Unpacking Boundary-Spanning Search and Green Innovation for Sustainability: The Role of AI Capabilities in the Chinese Manufacturing Industry

Abstract

Achieving the dual carbon goal and addressing escalating environmental challenges requires that manufacturing enterprises in China must pursue sustainability via green innovation strategies. A key rationale for green innovation is to overcome boundaries and acquire knowledge through boundary-spanning search. Additionally, leveraging artificial intelligence (AI) capabilities provides technical support throughout the innovation process. Thus, both boundary-spanning search and AI capabilities are crucial for achieving sustainability objectives. Drawing on organizational search and knowledge management theories, this paper aims to analyze how dual boundary-spanning search affects sustainability performance and green innovation. It also examines the moderating role of AI capabilities and constructs a moderated mediation model. We analyzed questionnaire data collected from 171 Chinese manufacturing companies over a 13-month period, employing hierarchical regression and bootstrap sampling methods using SPSS 27.0. Our findings reveal that both prospective and responsive boundary-spanning searches significantly enhance corporate sustainability performance. Furthermore, green innovation acts as a positive partial mediator between dual boundary-spanning search and corporate sustainability performance. Notably, AI capabilities positively moderate the relationship between dual boundary-spanning search and green innovation. They also strengthen the mediating effect of green innovation on the link between dual boundary-spanning search and corporate sustainability performance. Based on these findings, more resources should be allocated to boundary-spanning search while encouraging enterprises to pursue green innovation and develop AI capabilities. These efforts will provide robust support for sustainability performance in the manufacturing sector.

1. Introduction

The United Nations’ Sustainable Development Report 2024 indicates that global implementation of the 2030 Agenda for Sustainable Development is significantly lagging. Nearly 83% of the sustainable development goals (SDGs) are not on track in achieving their expected progress. The report underscores that if countries do not convert their verbal commitments into concrete interventions, achieving the SDGs will remain unlikely. Among the 167 countries and regions evaluated, China ranks 68th with a total score of 70.85. Despite some progress in achieving the SDGs, air pollution remains a critical challenge due to China’s rapid urbanization and industrialization. The Chinese government has acknowledged the existing challenges and made a firm commitment to achieve a peak in carbon emissions by 2030 and achieve carbon neutrality by 2060. As key stakeholders in environmental change, manufacturing enterprises must actively pursue green innovation opportunities to enhance their sustainability performance. Corporate sustainability performance refers to a company’s overall effectiveness in sustainable development, highlighting the balance between economic growth and ecological protection. Boundary-spanning search is a crucial aspect of organizational search behavior, essential for acquiring diverse knowledge necessary for green innovation. By fostering this process, enterprises can gain new momentum in enhancing their sustainability performance [1]. This paper categorizes boundary-spanning search into two dimensions: prospective and responsive search. These categories are defined by different search timings and objectives, aimed at clarifying their effects on sustainability. A review of the existing literature indicates that the majority of studies on boundary-spanning search focus on its link to innovation behavior [2,3] and innovation performance [4,5]. However, few studies examine its impact on corporate sustainability performance. Scholar Wang (2022) [6] has highlighted how essential boundary-spanning search is for the sustainability of science and technology start-ups. He suggests exploring the driving mechanisms of sustainability performance from the perspective of organizational learning [6]. However, his research overlooks the urgency for manufacturing enterprises to improve sustainability and fails to classify boundary-spanning search based on specific criteria. Therefore, it is essential to clarify how boundary-spanning search and its various dimensions affect corporate sustainability performance.

Motivated by the dual carbon goals, green innovation has become a strategic approach for enterprises to tackle environmental challenges and strengthen their green competitiveness. As a knowledge-intensive innovation activity, green innovation synergistically optimizes resource utilization while promoting environmental protection [7]. As a result, it serves as an essential method for enterprises to develop sustainable competitive advantages. In comparison to traditional innovation activities, the green innovation process is more cutting-edge and professional. Therefore, enterprises that adopt green innovation rely heavily on diverse knowledge resources [8,9], particularly in areas related to pollution control, energy conservation, and emission reduction [10]. According to the knowledge-based view, enterprises can create cross-organizational knowledge flow channels through effective boundary-spanning searches [11]. This approach allows them to obtain and integrate relevant technical resources and information. Ultimately, it helps overcome challenges in developing green products and processes, facilitating the accomplishment of sustainable development goals. However, few studies have examined the role of green innovation as a link between boundary-spanning search and sustainability performance. To acquire the complex knowledge necessary for green innovation, manufacturing enterprises must first overcome technological bottlenecks. They need to seek heterogeneous knowledge and novel technologies across various industries actively. Additionally, learning from peers’ experiences with green processes is crucial. Enhancing existing knowledge cooperation systems can further reduce the failure rate of green innovation. Thus, it is essential to clarify how boundary-spanning search impacts green innovation across different dimensions. Analyzing the mediation role of green innovation between dual boundary-spanning search and corporate sustainability performance is also important.

While boundary-spanning search is essential, the effective utilization of acquired knowledge by enterprises also relies on their internal capabilities. According to the “2024 China Artificial Intelligence Industry Research Report (VI)”, China’s core AI industry will reach CNY 269.7 billion by 2024. The AI industry is positioned to be a central catalyst for digital economy growth and sustainability [12,13]. AI capabilities refer to an enterprise’s ability to effectively leverage AI technology. In the digital intelligence era, Malik (2021) underscores the necessity for enterprises to fully harness AI capabilities to facilitate intelligent knowledge search and absorption [14]. Based on information processing theory, enterprises can gain real-time data analysis and efficient transmission by utilizing artificial intelligence and big data analytics [15]. This optimization enhances their information systems and effectively transforms knowledge obtained from boundary-spanning search. Higher AI capabilities correlate with a greater potential for boundary-spanning search to foster green innovation. While existing studies confirm the driving role of AI capabilities in open innovation [16] and breakthrough technological innovation [17], few have explored their impact on green innovation. Therefore, clarifying the moderating role of AI capabilities in the relationship between boundary-spanning search and green innovation is essential.

In summary, enterprises can enhance their sustainability performance by implementing boundary-spanning search, fostering green innovation, and developing AI capabilities. Thus, this paper addresses an existing research gap by applying organizational search theory and knowledge management theory. Currently, research on the interconnections between boundary-spanning search and corporate sustainability performance remains in its infancy. This is especially true when considering the differences among various types of boundary-spanning searches. Moreover, scholars have not yet investigated the effect of AI capabilities in green innovation. Specifically, this paper presents four primary objectives. The first objective focuses on exploring the significance of boundary-spanning search and its two dimensions, as well as evaluating their effects on sustainability performance. The second objective aims to probe the mediation role of green innovation within the context of boundary-spanning search and its resultant influence on sustainability performance. The third objective seeks to assess the moderating influence of AI capabilities on the interaction between boundary-spanning search and green innovation. Finally, this paper aims to determine whether AI capabilities strengthen the mediation effect of green innovation between boundary-spanning search and sustainability performance. This objective ultimately tests for the presence of a moderated mediation effect. After analyzing 171 manufacturing enterprises, we conclude that boundary-spanning search (including prospective and responsive boundary-spanning search) directly drives sustainability performance. Furthermore, green innovation partially mediates the relationship between them. Moreover, this paper finds that AI capabilities positively moderate the relationship between boundary-spanning search and green innovation. They also strengthen the mediating role of green innovation in the connection between boundary-spanning search and sustainability performance.

This paper makes marginal contributions in three main aspects: (1) It integrates boundary-spanning search, green innovation, corporate sustainability performance, and AI capabilities into a unified analytical framework. This integration enhances the understanding of the factors influencing sustainability performance. (2) This study innovatively decomposes boundary-spanning search into prospective and responsive dimensions. It systematically examines how these dimensions differentially impact green innovation and sustainability performance, addressing a gap in existing research. (3) AI capabilities are introduced in this paper as a moderating variable, providing new insights into enhancing green innovation efficiency and corporate sustainability performance. The focus of antecedents for promoting green innovation and sustainability has shifted from traditional digital capabilities centered on data management to AI capabilities. The research findings yield actionable insights for corporate management and government policy.

This paper is structured in the following manner: Firstly, we conduct a literature review and propose the research hypotheses, establishing a moderated mediation model. Secondly, the research design and sample sources are outlined, along with the questionnaire design and data collection process. Thirdly, we employ a series of scientific tests on the questionnaire data. This is followed by mechanism testing, which further validates the research hypotheses through robustness testing. Finally, we discuss the conclusions and provide management insights for businesses and governments. We also identify the limitations of this paper and suggest future research directions.

2. Theoretical Background and Hypothesis

2.1. Theoretical Background

Organizational search behavior refers to a problem-solving approach focused on the learning process of information retrieval. In complex and dynamic environments, organizations employ this process to identify opportunities and address problems, characterized by irreversibility and contingency. Consequently, organizational search has become foundational to organizational learning and behavior theory. The concept of boundary-spanning search was first introduced by Rosenkopf and Nerkar (2001) [18], who defined it as a strategic action based on organizational search theory. In this action, an enterprise actively transcends organizational or technological boundaries to seek, acquire, and integrate diverse external knowledge resources. Green innovation typically necessitates the integration of various types of knowledge and cross-domain applications [19]. Therefore, enterprises engage in boundary-spanning search to obtain innovative knowledge from diverse fields and industries. They also aim to obtain established environmental protection technologies held by a limited number of firms in the industry. The novelty and complexity of environmental protection technologies push manufacturing enterprises to surpass their competitors. They must expand their knowledge into unfamiliar areas to gather stakeholder needs and generate new ideas. At the same time, these enterprises need to monitor their competitors closely. They should gain a deep understanding of existing knowledge and technologies. They must also assimilate established practices and integrate these insights into their production and operational processes [20]. Boundary-spanning search is categorized into two dimensions: prospective and responsive, based on the timing and objectives of the search [21]. This framework enhances the understanding of the intricate relationship between boundary-spanning search and corporate sustainability performance. Prospective boundary-spanning search highlights the need for enterprises to seize market and technological opportunities. Companies must maintain a competitive edge. They should explore boundaries for cutting-edge and diverse knowledge resources that are still emerging but demonstrate high potential value [22]. Although these actions are challenging due to the difficulties of knowledge absorption, they can yield significant first-mover advantages. Responsive boundary-spanning search refers to companies reinforcing their existing position by monitoring competitors to acquire mature knowledge resources [23]. Their actions are reactive, and although knowledge gained from experiential learning carries a risk of depreciation [24], it can significantly enhance existing products or processes.

Sustainable development centers on fulfilling the requirements of today’s generation without hindering the capacity of future generations to fulfill their own needs. Ahi (2018) noted that enterprises must adhere to social and environmental standards while pursuing economic benefits [25]. They should also commit to the integrated development of economic, social, and environmental performance. Corporate sustainability performance is a comprehensive measure of organizational performance, influenced by various complex factors. The existing literature has identified significant antecedents of sustainability performance, with advances in digital technology [26,27] and green technology [28,29] being key drivers. To achieve sustainable development, enterprises must consistently pursue green innovation in their products and processes. This approach helps them meet environmental and social requirements while maintaining economic benefits [30]. Knowledge management theory emphasizes the critical importance of searching, integrating, and sharing knowledge for corporate innovation. The sustainable development capacity of enterprises, particularly their green innovation performance, relies heavily on effective knowledge management practices. Green innovation involves significant uncertainty, and enterprises frequently encounter unpredictable situations. The ability to extract applicable knowledge in a timely and effective manner directly influences the output of green innovation [31]. Thus, enterprises must build their knowledge management capabilities for knowledge acquisition and utilization. This will help them navigate complex business environments and ultimately enhance their sustainability performance.

Information processing theory emphasizes how organizations acquire, process, and utilize information to support decision-making and innovation. This aligns closely with the core function of AI capabilities; AI is fundamentally a tool for processing information. AI capabilities include data analysis, understanding, learning, reasoning [32], and the ability to coordinate resources in a changing environment to tackle complex problems [33]. Tarafdar (2019) emphasizes that the essence of AI algorithms lies in providing tentative solutions based on probabilistic predictions [34]. The realization of their value depends on how well enterprises construct and apply this technology. Scholar Halbusi (2025) [35] notes that AI, as an emerging technology, has excellent potential. It can help enterprises navigate uncertainties in green innovation and boost the efficiency of knowledge discovery [35]. In unfamiliar areas, AI capabilities can identify potential patterns and predict market trends by actively exploring cross-disciplinary knowledge [36], thereby providing managers with valuable insights. Furthermore, AI-generated exploratory solutions significantly reduce the risks and costs associated with exploring uncertain areas. Such risk mitigation encourages companies to transcend organizational boundaries, enabling them to seize emerging opportunities and foster disruptive green innovation. In established fields, AI capabilities excel at analyzing predictable problems using existing data. For example, AI can quickly respond to explicit demands by automatically screening large volumes of creative ideas and optimizing resource allocation processes [37]. This capability to efficiently execute systematic tasks, such as IBM Watson’s insight mining, significantly enhances the accuracy and efficiency of utilizing existing knowledge. Research indicates that the lack of AI capabilities is currently the most significant challenge for enterprises adopting AI [33,38]. This deficit directly affects their green innovation performance and learning effectiveness.

2.2. The Impact of Boundary-Spanning Search on Corporate Sustainability Performance

Organizational search theory explains how boundary-spanning search expands the ways manufacturing enterprises can obtain resources from different value chain networks [18]. This approach offers innovative strategies for enhancing sustainability performance in resource-scarce environments. Garetti (2012) highlighted that achieving sustainable development goals necessitates the use of diverse resources [39]. However, manufacturing enterprises often struggle to fulfill these requirements using their internal resource reserves. Consequently, efforts to improve resource stock and application capabilities through boundary-spanning search have become increasingly crucial. This paper contends that the role contribution of boundary-spanning search to enhance corporate sustainability performance can be outlined in two key aspects.

First, proactive boundary-spanning search enhances sustainability performance by more effectively identifying and mitigating risks. This proactive search strategy enhances opportunities for companies to engage with key resource owners [40], enabling them to manage both long-term and potential risk factors effectively. Sustainable development represents a long-term strategic orientation for enterprises. The risks associated with this process primarily stem from environmental changes and public opinion [41]. Enterprises that possess proactive boundary-spanning search capabilities are more effective at identifying environmental risks that threaten corporate profitability and green initiatives. By recognizing these risks, companies can implement effective mitigation measures. Additionally, actively perceiving and responding to the public’s changing evaluations of themselves and their competitors can motivate companies to embrace social responsibility. This approach helps them continuously improve their sustainability performance while fostering a positive social image.

Second, responsive boundary-spanning search can lead to substantial improvements in product and service quality. When industry benchmarks are referenced during boundary-spanning search, differentiated resources can be acquired to address the needs of diverse customer segments. Compared to prospective boundary-spanning search, responsive boundary-spanning search has well-defined objectives. This clarity enables companies to rapidly translate the knowledge gained into improvements in product performance and increased sales. It is essential to note that, in terms of environmental governance, manufacturing enterprises encounter greater social expectations than commercial enterprises [42]. Responsive boundary-spanning search allows companies to acquire insights from their competitors’ strengths while purposefully refining and enhancing their existing knowledge systems. This search strategy optimizes business processes and minimizes potential trial-and-error costs. It reduces unnecessary consumption during economic growth and achieves a closed loop of sustainable development. Duan (2023) noted that conducting two forms of boundary-spanning search simultaneously can prevent companies from falling into the trap of merely over-relying on new knowledge or existing knowledge [43]. Therefore, companies are likely to enhance their sustainability performance through dual boundary-spanning search. Consequently, this paper introduces the following hypotheses:

H1a. 

Prospective boundary-spanning search positively impacts corporate sustainability performance;

H1b. 

Responsive boundary-spanning search positively impacts corporate sustainability performance.

2.3. The Impact of Boundary-Spanning Search on Green Innovation

Green innovation activities fundamentally aim to achieve technological breakthroughs by searching for and integrating diverse and novel knowledge resources [19]. Enterprises’ boundary-spanning search behaviors facilitate the transfer and recombination of green knowledge across regions and industries; thus, an essential knowledge base for green innovation is established [44]. The significance of boundary-spanning search in promoting green innovation is mainly reflected in two key aspects.

First, prospective boundary-spanning search can enable companies to gain a first-mover advantage in green innovation. By adopting this strategy, companies can outpace their competitors in identifying potential customer needs and be the first to extract valuable new knowledge from the market. Moreover, by leading in the acquisition of new knowledge and technologies in emerging fields, companies can achieve technological advances that are hard to replicate. This approach limits competitors’ opportunities to imitate them. In summary, through prospective boundary-spanning search, companies can position themselves at the forefront of energy-saving technology development and environmental protection practices, leading to breakthrough green innovations.

Second, a responsive boundary-spanning search can help enterprises form synergies for green innovation. Yang (2021) noted that responsive boundary-spanning search strategies can enhance the efficiency of information exchange and circulation within enterprises [21]. As knowledge sharing increases, internal cohesion within organizations strengthens, thereby creating the necessary conditions for collaborative green innovation. Employees from different departments are more likely to collaborate when they are familiar with existing green knowledge and skills. They can also provide innovative suggestions for improvement. Furthermore, the responsive boundary-spanning search enhances green innovation ideas and solutions. This approach leverages the knowledge of peers, benefiting from the spillover effect of knowledge related to green innovation. By observing the practices of peer companies, organizations can learn about effective marketing models and risk avoidance strategies in green innovation. This strategy aids in lowering costs related to green product research and development, as well as trial marketing, ultimately leading to a higher success rate for green innovation. Therefore, companies will likely promote green innovation through dual boundary-spanning search. Consequently, this paper presents the following hypotheses:

H2a. 

Prospective boundary-spanning search positively impacts green innovation.

H2b. 

Responsive boundary-spanning search positively impacts green innovation.

2.4. Mediation Effects of Green Innovation

To foster sustainability, enterprises must consistently enhance their green competitiveness [45]. Green innovation is vital for enabling companies to tackle environmental, social, and economic challenges. These innovation activities utilize knowledge management in areas such as pollution reduction and energy conservation. They convert essential technological resources and diverse knowledge gained from boundary-spanning searches into practical, environmentally friendly products and processes. Therefore, green innovation can greatly enhance corporate sustainability performance. The mediating mechanism of green innovation between dual boundary-spanning search and corporate sustainability performance is primarily evident in two aspects.

First, prospective boundary-spanning emphasizes exploring new knowledge and developing new products [22]. This approach can disrupt reliance on traditional learning methods and innovation paths in corporate green innovation, providing long-term support for sustainable development. Through prospective boundary-spanning, more flexible learning modes and innovative problem-solving strategies can be developed by enterprises. This strategy stimulates creative thinking among employees and reduces path dependence associated with traditional green innovation. Additionally, prospective boundary-spanning is crucial in overcoming the challenge of green innovation homogenization, as it encourages companies to develop diverse eco-friendly products and services. The green attributes of these products address consumers’ increasing environmental awareness while also satisfying their diverse consumption preferences. As scholar Sarkar (2013) emphasized, the long-term momentum for sustainable development can only be provided by creating products that are both environmentally friendly and functionally diverse [44].

Second, responsive boundary-spanning search emphasizes enhancing existing knowledge and products. This strategy enables companies engaged in green innovation to share knowledge and upgrade their skills, thereby helping to reduce resource consumption during sustainable development. By utilizing responsive boundary-spanning search, companies can acquire mature knowledge validated by their competitors. This process enhances the efficiency of green innovation and significantly lowers associated costs [46]. Additionally, this search strategy increases the company’s overall awareness of environmental protection and sustainable development by sharing expertise and improving employee skills [47]. This deeper understanding encourages employees to voluntarily adopt resource conservation practices in their production activities, leading to greener and more eco-friendly products. To achieve sustainability, manufacturing corporations must not only develop and adopt advanced green technologies but also optimize their existing technologies [48] to improve resource and waste utilization efficiency. Thus, green innovation is likely to serve as a “bridge” between dual boundary-spanning search and enhanced corporate sustainability performance. Consequently, this paper presents the following hypotheses:

H3a. 

Green innovation partially mediates the relationship between prospective boundary-spanning search and corporate sustainability performance;

H3b. 

Green innovation partially mediates the relationship between responsive boundary-spanning search and corporate sustainability performance.

2.5. Moderating Effects of AI Capabilities

Information processing theory suggests that AI capabilities have significant potential to assist enterprises in managing the risks linked to green innovation [49]. The complexity of environmental protection technologies, along with the fluctuating demand for green products, raises the risks involved in green innovation activities. AI capabilities offer technical support for boundary-spanning search, actively predict market trends, and optimize knowledge processing flows. Therefore, as AI capabilities continue to increase, the effect of dual boundary-spanning search in enhancing green innovation becomes increasingly significant. The moderating role of AI capabilities in dual boundary-spanning searches and green innovation is primarily evident in two key dimensions.

First, AI capabilities positively influence the relationship between prospective boundary-spanning search and green innovation. Specifically, when companies have strong AI capabilities, they can track consumer discussions and purchasing behaviors in real time. Such AI-driven insights allow them to effectively identify consumers’ actual green needs and provide practical solutions for green innovation outcomes [50]. In other words, higher levels of AI capabilities amplify the beneficial influence of prospective boundary-spanning search on green innovation. Conversely, companies with limited AI capabilities struggle to utilize large models and channels, such as internet platforms, for interaction and collaboration with external entities. This limitation hinders their ability to overcome knowledge barriers. Additionally, when confronted with knowledge obtained through prospective boundary-spanning search, enterprises face challenges in identifying valuable information quickly. This difficulty undoubtedly slows down the progress of their green innovation activities.

Second, AI capabilities positively influence the relationship between responsive boundary-spanning search and green innovation. When enterprises possess strong AI capabilities, they can leverage knowledge gained from this search strategy to engage in iterative learning [51]. This ability allows them to complete complex yet predictable tasks independently [52]. As a result, the efficiency of knowledge flow within the organization improves, and new insights into existing knowledge emerge. In other words, higher levels of AI capabilities can lower the costs related to green innovation while simultaneously enhancing the competitiveness of its outcomes. Conversely, low AI capabilities may hinder managers from avoiding subjective interference and making objective, efficient decisions [53]. Such inefficient and low-intelligence knowledge management and decision-making systems cannot effectively control green innovation. This limitation hinders the optimization and advancement of green innovation processes. Consequently, this paper presents the following hypotheses:

H4a. 

AI capabilities positively moderate the relationship between prospective boundary-spanning search and green innovation;

H4b. 

AI capabilities positively moderate the relationship between responsive boundary-spanning search and green innovation.

Hypotheses H3a and H4a suggest that AI capabilities can impact the mediation effect of green innovation on prospective boundary-spanning search and corporate sustainability performance. By developing AI capabilities, enterprises can optimize knowledge retrieval and data utilization processes. This enhancement promotes green innovation through enhanced prospective boundary-spanning search, ultimately leading to improved corporate sustainability performance. Additionally, hypotheses H3b and H4b indicate that AI capabilities can impact the mediation effect of green innovation on responsive boundary-spanning search and corporate sustainability performance. The application of AI technology enhances the effectiveness of knowledge integration, which reduces the costs associated with green innovation. This is achieved by strengthening responsive boundary-spanning searches, ultimately leading to improved corporate sustainability performance. Consequently, this paper presents the following hypotheses:

H5a. 

AI capabilities influence the relationship between prospective boundary-spanning search and corporate sustainability performance by positively moderating the partial mediating effect of green innovation;

H5b. 

AI capabilities influence the relationship between responsive boundary-spanning search and corporate sustainability performance by positively moderating the partial mediating effect of green innovation.

Based on the literature review and research hypotheses above, this paper constructs the theoretical model shown in Figure 1. The model elucidates the key assumptions. This work focuses on the following: the intrinsic relationships among dual boundary-spanning search, green innovation, corporate sustainability performance, and AI capabilities. The paths depicted in the figure clearly delineate the ten research hypotheses put forth in this paper. These hypotheses encompass direct effects, mediating effects, moderating effects, and moderated mediating effects, which will be empirically tested in the subsequent sections.

3. Research Methods

3.1. Sample Selection and Data Collection

Compared to other types of enterprises, the market and the general public have higher expectations for the sustainability of manufacturing firms [54]. This is particularly evident in the urgent demand for green innovation products, technological upgrades, and enhanced sustainability performance. Therefore, this paper focuses on Chinese manufacturing enterprises, selecting samples from the provinces of Shanxi, Shandong, Hebei, and Henan. Senior and middle management, as well as technical research and development personnel from each enterprise, were invited to complete a questionnaire. The questionnaire was developed using established research scales from abroad and translated into Chinese following Brislin’s (1980) two-way translation procedure [55]. First, a preliminary survey was conducted, distributing 30 paper questionnaires to MBA students from 10 manufacturing enterprises affiliated with our institution. After collecting the responses, minor revisions were made to enhance the logic and clarity of certain sentences based on the feedback. These refinements produced a finalized questionnaire that more precisely captured the intended variable definitions. Second, electronic questionnaires were distributed to companies listed in the 2024 Chinese Manufacturing Enterprises Directory using their email addresses. Finally, the questionnaires were distributed to companies and collected via email through the social networks of the research team members. The distribution period for the questionnaires spanned from March 2024 to April 2025, totaling 13 months. In two rounds, 230 questionnaires were distributed. Following the exclusion of incomplete and invalid responses, a total of 171 valid questionnaires were obtained, yielding a valid response rate of 74.3%. The characteristics of the samples are outlined in

Table 1. Characteristics of sample enterprises.

Variate	Characteristics	No.	%
Enterprise age	Within 5 years	38	22.2%
	6 to 10 years	52	30.4%
	11 to 15 years	43	25.1%
	16 to 20 years	14	8.2%
	More than 20 years	24	14.1%
Enterprise scale	Less than 100 people	62	36.3%
	101 to 500 people	49	28.7%
	501 to 1000 people	26	15.2%
	1001 to 2000 people	20	11.7%
	More than 2000 people	14	8.1%
Ownership of enterprise	State-owned enterprise	73	42.7%
	Private enterprise	98	57.3%

.

3.2. Questionnaire Design

This paper primarily employs established foreign scales to measure variables. The quality of the questionnaire was rigorously controlled through multiple two-way translations and expert reviews. Additionally, the number of questions and their relevance to the study were adjusted based on the results of preliminary research. All variables were measured utilizing a five-point Likert scale. Each variable, including boundary-spanning search, green innovation, sustainability performance, and AI capabilities, was assigned a value from 1 to 5. These values correspond to “strongly disagree” and “strongly agree.” Specifically, boundary-spanning search is assessed from two dimensions: prospective and responsive. The scale developed by O’Cass [56] and Kumar [57] was utilized. Prospective boundary-spanning search comprises four items, with a Cronbach’s α value of 0.846, while responsive boundary-spanning search also comprises four items, with a Cronbach’s α value of 0.828. For green innovation, the scale developed by Chan [58] was employed. This scale consists of five items, with a Cronbach’s α value of 0.870. For sustainability performance, the scale developed by Bansal [59] was utilized. This scale comprises six items, with a Cronbach’s α value of 0.903. For AI capabilities, six items were designed, yielding a Cronbach’s α value of 0.853, which references the scale developed by Mikalef and Gupta [32]. All items from the scales utilized in this paper are detailed in

Table 2. Variable measurement items.

Variable	Code	Subfactor	Reference
PBS	PBS1	Your enterprise can identify new markets and customer segments before competitors do.	O’Cass et al. [56]
	PBS2	Your enterprise can create new products with distinctive features that outpace competitors.
	PBS3	Your enterprise can explore innovative approaches better to satisfy customer needs before others in the market.
	PBS4	Your enterprise can adopt advanced product development processes and technologies ahead of your competitors.
RBS	RBS1	Your enterprise strengthens its current market position by monitoring competitors.	Kumar et al. [57]
	RBS2	Your enterprise enhances the efficiency of its current products by benchmarking against competitors.
	RBS3	Your enterprise elevates the standard of its products through closely monitoring competitors.
	RBS4	Your enterprise optimizes the development processes and technologies of its existing products by following industry trends set by competitors.
GI	GI1	Your enterprise leads the industry in energy conservation and emission reduction efforts.	Chan et al. [58]
	GI2	Your enterprise’s sales of green innovation products are experiencing a notable increase as a percentage of total sales.
	GI3	Your enterprise has successfully established a positive social image through its commitment to green innovation.
	GI4	Your enterprise is actively engaged in researching and developing innovative green technologies and products.
	GI5	Your enterprise consistently improves its manufacturing processes to adhere to elevated standards of sustainable production.
CSP	CSP1	Your enterprise places significant importance on environmental protection initiatives.	Bansal [59]
	CSP2	Your enterprise has made considerable efforts to safeguard the environment.
	CSP3	Your enterprise actively engages in various environmental protection projects.
	CSP4	Your enterprise regularly reviews its environmental performance to ensure compliance and improvement.
	CSP5	Your enterprise achieves strong economic performance within the industry.
	CSP6	Your enterprise has earned social recognition for its commitment to environmental protection.
AIC	AIC1	Your enterprise can consolidate data from various internal sources into a centralized database for easy access.	Mikalef and Gupta [32]
	AIC2	Your enterprise can combine external data with internal data to facilitate high-value analysis of the business environment.
	AIC3	Your enterprise possesses the capability to share data across different business units and organizational boundaries.
	AIC4	Your enterprise can effectively prepare and clean AI data while identifying and addressing any errors present in the data.
	AIC5	Your enterprise has invested in scalable data storage infrastructure to support its needs.
	AIC6	Your enterprise has explored or adopted cloud services to process data and implement AI and machine learning solutions.

Notes: PBS, prospective boundary-spanning search; RBS, responsive boundary-spanning search; GI, green innovation; CSP, corporate sustainability performance; AIC, AI capabilities.

.

4. Empirical Analysis Results

4.1. Reliability and Validity Test

First, the reliability analysis conducted using SPSS 27.0 showed (see

Table 3. Reliability and validity analysis results.

Variate	Item	Load	Cronbach’s α	CR	AVE	Overall Cronbach’s α
Prospective boundary-spanning search	PBS1	0.780	0.846	0.848	0.582	0.913
	PBS2	0.757
	PBS3	0.754
	PBS4	0.760
Responsive boundary-spanning search	RBS1	0.808	0.828	0.829	0.549
	RBS2	0.712
	RBS3	0.754
	RBS4	0.685
Green innovation	GI1	0.797	0.870	0.901	0.646
	GI2	0.751
	GI3	0.830
	GI4	0.819
	GI5	0.820
Corporate sustainability performance	CSP1	0.808	0.903	0.903	0.609
	CSP2	0.744
	CSP3	0.823
	CSP4	0.766
	CSP5	0.818
	CSP6	0.718
AI capabilities	AIC1	0.783	0.853	0.870	0.528
	AIC2	0.714
	AIC3	0.694
	AIC4	0.764
	AIC5	0.682
	AIC6	0.719

) that five variables, including prospective boundary-spanning search, and the overall Cronbach’s α of the questionnaire surpassed 0.8. These findings confirm that the questionnaire meets the requisite reliability standards for subsequent analysis. Second, validity is assessed through convergent and discriminant validity. Since all scales are based on well-established international research, confirmatory factor analysis (CFA) is conducted. The results in Table 3 show that all factor loadings exceed 0.6, composite reliability (CR) values exceed 0.8, and the average variance extracted (AVE) is above 0.5. These findings confirm that the scales demonstrate good convergent validity. Additionally, the model fit results show that the five-factor model (χ²/df = 1.276, CFI = 0.967, TLI = 0.963, IFI = 0.968, RMSEA = 0.040) exhibits a significantly better fit than alternative models. This discovery strongly supports the excellent discriminant validity of the measurement model.

4.2. Common Method Bias and Collinearity Diagnostics

All observed variables are derived from the same questionnaire and sample collection. To avoid potential common method bias, this paper employs three methods to control this risk throughout the process. First, during the questionnaire design stage, this paper ensures that all participants received guarantees of anonymity. It also places explanatory and explained variables in different sections of the questionnaire to create a psychological isolation zone. Second, in the post hoc validation phase, this paper applies Harman’s single-factor test. The unrotated exploratory factor analysis revealed five factors with eigenvalues greater than 1, aligning with the number of variables in this paper. No common factors are identified. These five factors account for 68.0% of the total variance. The cumulative contribution rate of the first extracted factor is 32.55%, which falls short of the threshold of 40%. This preliminary result indicates that the sample data is not affected by common method bias. Third, this paper further employs the Unmeasured Latent Construct Method (ULMC) for verification after incorporating common method factors into the CFA model. Results presented in

Table 4. Confirmatory factor analysis and ULMC test results.

Model	χ2	df	χ2/df	TLI	CFI	RMSEA
one-factor model a	1351.699	275	4.915	0.472	0.516	0.152
two-factor model b	1217.749	274	4.444	0.535	0.576	0.142
three-factor model c	1106.622	272	4.068	0.586	0.625	0.134
four-factor model d	734.248	269	2.730	0.767	0.791	0.101
five-factor model e	338.119	265	1.276	0.956	0.962	0.043
ULMC model	316.382	250	1.266	0.964	0.970	0.040
change amount	23.555	15	0.010	0.008	0.008	0.003

Notes: ^a PBS + RBS + GI + CSP + AIC; ^b PBS, RBS + GI + CSP + AIC; ^c PBS, RBS, GI + CSP + AIC; ^d PBS, RBS, GI, CSP + AIC; ^e PBS, RBS, GI, CSP, AIC.

indicate that after introducing common method factors, the changes in RMSEA (Root Mean Square Error of Approximation), TLI (Tucker–Lewis Index), and CFI (Comparative Fit Index) are all less than 0.01. Such minimal changes indicate that the model has not significantly improved [60]. These findings further confirm that the sample data does not exhibit common method bias and that its quality meets research standards. Additionally, this paper uses SPSS 27.0 to conduct multicollinearity tests on the sample data. The maximum VIF value in each model is below 3, indicating that the sample data do not exhibit serious multicollinearity issues.

4.3. Correlation Analysis

A correlation analysis is conducted on prospective boundary-spanning search, reactive boundary-spanning search, green innovation, sustainability performance, and AI capabilities. As shown in

Table 5. Correlation analysis of variables.

Variate	MV	SD	1	2	3	4	5
PBS	3.684	0.737	1
RBS	3.664	0.798	0.454 ***	1
GI	3.463	0.757	0.526 ***	0.539 ***	1
CSP	3.825	0.794	0.432 ***	0.438 ***	0.430 ***	1
AIC	4.298	0.555	0.106	0.150	0.182 *	0.184 *	1

Notes: * p < 0.05, *** p < 0.001.

, at the 0.001 significance level, prospective boundary-spanning search (r = 0.432) and reactive boundary-spanning search (r = 0.438) are significantly and positively correlated with sustainability performance. Additionally, prospective boundary-spanning search (r = 0.526) and reactive boundary-spanning search (r = 0.539) are significantly and positively correlated with green innovation. Furthermore, green innovation (r = 0.430) is significantly and positively correlated with corporate sustainability performance. In summary, the correlation coefficients among the primary research variables are all below the critical threshold of 0.7. These findings suggest that the relationships between the research variables are consistent with the theoretical assumptions, establishing a preliminary basis for further hypothesis verification.

4.4. Hypothesis Testing

4.4.1. Main Effects Tests

This paper examines the relationship between prospective and responsive boundary-spanning search, green innovation, and corporate sustainability performance, utilizing SPSS 27.0 for stratified linear regression analysis. The regression results are summarized in

Table 6. Regression results for direct effects and mediation effects.

Variate	CSP						GI
	M1	M2	M3	M4	M5	M6	M7	M8
age	0.146	0.133	0.075	0.155 *	0.143	0.107	−0.036	−0.113
size	0.037	0.011	0.047	−0.036	−0.027	−0.005	0.138	0.181 *
ownership	−0.013	−0.079	−0.002	−0.057	−0.085	−0.035	0.024	0.115
PBS		0.437 ***			0.292 *		0.517 ***
RBS			0.423 ***			0.267 ***		0.550 ***
GI				0.432 ***	0.280 ***	0.284 ***
R2	0.028	0.214	0.203	0.209	0.270	0.258	0.293	0.327
ΔR2	0.011	0.195	0.184	0.190	0.248	0.235	0.276	0.311
F	1.633	11.314 ***	10.574 ***	15.670 ***	10.982 ***	11.445 ***	17.166 ***	20.140 ***
Max VIF	1.287	1.290	1.303	1.316	1.414	1.485	1.290	1.303

Notes: * p < 0.05, *** p < 0.001.

. First, sustainability performance is designated as the dependent variable, followed by the inclusion of control variables. Finally, the independent variables (prospective and responsive boundary-spanning search) are incorporated into the regression equation. The findings of the M1 analysis reveal that the selected control variables do not significantly impact the dependent variable. This appropriate selection of control variables facilitated smooth subsequent hypothesis testing. M2 shows that prospective boundary-spanning search has a significant positive impact on sustainability performance (β = 0.437, p < 0.001), while M3 indicates that responsive boundary-spanning search also has a positive influence on corporate sustainability performance (β = 0.423, p < 0.001). Therefore, H1a and H1b are supported.

4.4.2. Mediation Effects Tests

First, sustainability performance and green innovation are designated as dependent variables, with control variables added afterwards. Next, we include independent variables, specifically prospective and responsive boundary-spanning search. Finally, we add the mediating variable, green innovation, to the regression equation. The regression results are summarized in Table 6. Models M7 and M8 indicate that both prospective boundary-spanning search (β = 0.517, p < 0.001) and responsive boundary-spanning search (β = 0.550, p < 0.001) have a significant positive impact on green innovation. These findings validate H2a and H2b. Model M4 results demonstrate that green innovation significantly improves corporate sustainability performance (β = 0.432, p < 0.001). Model M5 results reveal that, after incorporating green innovation as a mediating variable, it has a positive effect on corporate sustainability performance (β = 0.280, p < 0.001). At the same time, a significant positive relationship remains between prospective boundary-spanning search (β = 0.292, p < 0.05) and sustainability performance, although the effect has notably diminished. Similarly, M6 analysis results indicate that, with green innovation included as a mediating variable, it significantly enhances sustainability performance (β = 0.284, p < 0.001). Responsive boundary-spanning search also maintains a positive correlation with corporate sustainability performance (β = 0.267, p < 0.001), although its effect has notably decreased. These findings imply that green innovation partially mediates the relationship between both prospective and responsive boundary-spanning search and corporate sustainability performance. This outcome confirms H3a and H3b.

4.4.3. Moderating Effects Tests

To examine the moderating effect of AI capabilities, we center the relevant variables prior to conducting the stratified regression analysis. First, we include green innovation as the dependent variable within the regression model. Next, we add control variables and independent variables, specifically proactive and responsive boundary-spanning search. Finally, we incorporate the moderating variable, AI capabilities. Lastly, we add the interaction terms for proactive and responsive boundary-spanning searches with AI capabilities to the regression model. The regression results are summarized in

Table 7. Regression results for moderating effects and moderating mediating effects.

Variate	GI		CSP
	M9	M10	M11	M12
age	−0.067	−0.096	0.154 *	0.122
size	0.134 *	0.146 *	−0.026	−0.003
ownership	0.037	0.083	−0.084	−0.042
PBS	0.474 ***		0.234 **
RBS		0.529 ***		0.235 **
GI			0.180 **	0.154 *
AIC	0.102	0.037	0.122	0.133
PBS × AIC	0.403 ***
RBS × AIC		0.237 ***
GI × AIC			0.214 **	0.253 ***
R2	0.467	0.386	0.313	0.314
ΔR2	0.447	0.364	0.284	0.284
F	23.904 **	17.192 ***	10.625 ***	10.645 ***
Max VIF	1.294	1.308	1.651	1.795

Notes: * p < 0.05, ** p < 0.01, *** p < 0.001.

. Model 9 indicates that the interaction between prospective boundary-spanning search and AI capabilities significantly enhances green innovation (β = 0.403, p < 0.001). This indicates that AI capabilities serve as a positive moderator in the relationship between prospective boundary-spanning search and green innovation. Similarly, Model 10 reveals that the interaction between responsive boundary-spanning search and AI capabilities significantly enhances green innovation (β = 0.237, p < 0.001). This evidence illustrates that AI capabilities positively moderate the relationship between responsive boundary-spanning search and sustainability performance. Therefore, we conclude that hypotheses H4a and H4b are substantiated. Additionally, we illustrate how different levels of AI capabilities affect the relationships between prospective and responsive boundary-spanning searches and green innovation. Figure 2 and Figure 3 illustrate these effects. In Figure 2, the slope for high-level AI capabilities is steeper than that for low-level AI capabilities. This finding shows that higher AI capabilities significantly boost the positive effect of prospective boundary-spanning search on green innovation. Similarly, Figure 3 reveals that the slope for high-level AI capabilities is greater than that for low-level AI capabilities. This suggests that higher AI capabilities also amplify the positive influence of responsive boundary-spanning search on green innovation.

4.4.4. Moderated Mediation Effects Tests

Model M11 incorporates the moderating variable, AI capabilities, and its interaction term with green innovation (GI × AIC), into Model M5 presented in Table 6. The results in Table 7 indicate that green innovation exhibits a significant positive impact (β = 0.180, p < 0.01), and the interaction term (GI × AIC) also reveals a significant positive effect (β = 0.214, p < 0.01). Therefore, H5a is confirmed. This suggests that prospective boundary-spanning search affects corporate sustainability performance via green innovation. The initial phase of this process is enhanced by AI capabilities, establishing a moderating mediating effect. Similarly, Model M12 incorporates the moderating variable, AI capabilities, and its interaction term with green innovation (GI × AIC), into Model M6 presented in Table 6. The results indicate that green innovation exhibits a significant positive effect (β = 0.154, p < 0.05), and the interaction term (GI × AIC) also demonstrates a significant positive effect (β = 0.253, p < 0.001). Consequently, H5b is validated, which suggests that responsive boundary-spanning search affects corporate sustainability performance via green innovation. The initial phase of this process is enhanced by AI capabilities, establishing a moderating mediating effect.

4.4.5. Robustness Tests

Additional robustness tests are performed using the Bootstrap method to analyze the direct, mediation, moderating, and moderated mediation effects described above. We perform 5000 repeated samplings and construct a 95% confidence interval. The dependent variable (sustainability performance), independent variables (prospective and responsive boundary-spanning search), mediating variable (green innovation), and moderating variable (AI capabilities) are then incorporated into the corresponding model. The results of the robustness tests are presented in

Table 8. Robustness test results for direct effects and mediation effects.

Effect Type	Path Relationship	Effect Size	SE	LLCI	ULCI
Direct effect	PBS → CSP	0.314	0.085	0.147	0.482
Indirect effect	PBS → GI → CSP	0.156	0.055	0.049	0.264
Total effect	PBS → CSP	0.470	0.075	0.322	0.619
Direct effect	RBS → CSP	0.266	0.081	0.106	0.426
Indirect effect	RBS → GI → CSP	0.156	0.057	0.055	0.279
Total effect	RBS → CSP	0.422	0.070	0.284	0.560

,

Table 9. Robustness test results for moderating effects.

Dependent Variable	Independent Variables	R2	F	B	LLCI	ULCI
GI	PBS	0.467	23.904 ***	0.487	0.368	0.605
	AIC			0.139	−0.016	0.294
	PBS × AIC			0.838	0.602	1.074
	RBS	0.386	17.192 ***	0.502	0.385	0.620
	AIC			0.046	−0.127	0.218
	RBS × AICI			0.426	0.200	0.652

*** p < 0.001.

and

Table 10. Robustness test results for moderated mediation effects.

	Indirect Effect	SE	LLCI	ULCI
Index of moderated mediation (start with prospective boundary-spanning search)	0.246	0.096	0.066	0.444
Conditional indirect effect at AI capabilities = M ± 1SD
M + 1SD	0.280	0.098	0.080	0.466
M − 1SD	0.006	0.029	−0.067	0.053
Index of moderated mediation (start with responsive boundary-spanning search)	0.127	0.056	0.028	0.243
Conditional indirect effect at AI capabilities = M ± 1SD
M + 1SD	0.262	0.220	0.080	0.265
M − 1SD	0.080	0.039	0.021	0.173

.

Table 8 shows that the direct effect of prospective boundary-spanning search on sustainability performance is 0.314, with a confidence interval of [0.147, 0.482]. This interval excludes 0, indicating a significant direct effect, which further supports hypothesis H1a. The indirect effect of prospective boundary-spanning search on sustainability performance through green innovation is 0.156, with a confidence interval of [0.049, 0.264]. This interval excludes 0, indicating a significant mediating effect, which further supports hypothesis H3a. The direct effect of responsive boundary-spanning search on sustainability performance is 0.266, with a confidence interval of [0.106, 0.426]. This interval does not include 0, indicating a significant direct effect, which further supports hypothesis H1b. The indirect effect of responsive boundary-spanning search on sustainability performance through green innovation is 0.156, with a confidence interval of [0.055, 0.279]. This interval does not include 0, indicating a significant mediating effect, which further supports hypothesis H3b.

Table 9 shows that the interaction term (PBS × AIC) between prospective boundary-spanning search and AI capabilities has an effect coefficient of 0.838 on green innovation. The confidence interval for this effect is [0.602, 1.074], which excludes 0. This finding indicates that AI capabilities significantly regulate the positive relationship between prospective boundary-spanning search and green innovation. Thus, it provides further support for hypothesis H4a. Additionally, the effect coefficient of the interaction term between responsive boundary-spanning search and AI capabilities (RBS × AIC) on green innovation is 0.426. The confidence interval is [0.200, 0.652], which also excludes 0. This finding illustrates that AI capabilities significantly regulate the positive relationship between responsive boundary-spanning search and green innovation. Therefore, it further supports hypothesis H4b.

Table 10 shows that under the mediation effect of green innovation, the indirect effect of “prospective boundary-spanning search → green innovation → sustainability performance” is not significant when AI capabilities are low. The confidence interval for this effect is [−0.067, 0.053], indicating that it includes 0. However, when AI capabilities are high, the indirect effect becomes significant, with a confidence interval of [0.080, 0.466] that excludes 0. Furthermore, the moderated mediation effect is significant, with a determination index of 0.246. The confidence interval for this effect is [0.066, 0.444], indicating that it also excludes 0. These findings indicate that AI capabilities significantly moderate the mediation effect of green innovation on the relationship between prospective boundary-spanning search and sustainability performance. These findings further validate hypothesis H5a. Additionally, under the mediating effect of green innovation, the indirect effect of “responsive boundary-spanning search → green innovation → sustainability performance” is significant when AI capabilities are low. The effect value in this case is 0.080, and the confidence interval is [0.021, 0.173], which does not include 0. When AI capabilities are high, the indirect effect remains significant and increases to an effect value of 0.220. The confidence interval for this effect is [0.080, 0.265], indicating that it also excludes 0. Moreover, the moderated mediation effect is significant, with an indicator value of 0.127 and a confidence interval of [0.028, 0.243], which excludes 0. These results suggest that AI capabilities significantly moderate the mediation effect of green innovation on the relationship between responsive boundary-spanning search and sustainability performance. These findings further validate hypothesis H5b. In summary, the conclusions of this paper are relatively robust.

5. Conclusions, Implications and Limitations

5.1. Conclusions

This paper focuses on Chinese manufacturing enterprises and explores the impact of boundary-spanning search on corporate sustainability performance. It also analyzes the mediation role of green innovation and the moderating effect of AI capabilities through empirical analysis.

First, the main effect analysis of the research model reveals that both prospective boundary-spanning search (β = 0.437, p < 0.001) and responsive boundary-spanning search (β = 0.423, p < 0.001) significantly enhance corporate sustainability performance. Additionally, both prospective (β = 0.517, p < 0.001) and responsive boundary-spanning search (β = 0.550, p < 0.001) positively impact green innovation.

Second, mediation analysis of the research model reveals that green innovation partially mediates the relationship between prospective boundary-spanning search and corporate sustainability performance. Similarly, green innovation also partially mediates the linkage between responsive boundary-spanning search and corporate sustainability performance. The indirect effects are both 0.156, with confidence intervals that do not include 0. As shown in Table 6, green innovation accounts for 33.19% of the partial mediating effect between prospective boundary-spanning search and sustainability performance. For responsive boundary-spanning search, green innovation accounts for 36.97% of the partial mediating effect on sustainability performance. The mediation effect of responsive boundary-spanning search is slightly higher than that of prospective boundary-spanning search. This difference in mediating effects may be due to the precise objectives of responsive boundary-spanning search activities. Such clarity allows the knowledge gained to be more rapidly transformed into improvements in product performance and increases in sales.

Third, the moderation analysis of this research model indicates that AI capabilities enhance the positive effects of both prospective boundary-spanning search (β = 0.403, p < 0.001) and responsive boundary-spanning search (β = 0.237, p < 0.001) on green innovation. Additionally, AI capabilities strengthen the partial mediating influence of green innovation on both types of boundary-spanning search and sustainability performance. The determination indices for the two segments of the moderated mediating effect are 0.246 and 0.056, with confidence intervals that exclude 0. As illustrated in Table 8, when AI capabilities are high, green innovation plays a more significant mediating role between both prospective and responsive boundary-spanning search and sustainability performance. Conversely, when AI capabilities are low, this mediating effect is either insignificant or minimal.

5.2. Theoretical Implications

In this paper, a moderated mediation model is established to examine how boundary-spanning search and green innovation enhance sustainability performance. The exploration of the mediation and moderating roles of green innovation and AI capabilities represents a novel contribution to the field. Three theoretical implications are outlined in the research.

Initially, a theoretical framework is provided for analyzing the relationship between dual boundary-spanning search and sustainability performance, drawing on organizational search and knowledge management theory. Hypotheses H1a and H1b reveal that both prospective and responsive searches positively affect sustainability performance. These results support Wang’s (2022) assertion regarding the connection between boundary-spanning search and sustainable performance [6], thereby enriching research on the driving factors of sustainability performance. The mechanisms through which these two types of boundary-spanning search impact sustainability performance vary depending on timing. Specifically, prospective boundary-spanning search offers new impetus for breakthrough green innovation and enhances sustainability by exploring new knowledge. This aligns with the conclusions of Xue (2024) and Appiah (2024), which identify knowledge restructuring as a driver of the green transition [61,62]. Conversely, responsive boundary-spanning search mitigates technological and market uncertainty in the green innovation process. It does this by absorbing the proven and mature experiences of competitors. This approach steadily enhances sustainable performance and further supports Yang’s (2021) argument that imitative learning reduces innovation risks [21]. Therefore, enterprises must balance both search strategies. Over-reliance on a single approach could hinder the efficiency of green innovation and the achievement of sustainability goals [63]. Furthermore, sustainability performance is influenced by multiple factors [64]; while boundary-spanning search is crucial, future research should explore additional potential influencing factors.

Next, the indirect hypotheses H2a, H2b, H3a, and H3b validate the mediation role of green innovation between dual boundary-spanning search and sustainability performance. The findings reveal that both prospective and responsive boundary-spanning searches exert direct and indirect positive effects on sustainability performance through green innovation. Specifically, green innovation activities facilitate the creation of viable green products and processes by leveraging heterogeneous knowledge acquired through boundary-spanning search. In other words, to enhance sustainability performance through green innovation, it is essential to apply both boundary-spanning search strategies comprehensively. Moreover, in contexts where organizations frequently engage with external stakeholders, green innovation is more likely to enhance sustainability performance [65]. Thus, companies should leverage effective search strategies and develop AI capabilities to support their green innovation activities. This study contributes to the literature by highlighting the critical role of corporate green innovation behaviors in achieving sustainability from a knowledge management perspective. These findings extend the research of Seman (2019) and Kanan (2023) [66,67], which previously demonstrated the vital role of green innovation in navigating dynamic environments and achieving sustainability in such contexts.

Finally, the results for hypotheses H4a, H4b, H5a, and H5b indicate that AI capabilities effectively moderate the relationship between dual boundary-spanning search and green innovation. Additionally, AI capabilities indirectly serve as a significant driving force for enhancing corporate sustainability performance by strengthening the mediation effect of green innovation. This finding aligns with the conclusions of Zhong (2025) and Almansour (2025) [68,69], which collectively support the critical role of AI capabilities in reducing resource waste and achieving sustainability. It is important to note that the positive impact of AI capabilities on green innovation and sustainability performance is not isolated [70]; instead, it is linked to interactions with technical methods such as boundary-spanning search. Currently, there are limited studies examining the impact of AI capabilities on green innovation and sustainability performance, thereby enriching the literature on this topic. In summary, enterprises need to develop AI capabilities, thereby enhancing the efficiency of utilizing knowledge obtained through boundary-spanning search. Consequently, it not only fosters green innovation but also reinforces its mediating effect between both prospective and responsive boundary-spanning search and sustainability performance. Ultimately, this leads to more effective improvements in sustainability performance.

5.3. Practical Implications

Integrating the research findings offers a synthesis of the practical implications. We present recommendations for both enterprises and the government based on three key conclusions.

(1) Enterprise managers ought to adopt strategic initiatives focused on dual boundary-spanning search, green innovation, and robust AI capabilities to advance sustainability objectives.

Implement Dual Boundary-spanning Search Strategies to Enhance Knowledge Integration Efforts: To achieve sustainability, enterprises should balance prospective and responsive boundary-spanning searches to avoid excessive consumption of corporate energy on a single search strategy. It is advisable for them to establish professional search teams and create tailored knowledge bases that address specific business needs. Additionally, managers should cultivate awareness of boundary-spanning practices and actively seek diverse external information, including technological trends and market shifts. The core of sustainable discovery performance lies in effectively integrating knowledge resources to realize their value. Thus, enterprises should utilize knowledge management systems and hybrid clouds, which promote efficient two-way communication and continuously improve knowledge integration efficiency within the organization.

Prioritize Green Innovation in Improving Sustainability: enterprises need recognize that green innovation is not just about meeting rigid compliance requirements, but a long-term strategic approach to driving sustainability. To achieve this, green concepts should be integrated into the entire value chain, including R&D, production, and logistics (e.g., developing biodegradable product packaging and building closed-loop recycling systems). Additionally, companies should establish platforms for sharing green knowledge while systematically collecting advanced green technologies, environmental standards, and best practices. Only by taking these steps can employees proactively monitor and reduce pollution in their roles, transforming sustainable goals into daily actions.

Adopt AI Technology to Enhance Channels for Green Transformation: This involves collaborating with AI firms to lower application thresholds and costs while integrating digital tools, like big data analysis and intelligent monitoring, into the research of green products and technologies. Leveraging AI will enable precise resource allocation, efficient knowledge flow, and informed decision-making, fostering a unique and sustainable competitive advantage. Moreover, AI technology providers should engage deeply with manufacturing enterprises in China, forming strong partnerships to develop specialized AI optimization models for processes with high material consumption and emissions. Additionally, exploring AI-based models for charging focused on green product premiums can further integrate sustainability performance for both parties.

(2) Regarding policy recommendations for the government, it is essential to implement effective interventions for sustainability. Policymakers should systematically enhance knowledge exploration and develop AI capabilities that facilitate green innovation.

Strengthen the Protection of Enterprise Knowledge Resources: Introducing policies that protect both explicit and tacit knowledge is urgent. These policies will reasonably guide enterprises in preventing the inappropriate spillover of benefits from intellectual assets. At the same time, the government should take firm action against enterprises that violate intellectual property rights. Such interventions will ensure that external knowledge obtained through boundary-spanning searches can generate economic returns.

Exert Pressure on Enterprises for Green Innovation from Both Public and Policy Angles: Initially, the government should work to raise public awareness of low-carbon development. This effort can include encouraging public participation in monitoring enterprises’ pollution behaviors and environmental performance, which will urge them to prioritize building a green image. Furthermore, penalties for polluting enterprises that cause environmental damage should be increased within the judicial process. Simultaneously, enterprises that excel in energy conservation and emission reduction should receive substantial incentives, including tax reductions and preferential procurement rights. These measures aim to enhance the internal motivation of enterprises to pursue green innovation.

Provide Vigorous Support for Enhancing AI Capabilities: The government should encourage enterprises to build their AI capabilities without delay. At the national strategy level, it is crucial to promote the integration of AI technology with the tangible economy by providing financial and policy support for research and development in intelligent technologies within manufacturing and AI companies. This initiative will cultivate a market environment conducive to AI. Furthermore, actively promoting successful examples of AI integration in green manufacturing will help mitigate the risks and uncertainties that manufacturing enterprises face in adopting new technologies.

5.4. Limitations and Future Research Directions

This study integrates organizational search theory with knowledge management theory. It examines the influence mechanisms among variables, including boundary-spanning search, green innovation, sustainability performance, and AI capabilities. However, three limitations are identified that suggest areas for improvement in future research.

Firstly, all variable data are collected through questionnaire surveys, which have inherent limitations in capturing the dynamic development processes of enterprises. Future research could be enhanced by incorporating panel data or longitudinal surveys.

Secondly, the current approach to boundary-spanning search is limited, focusing only on prospective and responsive perspectives regarding their impact on sustainability performance. In future research, the dimensions of boundary-spanning search could be categorized based on objectives or breadth. For instance, boundary-spanning search could be categorized into market knowledge and technical knowledge to broaden the research perspective.

Lastly, the findings of this paper confirm that green innovation partially mediates the relationship between dual boundary-spanning search and corporate sustainability performance. Additionally, it highlights the moderating role of AI capabilities in the connection between dual boundary-spanning search and green innovation. However, the role of boundary-spanning search in corporate sustainability performance involves a complex process of influence. Factors such as executive team characteristics, green dynamic capabilities, and environmental dynamism can significantly influence this process. These potential influence mechanisms require further investigation and exploration in future studies.