on green innovation. Therefore, for enterprises with high market competition, it is more important to promote green innovation by improving artificial intelligence.

6. Research Conclusions and Insights

6.1. Conclusions

Based on socio-technical systems theory and upper echelons theory, this study systematically examines the impact of artificial intelligence (AI) technology on corporate green innovation and the boundary role of managerial cognition. Using panel data from Chinese A-share listed companies (2012–2024) and textual analysis, we find:

First, AI significantly promotes green innovation. As a general-purpose technology, AI drives green technology exploration through its self-learning capabilities and strong generality, extending green innovation drivers beyond institutional pressures to technological endogeneity.

Second, managerial cognition serves as a critical boundary condition. Managerial green cognition, innovation cognition, and long-termism all positively moderate the AI–green innovation nexus. High green cognition reframes AI as a strategic engine for sustainability; strong innovation cognition enables tolerance for uncertainty and absorptive capacity building; and long-termism provides strategic patience against short-term volatility.

Third, AI operates through three transmission mechanisms: information transparency (enhancing carbon data quality and compressing greenwashing space), compliance internalization (embedding environmental regulations into digital compliance systems), and value creation (converting environmental data into economic assets and green business models).

Fourth, significant industry heterogeneity exists. AI exhibits stronger marginal effects in non-high-tech industries and under intense market competition, offering leapfrogging opportunities to bridge green innovation gaps across technological divides.

6.2. Practical Implications

The findings of this study offer important implications for corporate managers, policymakers, and investors.

First, managers should transcend the narrow perception of AI as merely a tool for cost reduction and efficiency gains, instead establishing a cognitive framework that views AI as an architect of green strategy. Second, managers must consciously cultivate their green cognition, innovation cognition, and long-term orientation through proactive learning, cross-industry exchanges, and engagement with sustainability issues, thereby building the cognitive foundations for deeply integrating AI with green innovation. Specifically, they should incorporate green value dimensions into AI project evaluations, tolerate early-stage trial-and-error costs, and establish long-term incentive mechanisms aligned with AI technology iteration cycles.

Second, when promoting the “AI+” initiative, governments should fully account for industry heterogeneity. Policymakers should increase inclusive investment in AI infrastructure to lower technology adoption barriers. Simultaneously, they should improve ESG disclosure standards and green finance incentive mechanisms to provide institutional support for the formation of managerial long-termism. Furthermore, establishing demonstration case libraries of AI-enabled green transformation could guide corporate AI investment toward green innovation through “cognitive inspiration” rather than mere regulatory pressure.

Finally, when assessing corporate AI investment value, capital markets should look beyond technology input intensity and focus instead on management cognitive characteristics and strategic intent. Managers’ emphasis on green issues, innovation orientation, and long-termism expressed in MD&A sections can serve as effective signals for predicting whether AI technology will translate into substantive green innovation outcomes.

6.3. Limitations and Future Research

This study has several limitations that directly suggest avenues for future research. First, regarding measurement validity, while text-based word frequency indicators offer objectivity and reproducibility, they may inadequately capture the full depth, complexity, and strategic ambiguity of managerial cognition. To address this, future research could adopt mixed-method designs, integrating qualitative approaches such as in-depth interviews and longitudinal case studies with quantitative analysis to illuminate the micro-psychological and organizational processes through which managerial cognition shapes AI technology absorption.

Second, the sample is limited to Chinese A-share listed companies, which constrains generalizability to non-listed SMEs or different institutional environments. Future research could extend to unlisted firms and emerging economies to examine how organizational scale and ownership structure moderate the AI–green innovation relationship, and to validate whether managerial cognition operates similarly across distinct cultural and institutional settings.

Finally, given the rapid iteration of AI technology, our data spanning 2012–2024 may not fully reflect the disruptive impact of emerging technologies such as Generative AI. Moving forward, scholars should closely monitor how next-generation AI technologies—including large language models and digital twins—influence green innovation paradigms and correspondingly transform managerial cognitive frameworks. Furthermore, future work could trace the long-term consequences of AI-driven green innovation on firm value, supply chain resilience, and carbon neutrality pathways, thereby constructing a complete “technology—cognition—innovation—performance” causal chain.