Artificial Intelligence as a Catalyst for Sustainable Tourism: A Case Study from China

Abstract

The tourism industry’s explosive growth has triggered severe carbon emission issues, making enhancing tourism carbon efficiency (TCE) a pressing concern for achieving sustainable tourism development. The widespread application of artificial intelligence (AI) in tourism presents new opportunities. This study applies the Environmental Kuznets Curve (EKC) theory to examine the pathways and mechanisms of AI’s impact on TCE, with a focus on China. The findings reveal that AI significantly enhances TCE, where improvements in tourism labor productivity, the rationalization of the tourism industry structure, and advancements in tourism technology are the key channel mechanisms. Heterogeneity tests indicate that AI substantially boosts TCE in eastern developed regions and areas with deficient tourism resource endowments. Furthermore, AI exhibits significant spatial spillover effects, enhancing both local and neighboring regions’ TCE. These insights provide crucial policy implications for utilizing AI to promote China’s sustainable tourism industry.

1. Introduction

Tourism is a major economic activity and has emerged as one of the most dynamic and rapidly expanding industries globally [1]. According to the UNWTO, international tourism revenue reached USD 1.481 trillion in 2019, a 2.5% increase from 2018 [2]. But as a result of the industry’s quick growth, energy consumption and carbon emissions have gone up [3]. Research indicates that tourism’s contribution to global greenhouse gas emissions is nearly 8% [4], significantly higher than many other economic sectors, raising concerns about environmental sustainability. This is especially critical in China, which has the biggest domestic tourism market in the world. Therefore, enhancing the tourism carbon efficiency (TCE) [4,5] to achieve sustainable development goals has become an urgent practical and research issue. Meanwhile, artificial intelligence (AI) is opening up new possibilities in the tourism sector due to the quick development and expanding use of intelligent automation technology [6].

As a machine or technology capable of performing tasks that require human intelligence [7], AI contributes to fostering economic development [8] and environmental protection [9], and is becoming a key factor in sustainable development. Given its information-intensive [10] and experiential nature [11], the tourism sector is becoming a significant application scenario for AI [12], including autonomous vehicles, hotel e-concierge services, virtual tour guides, robot chefs, and smart chatbots [13], as well as projects like China’s Digital Forbidden City, Digital Dunhuang, and Time Machine Europe [14]. AI has a broad impact on tourism; from a management perspective, it enhances the insights and predictive capabilities of tourism authorities regarding large volumes of information [15]; from a supply perspective, it optimizes business processes and saves costs [10]; from a demand perspective, it encourages tourists to reduce high-energy, high-emission travel activities [10]. As AI’s application in tourism widens, further analysis of its impact on TCE is urgently needed.

Tourism labor productivity, the tourism industry structure, and the tourism technological level can facilitate production in tourism enterprises and reduce their energy consumption and carbon emissions [16,17,18,19,20], thereby influencing TCE. Evidence suggests that AI has been shown to replace repetitive tasks [15], inspire optimal employee performance [21], and enhance labor productivity in the tourism sector [15]. Simultaneously, AI improves factor flows [22], which helps to optimize the structure of the tourism sector [17]. Furthermore, AI can enhance the tourism technological level by optimizing resource allocation [11] and improving energy efficiency [23]. Thus, enhancing tourism’s labor productivity and optimizing the tourism industry structure and tourism technological level are crucial for AI to promote TCE.

Three major categories can be used to group the existing research that is pertinent to this subject. The first group encompasses studies focusing on AI. AI has sparked a new wave of industrial, technological, and societal transformations, drawing considerable attention from the academic community. The effects of AI on labor structure [24], productivity [25], and economic growth [8] are becoming more and more prominent in the literature that is currently available. As environmental issues become deeply ingrained in societal consciousness, some scholars have begun to explore the pathways through which AI contributes to low-carbon development. Research has revealed that AI exhibits significant emission reduction effects [26], and its theoretical positioning as a key driver of sustainable economic growth has received empirical support [22,26]. The second category involves research on TCE. Existing studies generally agree that TCE serves as a critical indicator of sustainable tourism development [27]. Extensive research has been conducted on the conceptual connotations, measurement methodologies, and influencing factors of TCE [4,5,27,28,29]. It is widely acknowledged that factors such as urbanization, environmental regulations, industrial structure, tourism agglomeration, tourism scale, openness, and technological innovation impact TCE [1,4,5], providing a solid theoretical foundation for this study. However, no research has thus far integrated AI and TCE into a unified analytical framework. This academic gap provides the impetus and direction for the current research. Studies on AI in the hotel and tourism industries make up the third category. Research indicates that AI has transformed the hospitality and tourism industry [14,30], acting as a catalyst [6,12]. AI reduces operational costs [14], enhances demand forecasting capabilities [31], and improves marketing and economic performance [7,32,33]. Furthermore, AI influences tourist perceptions [34], satisfaction [35], experiences [19,36], and pro-environmental behavior [37]. Studies have also demonstrated the impact of AI on tourism professionals, such as hotel employees [38] and tour guides [15]. Concurrently, other academics have looked at how AI might affect sustainable tourism development [10,12], suggesting that AI offers numerous potential solutions to sustainability challenges in tourism [13]. Overall, the focus of the literature currently available on AI in the hotel and tourism industry is gradually shifting from purely economic effects to sustainable development that balances economic and environmental outcomes [39], but it primarily relies on theoretical explanations [13,39], lacking empirical validation. In summary, this study endeavors to integrate AI and TCE into a unified analytical framework, conducting comprehensive theoretical analyses and rigorous empirical tests to fill the aforementioned academic gap.

There are four main research questions that this study focuses on: (1) What kind of effect does AI have on TCE? (2) Through what mechanisms does AI influence TCE? (3) Does AI’s effect on TCE exhibit heterogeneity? (4) Does AI demonstrate spatial spillover effects on TCE? Utilizing provincial-level panel data (2000–2022) from China, this research employs the fixed effects panel model, mediation effect model, and spatial Durbin model to analyze the AI-TCE link based on the Environmental Kuznets Curve (EKC) theory. Our findings reveal that AI positively enhances TCE. The empirical results demonstrate that AI promotes TCE through three key mediating channels: the improvements in tourism labor productivity, the rationalization of the tourism industry structure, and advancements in tourism technology. Furthermore, our analysis identifies significant heterogeneity in AI’s effects across different geographical regions and varying levels of tourism resource endowments. Notably, we document positive spatial spillover effects, indicating that AI development in one region contributes to TCE not only locally but also in adjacent areas. The results offer robust empirical support and policy implications for tourism enterprises and local governments seeking to implement AI-driven sustainable development strategies, contributing new insights to the literature on sustainable tourism pathways.

This study introduces four innovations. Firstly, it integrates AI and TCE within a unified analytical framework, examining AI as a determinant of TCE to enrich the existing literature. Secondly, drawing on EKC theory, it elucidates the mechanistic pathways—improvements in tourism labor productivity, optimization of tourism industry structure, and advancements in tourism technology—through which AI influences TCE, revealing novel avenues for AI-driven sustainable tourism development. Thirdly, it elucidates the heterogeneous effects of AI on TCE from regional and tourism resource endowment perspectives, offering policy recommendations for differentiated government interventions and fostering intelligent transformation in tourism. Finally, it examines how AI affects TCE spatially from the standpoint of spatial correlation, highlighting the need for coordinated regional development strategies in tourism.

This paper’s remaining sections are organized as follows: The materials and methods are described in Section 2. The results and discussion are shown in Section 3. The conclusions and implications are presented in Section 4.

2. Materials and Methods

2.1. Research Hypothesis

AI significantly impacts TCE by enhancing the carbon monitoring capabilities of regulatory bodies [9], improving the operational performance of tourism enterprises [10], and fostering pro-environmental behavior among tourists [37]. Specifically, AI technologies allow tourism authorities to automatically track, record, and trace tourism carbon footprints, enhancing both early warning and perception abilities, and increasing their precision in formulating and implementing carbon reduction policies [40], which is beneficial to destination sustainable development [14]. Moreover, hotels and tourism make extensive use of AI technologies like chatbots, virtual assistants (VAs), self-service kiosks, and natural language processing (NLP) [30]. These technologies change and reshape the production processes in terms of forecasting, production, promotion, and supply, while also reducing costs and improving productivity [14]. Furthermore, AI facilitates public access to environmental information, reshaping public attitudes toward the environment [41], and reducing high-energy and high-carbon activities [10]. However, emerging scholarship has raised concerns regarding the exponential growth in computational demands driven by AI development, which may consequently increase energy consumption and carbon emissions [42]. While these studies advocate for greater scrutiny of AI’s environmental impacts, they simultaneously acknowledge that certain industries may derive substantial benefits from the adoption of AI [43]. The prevailing academic consensus suggests that AI predominantly enhances tourism energy efficiency [13] and achieves emission reductions [6,44]. Building upon this foundation while accounting for the computational energy costs, our study posits that AI will remain beneficial for TCE, which leads us to propose Hypothesis 1.

Hypothesis 1:

AI enhances TCE.

The Environmental Kuznets Curve (EKC) theory was developed by Grossman and Kruger (1991) and suggests that there is a nonlinear link between environmental pollution and economic growth through three main mechanisms: the scale effect, the structural effect, and the technical effect [45]. This theory posits that during initial economic development stages, environmental quality may deteriorate due to production scale expansion and intensified resource consumption (scale effect). However, sustained economic growth enables industrial structure upgrading and technological advancement to generate positive impacts, enhancing resource utilization efficiency and promoting cleaner production technologies that ultimately counteract and potentially reverse the negative scale effects. An inverted U-shaped trajectory between environmental contamination and economic development is produced by this dynamic progression [46]. It is thus evident that the scale–structure–technology analytical framework constitutes a vital component of the EKC theory, fundamentally elucidating the intrinsic transmission mechanism through which economic growth influences environmental quality [47]. Drawing inspiration from EKC theory, this study examines AI’s impact on TCE through three channel mechanisms: tourism labor productivity, the tourism industry structure, and the tourism technological level.

AI can influence labor productivity [25]. AI improves tourism labor productivity by boosting revenue, reducing costs, and improving workforce quality, thus enhancing the industry scale. Specifically, AI enables the synergistic combination of multiple tourism elements, significantly enhancing the value added to tourism products and services. For instance, augmented reality (AR) has been applied in various historical tourism settings to create immersive spaces by linking advanced abstract concepts with physical reality [30], thereby enhancing visitor experiences while boosting tourism revenue. Secondly, AI supports destination marketing and consumer travel decision-making through intelligent tourism recommendation systems (TRSs) [32], reducing the indirect steps and distances between tourism supply and demand [34], and lowering transaction costs. Thirdly, AI enhances the quality of tourism workers through substitution and skill enhancement effects. AI replaces some low-skilled labor [38], such as travel consultants and tour guides [15], thereby enhancing the workforce retained [21]. Additionally, robotics assists tourism employees in working more efficiently, necessitating additional training and skill upgrading [21]. Overall, AI significantly enhances tourism labor productivity. As productivity rises, tourism enterprises can achieve greater output with the same workforce, thereby improving TCE. Consequently, this research proposes Hypothesis 2.

Hypothesis 2:

AI enhances TCE by promoting tourism labor productivity.

AI plays a pivotal role in facilitating structural transformation within industries [17], impacting TCE by rationalizing and upgrading the tourism industry structure. Notably, AI promotes production factor mobility among tourism subsectors [22], fostering coordinated development [17] and intensive production, thereby mitigating structural imbalances in sectoral proportions. Furthermore, intelligent development in tourism can steer industry sectors toward higher value-added development and lower carbon emissions, accelerating the process toward advanced, green development [26] and promoting a more advanced industry structure [17]. In tourism subsectors, sectors like accommodation and catering services emit less carbon compared to sectors like rail, road, air, and maritime transport [16], and optimizing the tourism industry structure can reduce overall emissions. The analytical results suggest that structural optimization of the tourism sector, encompassing both the rationalization and upgrading of industrial structure, constitutes a critical pathway for improving TCE, leading to the proposal of Hypothesis 3 and Hypothesis 4.

Hypothesis 3:

AI enhances the TCE through the rationalization of the tourism industry structure.

Hypothesis 4:

AI enhances the TCE through the upgrading of the tourism industry structure.

According to endogenous growth theory, technological advancement serves as the primary engine of economic expansion. Upgrades in tourism technology driven by AI foster industry development, while enhancing energy efficiency and reducing carbon emissions [20]. AI will elevate the tourism technology level through technology application, technology integration, and factor accumulation. First, intelligent development in tourism leverages the potential for energy savings and emissions reductions: AI-driven robots or virtual agents promote resource-saving behaviors among tourists by providing feedback on their water and energy usage during hotel stays [13]. Second, AI’s capabilities in digital algorithms, smart applications, and telecommunications, along with its precision, provide solid hardware and algorithmic support for enhancing the tourism technology level, such as improving the accuracy of tourism forecasts [30] and optimizing production to minimize resource wastage. Third, AI facilitates a “replacement of humans with machines”, not only accelerating the tourism economic scale but also directing more resources into the industry’s core sectors to foster innovation and enhance technological advancement.

Upgrading tourism technology is a crucial measure for improving TCE. On one hand, it enables transformative changes in traditional tourism production–supply models, enriching tourism products and enhancing supply capabilities [48]. On the other hand, it reduces reliance on primary resources like raw materials and capital, increasing the use of high-tech resources and high-quality human capital [49], and consequently reducing energy consumption. Thus, Hypothesis 5 is proposed.

Hypothesis 5:

AI enhances TCE by advancing the tourism technological level.

The first law of geography states that everything in space is connected to everything else. Theoretically, AI not only impacts TCE within a local region but also influences the TCE of neighboring regions. Firstly, the siphon effect posits that if local AI development is advanced, local enterprises can attract high-quality capital and labor from neighboring areas by offering higher marginal returns [50]. This, driven by profit-seeking motives, may inhibit AI development and TCE improvement in those neighboring areas as high-quality production factors migrate. Secondly, the radiation effect suggests that AI technology transcends time and spatial limitations, enhancing the geographical permeability and facilitating its diffusion [50], thus affecting the TCE in adjacent areas. Therefore, Hypothesis 6 is proposed (Figure 1).

Hypothesis 6:

AI exhibits a spatial spillover effect on TCE.

2.2. Empirical Model

Building upon the established theoretical framework and the existing literature [24,51,52], this study employs a quantitative approach to examine the relationship between AI and TCE. Initially, a correlational design is utilized to preliminarily identify the correlation between AI and TCE [51,52]. Subsequently, a two-way fixed effects panel econometric model is employed to further assess the relationship between AI and TCE [24]. The specified model incorporates entity and time-fixed effects, effectively mitigating potential endogeneity concerns arising from unobserved individual heterogeneity and time trends. This rigorous specification enables the precise quantification of the causal relationship between the explanatory and explained variables while providing statistically robust evidence to evaluate our research hypotheses.

$$TC{E}_{it}=\alpha +{\beta }_{1}A{I}_{it}+{\displaystyle \sum _{j=1}^n{\beta }_j{X}_{it}^j}+{\gamma }_i+{\delta }_t+{\epsilon }_{it}$$

(1)

where the subscripts $i$ and $t$ denote the province and year, respectively. The value of $TCE$ is the dependent variable, while the value of $AI$ is the core explanatory variable; $X$ represents other control variables, with their measurement details discussed later. $n$ denotes the number of control variables. ${\gamma }_i$ and ${\delta }_t$ represent fixed effects for provinces and years, respectively, and ${\epsilon }_{it}$ denotes the random disturbance term.

2.3. Variables and Data

This study utilizes balanced panel data spanning 2000–2022 from 30 Chinese provinces (Taiwan, Hong Kong, Macao, and Tibet were not included because of data restrictions). The dataset’s structure is particularly well suited for two-way fixed effects panel model analysis. From an econometric perspective, the data possess both sufficient cross-sectional dimensions ($i=30$) and time-series dimensions ($t=23$), satisfying the large-sample requirements for reliable panel model estimation.

This study selects China as the case study subject based on two compelling rationales: First, as the world’s largest developing economy, China’s tourism sector generated over 160 million tons of carbon emissions in 2019 [29], demonstrating significant emission reduction pressures that render the decarbonization and sustainable development of its tourism industry both urgent and paradigmatic [27]. Second, China has emerged as a global leader in AI development—evidenced by surpassing the United States in AI patent applications [53]—while pioneering smart tourism applications [14], thereby providing an ideal context for examining AI-TCE relationships. This unique combination of substantial carbon emission challenges and technological innovation capabilities establishes China as a seminal case for investigating sustainable tourism pathways.

The research draws upon multiple authoritative data sources, including the China Statistical Yearbook, China Tourism Statistical Yearbook, China Culture, Heritage, and Tourism Statistical Yearbook, China Science and Technology Statistical Yearbook, China High-tech Industry Statistical Yearbook, China Domestic Tourism Sample Survey, International Tourist Sample Survey, China Labor Statistical Yearbook, various provincial yearbooks, and the International Federation of Robotics (IFR) website. Missing values for certain years were supplemented using linear interpolation. Apart from ratio data, fixed asset investment in tourism and total tourism revenue were deflated using the fixed asset price index and the consumer price index, with 2000 as the base year.

2.3.1. Explained Variables: TCE

Given the structural attributes of the tourism sector, three key input metrics were selected: tourism fixed assets investment, tourism employees, and tourism energy consumption [5,54]. Total tourism revenue is used as the desirable output, and tourism carbon emissions as the undesirable output. The TCE index system was developed (

Table 1. Input–output table of TCE.

Primary Indices	Secondary Indices	Indices’ Names	Units
Inputs	Capital	Tourism fixed assets investment	Hundred million CNY
	Labor force	Tourism employees	Person
	Energy consumption	Tourism energy consumption	MJ
Outputs	Desirable output	Total tourism revenue	Hundred million CNY
	Undesirable output	Tourism carbon emissions	10,000 tons

), and the value of TCE was determined using the Super-SBM model [4]. The detailed calculation process for tourism energy consumption and carbon emissions is provided in Supplemental Material S1.

2.3.2. Explanatory Variable: AI

Reliable AI measurement remains challenging in China due to insufficient direct statistical records. Some literature uses the density of robot installations as a proxy [9]. However, robots do not necessarily represent AI, and many robotics statistics may not accurately reflect regional levels of AI [55]. Moreover, AI largely constitutes an intangible asset rather than a consumable product, making the intellectual contributions of smart software and software developers difficult to measure objectively. Therefore, this study draws on relevant research [26,56] and the “Intelligent Manufacturing Capability Maturity Model White Paper (Version 1.0)” published by the China Electronics Technology Standardization Institute. Considering data availability, it constructs an AI evaluation index system with three primary indicators: intelligent infrastructure, intelligent technology, and intelligent outputs (

Table 2. The evaluation index system of AI.

Primary Indices	Secondary Indices	Indices’ Names	Units	Properties
Intelligent infrastructure	Intelligent talent	Number of employees in information transmission, software and information technology services	Person	Benefits
	Intelligent capital	Investment in fixed assets in information transmission, software and information technology services	Ten thousand CNY	Benefits
Intelligent technology	Intelligent technology development	Number of high-tech enterprises	Piece	Benefits
	Intelligent technology service	Primary business revenue of high-tech enterprises	Hundred million CNY	Benefits
Intelligent outputs	Intelligent economic benefit	The proportion of revenue from new product sales in the primary business revenue of industrial enterprises	%	Benefits
		Average output value per capita in the electronics and telecommunications equipment manufacturing industry	Hundred million CNY per capita	Benefits
	Intelligent environmental benefit	Energy consumption per unit of GDP	Ton of coal equivalent per 10,000 CNY	Costs

). Specifically, intelligent infrastructure forms the physical foundation and basic security for AI development; intelligent technology, which is backed by fundamental technologies like cloud computing, big data, and Internet +, drives AI advancement; and intelligent outputs reflect the comprehensive benefits of AI, including economic and environmental benefits, serving the fundamental goal of marketization. The entropy weighting approach [56], which is renowned for its objectivity, is used in this study to compute AI.

Instrument variables. The instrumental variable approach is used in this paper to address endogeneity problems, including reverse causality and omitted factors. Initially, drawing from Fu et al. [57], the number of post offices in 1984 was used as an instrumental variable for AI. These historical data, closely related to AI development yet weakly correlated with TCE, satisfy the relevance and exogeneity requirements for an instrumental variable. However, since these data are cross-sectional and do not reflect time trends, this study multiplies the number of post offices in 1984 by the year to introduce a time trend and dynamic character, creating the first instrumental variable. Secondly, following the relevant literature [40], the second instrumental variable is the lagged value of the core explanatory variable.

Mechanism variables. Tourism labor productivity (TLP): Following Chen et al. [58], TLP is calculated as the ratio of total tourism revenue to total tourism employees.

Tourism structure rationalization (TSR): Star-rated hotels, travel agencies, and scenic spots make up the tourism economic sectors. Utilizing the modified Theil index model [26], based on the revenue and employment numbers of these three sectors, the study calculates the index of TSR. A lower TSR value, approaching zero, indicates a more rational industry structure.

Tourism structure upgrading (TSU) is assessed by the aggregate of the product of the revenue share and the corresponding labor productivity of each tourism sector [26]. A higher TSU value indicates a higher level of industry structure upgrading.

Tourism technological level (TTEC): Based on He, Zha, and Loo [20], the TTEC is determined by dividing total tourism revenue by total tourism energy consumption.

Control variables. Urbanization, openness, industry structure, environmental regulation, and tourism agglomeration affect tourism development and carbon emissions [1,4,5]. The ratio of urban to total population (Urban), the ratio of total imports and exports to GDP (Open), the location entropy of total tourism revenue and GDP (Agg), the ratio of tertiary industry output to GDP (Str), and the ratio of environmental governance investment to GDP (Regl) are used as control variables in the study to measure these factors and guarantee the accuracy of the estimates. Descriptive statistics for the primary variables are shown in

Table 3. Descriptive statistics of variables.

Variable	Description	Observations	Mean	SD	Minimum	Maximum
TCE	Tourism carbon efficiency	690	0.2091	0.2072	0.0071	2.8345
AI	Artificial intelligence	690	0.0843	0.0971	0.0038	0.7595
Urban	Urbanization	690	52.4706	15.5883	13.8850	94.1516
Open	Openness	690	29.7020	35.9038	0.7637	172.1482
Agg	Tourism agglomeration	690	1.0374	0.5329	0.0104	4.3657
Str	Industrial structure	690	43.7716	9.5605	28.6000	83.9000
Regl	Environmental regulation	690	1.2624	0.7751	0.0994	9.2165
TLP	Tourism labor productivity	690	211.6824	200.5875	3.8160	1240.8311
TSR	Tourism structure rationalization	690	0.9833	0.8792	0.0444	11.7255
TSU	Tourism structure upgrading	690	6.1137	7.7479	0.1051	64.4524
TTEC	Tourism technological level	690	3.2557	2.6893	0.0459	19.2205

, and the correlation test results are shown in Figure 2.

3. Results and Discussion

3.1. Results

3.1.1. The Values of Artificial Intelligence and Tourism Carbon Efficiency

This study plotted the annual average trends of AI and TCE (Figure 3a), where TCE exhibited a fluctuating growth pattern before 2020, similar to trends observed by Wang and Luo [4]. But a downward trend started in 2020 as a result of the COVID-19 pandemic’s effects. AI also showed a growth trend before 2020, akin to findings by Meng, Xu, and Zhang [26], with notably accelerated growth post-2020. This suggests that the COVID-19 pandemic significantly accelerated AI development [33], as the contact-reducing and safety-enhancing characteristics of AI technologies garnered increased support for their advancement. Additionally, both variables exhibited broadly similar trends, suggesting a potential positive correlation, yet the true nature of this relationship requires further rigorous empirical testing.

The study also mapped the average spatial distributions of AI and TCE (Figure 3b,c), revealing a pronounced regional imbalance. High values of TCE were predominantly found in the central regions, mirroring findings by Liu, Gao, and Tsai [5]. Areas of high AI were primarily concentrated in the eastern regions [26], indicating the necessity to analyze heterogeneity effects based on locational characteristics.

3.1.2. Baseline Analysis

The baseline regression findings of AI’s effect on TCE are shown in

Table 4. Baseline analysis result.

	(1)	(2)	(3)	(4)	(5)	(6)
Variable	TCE
AI	1.3378 ***	0.3193 **	0.7026 ***	0.6315 ***	0.6039 ***	0.5967 ***
	(0.1103)	(0.1338)	(0.1179)	(0.1251)	(0.1233)	(0.1295)
Urban			0.0010	0.0003	0.0013 *	0.0003
			(0.0007)	(0.0008)	(0.0007)	(0.0009)
Agg			0.1951 ***	0.1763 ***	0.2168 ***	0.1924 ***
			(0.0148)	(0.0150)	(0.0165)	(0.0170)
Str			0.0093 ***	0.0021	0.0084 ***	0.0023
			(0.0013)	(0.0019)	(0.0013)	(0.0019)
Open					−0.0013 **	−0.0005
					(0.0006)	(0.0006)
Regl					−0.0167 *	−0.0214 **
					(0.0090)	(0.0092)
Constant	0.0962 ***	0.1242 ***	−0.5074 ***	−0.1562 *	−0.4394 ***	−0.1316
	(0.0114)	(0.0288)	(0.0444)	(0.0874)	(0.0498)	(0.0876)
Year effect	NO	YES	NO	YES	NO	YES
Province effect	YES	YES	YES	YES	YES	YES
Adjusted R-squared	0.1453	0.3337	0.4119	0.4513	0.4180	0.4550
Observations	690	690	690	690	690	690

Note: * p < 0.1, ** p < 0.05, and *** p < 0.01. Standard error is enclosed in parentheses.

. While columns (2), (4), and (6) include both temporal and individual fixed effects, columns (1), (3), and (5) only include individual fixed effects. Overall, AI significantly enhances TCE, confirming Hypothesis 1. Taking column (6) as the benchmark, an average increase of 1 percentage point in AI is correlated with a 0.5967 percentage point increase in TCE, indicating that AI is a driver of sustainable tourism development. Agg shows a significantly positive effect on TCE among the control variables, which is consistent with existing research [1] suggesting that Agg enhances inter-firm collaboration through talent concentration, infrastructure development, and information sharing, thereby improving tourism output efficiency. The findings further corroborate study [27], indicating that at the current stage, Agg strengthens TCE by improving tourism resource utilization rates, optimizing industrial structure, and enhancing production efficiency. However, with the further intensification of agglomeration in the future, the potential agglomeration diseconomy may increase energy consumption and environmental pollution, and ultimately reduce TCE. Conversely, the effect of Regl on TCE is significantly negative, contrary to the findings of Wang and Luo [4]. This may be because current governmental environmental regulations have raised environmental standards. These heightened standards increase pollution control costs for tourism enterprises, thereby reducing TCE to some extent. Meanwhile, the coefficients for Urban, Open, and Str did not pass the significance test.

3.1.3. Robustness Checks

To guarantee the validity of the study’s findings, a number of robustness checks were carried out, with results detailed in Supplemental Material S2.

Endogeneity test. The instrumental variable approach is used in this work to handle any possible endogeneity concerns. The findings demonstrate the robustness of the baseline regression results by showing that the influence of AI on TCE remains considerably favorable even when endogeneity is controlled.

Alternative measures for independent and dependent variables. This paper uses alternative measures to mitigate measurement error effects on conclusions. AI was remeasured by PCA [26] or robot density [9]. TCE was remeasured using the EBM method [59]. The results show baseline regression robustness despite measurement errors.

Changing the model setting. Given TCE’s truncation (0–2.9) by Super-SBM, conventional models fail. Hence, a panel Tobit model [28] was used for re-estimation, showing robust baseline regression results.

Eliminating outliers. A re-regression was performed after a 1% bilateral trimming step for all variables to account for the potential influence of outliers on the results. The findings show that, following outlier corrections, the baseline regression results are reliable.

Eliminating the impact of COVID-19. China’s socioeconomic sector, particularly tourism, was severely impacted by COVID-19 [59]. When 2020–2022 data are excluded, the baseline regression results are still reliable.

3.1.4. Mechanism Analysis

Based on the theoretical analysis presented earlier, tourism labor productivity, the optimization of the tourism industry structure, and the tourism technological level were identified as the pathways through which AI impacts TCE. This section empirically analyzes these pathways using the mediation effect model from Luo and Feng [8]:

$$Me{d}_{it}=\alpha +{\beta }_{1}aA{I}_{it}+{\displaystyle \sum _{j=1}^n{\beta }_j{X}_{it}^j}+{\gamma }_i+{\delta }_t+{\epsilon }_{it}$$

(2)

where ${Med}_{it}$ represents the mechanism variables, primarily including tourism labor productivity (TLP), the rationalization of the tourism industry structure (TSR), the upgrading of the tourism industry structure (TSU), and the tourism technology level (TTEC). Other variables are consistent with Equation (1) (results shown in

Table 5. Mechanism analysis results.

	(1)	(2)	(3)	(4)
Variable	TLP	TSR	TSU	TTEC
AI	0.5613 *	−2.9064 ***	−0.2379	8.5984 ***
	(0.2962)	(0.6621)	(0.2898)	(1.0209)
Urban	−0.0037 *	0.0067	0.0057 ***	0.0027
	(0.0020)	(0.0044)	(0.0019)	(0.0068)
Open	−0.0048 ***	−0.0055 *	−0.0036 ***	0.0089 *
	(0.0013)	(0.0030)	(0.0013)	(0.0046)
Agg	0.7111 ***	−0.0202	−0.1843 ***	3.0245 ***
	(0.0389)	(0.0869)	(0.0380)	(0.1340)
Str	0.0054	0.0193 *	0.0097 **	−0.0103
	(0.0044)	(0.0099)	(0.0043)	(0.0153)
Regl	−0.0313	−0.1456 ***	−0.0878 ***	0.0480
	(0.0209)	(0.0468)	(0.0205)	(0.0722)
Constant	3.1019 ***	0.8140 *	1.0225 ***	−1.9866 ***
	(0.2004)	(0.4479)	(0.1961)	(0.6907)
Year effect	YES	YES	YES	YES
Province effect	YES	YES	YES	YES
Adjusted R-squared	0.8719	0.0710	0.4010	0.7558
Observations	690	690	690	690

Note: * p < 0.1, ** p < 0.05, and *** p < 0.01. Standard error is enclosed in parentheses.

).

Enhancing tourism labor productivity. Column (1) indicates that AI significantly enhances TLP, confirming that AI can improve TCE by increasing tourism labor productivity, thus validating Hypothesis 2. The enhancement in labor productivity allows businesses to produce more with the same amount of labor input [18]. AI improves the quality of tourism workers through substitution effects and skill enhancement effects [21,38]. Consequently, the government should focus on the cultivation and allocation of tourism talent, enhancing the quality of the tourism workforce to stimulate the development of the tourism industry.

Optimizing the tourism industry structure. Column (2) shows a significantly negative impact of AI on the TSR, indicating that AI enhances TCE by promoting the rationalization of the industry structure, thereby confirming Hypothesis 3. The rationalization of the tourism industry structure, which ensures a balanced proportion among tourism sectors, enhances production efficiency and reduces waste. AI enhances the mobility of production factors [22], promoting coordinated development [17] and intensive production in the tourism sector. Consequently, tourism regulatory bodies should leverage AI technology to guide the flow of these factors effectively, driving the rationalization of the tourism industry structure and enhancing TCE through more intensive production practices.

Column (3) indicates that there is no significant correlation between AI and the TSU, suggesting that the upgrading of the tourism industry structure has not yet become a mechanism through which AI influences TCE, and thereby not supporting Hypothesis 4. The underlying cause may reside in the tourism industry’s current lock-in at the lower tiers of the value chain, where transportation and accommodation dominate revenue shares while value-added products and services remain marginal [16]. Transitioning toward higher-value-added positions requires coordinated multi-stakeholder interventions, such as steering high-carbon tourism subsectors toward low-carbon transformation [60], and fostering innovation in value-creating tourism products and experiential services. Therefore, the government should implement comprehensive planning and suitable support policies to encourage high-carbon-emission sectors to switch to non-fossil fuels and foster innovation in tourism content, guiding the transformation of the tourism industry chain toward higher value.

Improving the tourism technology level. Column (4) reveals that AI significantly boosts the TTEC, demonstrating that AI can enhance TCE by improving technology levels, thus confirming Hypothesis 5. Enhanced technology levels enrich tourism products, increase supply capabilities [48], and reduce reliance on resources [48,49], thereby promoting TCE. AI provides a strong basis for the improvement of the tourism technology level [13] and fosters innovation through the accumulation of factors. Consequently, tourism enterprises should continuously improve and enhance their production technologies using AI, increasing energy efficiency.

3.1.5. Heterogeneity Analysis

Region. Location affects how AI affects TCE. Chinese provinces are divided into eastern, central, and western regions, based on research by Wang and Luo [4], and a segmented regression approach is used (

Table 6. Heterogeneity results.

	Eastern Region	Central Region	Western Region	Rich Resource Endowment	Deficient Resource Endowment
	(1)	(2)	(3)	(4)	(5)
Variable	TCE
AI	0.5010 ***	0.9696	0.2962	0.2760	0.8926 ***
	(0.0990)	(0.7260)	(0.4544)	(0.2884)	(0.1683)
Urban	0.0005	−0.0040	−0.0006	−0.0059	−0.0009
	(0.0008)	(0.0035)	(0.0014)	(0.0042)	(0.0007)
Open	−0.0005	−0.0019	0.0025	−0.0011	−0.0005
	(0.0005)	(0.0078)	(0.0018)	(0.0017)	(0.0005)
Agg	0.1321 ***	0.2180 ***	0.2660 ***	0.3129 ***	0.1532 ***
	(0.0216)	(0.0677)	(0.0224)	(0.0417)	(0.0166)
Str	0.0010	−0.0033	−0.0008	−0.0021	0.0080 ***
	(0.0025)	(0.0055)	(0.0028)	(0.0038)	(0.0020)
Regl	−0.0236 **	−0.0041	−0.0165	0.0305	−0.0326 ***
	(0.0101)	(0.0463)	(0.0100)	(0.0281)	(0.0078)
Constant	−0.1258	0.2574	0.0355	0.0915	−0.2417 **
	(0.1199)	(0.2609)	(0.1151)	(0.2115)	(0.0939)
Year effect	YES	YES	YES	YES	YES
Province effect	YES	YES	YES	YES	YES
Adjusted R-squared	0.5960	0.3774	0.5906	0.4438	0.4822
Observations	253	184	253	342	348

Note: ** p < 0.05, and *** p < 0.01. Standard error is enclosed in parentheses.

). Column (1) shows a significant positive effect at the 1% level, while columns (2) and (3) show positive but insignificant coefficients. Thus, AI significantly enhances TCE in the east but not in the center or west. This observation aligns with previous studies [1,49]. Compared to central and western areas, the eastern region has a more robust economy and advanced technology, facilitating the effectiveness of AI in improving TCE. Therefore, local governments should enhance infrastructure and promote regional economic development to create conditions favorable for technological advancements.

Tourism resource endowment. The impact of AI on TCE may exhibit heterogeneity across different levels of tourism resource endowment (Table 6). Complete measurement specifications for tourism resource endowment are documented in Supplemental Material S3. Regression results reveal distinct patterns: column (4) displays a positive but statistically insignificant coefficient for regions with rich resource endowments, while column (5) shows a significantly positive coefficient for endowment-deficient regions. This indicates AI’s capacity to enhance TCE in endowment-deficient regions significantly, but not in endowment-rich regions. This finding aligns with the results of Yang and Fik [61], suggesting that while tourism resources are fundamental to the industry’s development, AI can compensate for limitations due to scarce resources and is more energy-efficient, particularly benefiting endowment-deficient areas’ TCE. Thus, regional governments should tailor their strategies to encourage the penetration of AI into tourism, overcoming the constraints of resource scarcity.

3.1.6. Further Study: Spatial Effect Analysis

As economic interdependence between regions intensifies, the interactive effects among different areas become increasingly apparent: the tourism industry in one region may be influenced by the development of tourism in other regions, leading to spatial autocorrelation. Furthermore, AI’s permeability transcends regional boundaries, enabling cross-regional collaboration and generating spatial spillover effects, impacting tourism development elsewhere. In these situations, using a traditional panel model for regression analysis could skew the estimation results. After several tests, this research used the spatial Durbin model (SDM) to examine spatial impacts in order to solve this problem (see Supplemental Material S4 for details).

Given the presence of feedback effects, the regression coefficients in the SDM may not effectively reveal the impact of AI on TCE. Consequently, the spatial impacts are broken down in this research using partial differential approaches (

Table 7. Spatial spillover effect decomposition.

Variable	Direct Effect	Indirect Effect	Total Effect
AI	0.6836 ***	1.3078 ***	1.9914 ***
	(0.1325)	(0.4001)	(0.4132)
Urban	−0.0002	−0.0002	−0.0004
	(0.0008)	(0.0024)	(0.0026)
Open	−0.0003	0.0020	0.0017
	(0.0006)	(0.0015)	(0.0016)
Agg	0.2081 ***	−0.1504 **	0.0577
	(0.0156)	(0.0600)	(0.0607)
Str	0.0016	0.0054	0.0071
	(0.0018)	(0.0071)	(0.0075)
Regl	−0.0215 **	−0.1194 ***	−0.1409 ***
	(0.0088)	(0.0391)	(0.0409)

Note: ** p < 0.05, and *** p < 0.01. Standard error is enclosed in parentheses.

). The findings indicate that AI has a significant direct effect coefficient of 0.6836 and a spillover effect coefficient of 1.3078. This indicates that AI not only enhances TCE within the local region but also has a spillover effect, improving TCE in neighboring regions. This validates the robustness of the baseline regression results and supports Hypothesis 6.

3.2. Discussion

This study provides robust empirical evidence demonstrating that AI significantly enhances TCE, corroborating existing research highlighting AI’s potential to address sustainable tourism challenges [12,13]. Through rigorous empirical testing, we advance this theoretical understanding. Concurrently, our findings reveal that, even when accounting for the increased energy consumption and carbon emissions resulting from heightened AI computational demands, AI still significantly promotes TCE, underscoring the substantial benefits the tourism industry derives from the adoption of AI [43]. A plausible explanation for this outcome is that AI in the tourism sector primarily constitutes applied AI, which, in contrast to innovative AI, does not necessitate extensive algorithm training and, consequently, lacks significant computational and energy requirements, aligning with prior research [9]. Overall, AI primarily reduces energy consumption and carbon emissions in tourism, thereby fostering TCE by enhancing the carbon emission monitoring capabilities of tourism authorities [8], improving the operational efficiency of tourism enterprises [10], and promoting tourist pro-environmental behavior [37].

Mechanism analysis reveals that tourism labor productivity improvements, tourism industry structure rationalization, and tourism technology advancements serve as significant channels through which AI enhances TCE, consistent with EKC theory [45]. This demonstrates the critical influence of tourism’s scale, structure, and technology effects on AI-driven TCE improvement. However, industrial structure upgrading has not emerged as an effective channel, potentially due to tourism’s persistent low-end positioning, where transportation and accommodation dominate revenue shares while value-added products and services remain marginal [16]. The tourism industry’s transition toward higher value chain positions faces substantial challenges, necessitating coordinated multi-stakeholder interventions, including the low-carbon transformation of high-emission subsectors [60] and innovation in value-added tourism products and experiential services.

The results of the heterogeneity study show that AI’s impact on TCE is not uniform but exhibits disparities, with more pronounced effects in eastern China—a finding consistent with both regional development policies [1,49]. This spatial pattern likely stems from eastern regions’ stronger economic foundations and greater factor endowments, which enhance AI’s capacity to improve TCE. Notably, the effect proves most substantial in regions with limited tourism resource endowments, suggesting AI’s dual capacity to compensate for tourism resource deficiencies [61] while exhibiting superior energy-saving characteristics, thereby driving greater TCE improvements in these areas.

Further spatial effect analysis reveals that AI exhibits significant spillover effects, positively contributing to the TCE of neighboring regions. This finding contrasts with prior research [5], which argued that technological innovation lacks spatial spillover effects on TCE. The discrepancy likely arises because earlier studies measured technological innovation through patent applications, where intellectual property protections may restrict cross-regional technology diffusion. As an emerging disruptive technology, AI demonstrates inherent permeability and diffusivity [50], generating spatial spillover effects on the socioeconomic systems through inter-regional factor flows [62]. This technological characteristic ultimately enhances TCE in both local and neighboring regions.

4. Conclusions and Implications

4.1. Conclusions

This paper draws on the EKC theory, using panel data from 2000–2022 across 30 Chinese provinces, to develop and test AI’s influence on TCE and the mechanisms involved. The empirical findings indicate that AI has markedly enhanced TCE, with their validity retained following rigorous robustness tests. Moreover, improving tourism labor productivity, rationalizing the tourism industry structure, and enhancing the tourism technology level are three channels through which AI influences TCE. Additionally, AI is particularly effective in boosting TCE in eastern regions and areas with deficient tourism resource endowments. Furthermore, AI showed significant spatial spillover effects, significantly promoting TCE in both local and surrounding areas.

4.2. Implications

The findings yield significant managerial implications for tourism policymakers and practitioners. Our benchmark regression analysis robustly demonstrates AI’s substantial enhancement of TCE, underscoring its transformative potential in advancing sustainable tourism development. This empirical insight suggests that tourism authorities should strategically integrate AI development into sectoral policy frameworks to systematically promote smart tourism initiatives. Concurrently, tourism operators should be incentivized to adopt AI technologies for operational efficiency improvements, thereby reducing energy consumption and carbon emissions. Collectively, these measures provide actionable pathways for the tourism sector to contribute substantively to China’s broader carbon mitigation goals through subsystem-level emission reductions.

Secondly, the identification of tourism labor productivity improvements, tourism industry structure rationalization, and tourism technology advancements as key transmission mechanisms not only delineates concrete domains for managerial intervention but also validates the practical applicability of EKC theory in tourism contexts. To maximize AI’s potential in boosting TCE, policymakers should cultivate an enabling environment through (1) investing in AI-oriented education and training programs for tourism workforce development, (2) facilitating factor mobility and promoting structural rationalization of the tourism sector, and (3) propelling intelligent transformation in tourism to accelerate technological upgrading and energy efficiency gains.

Furthermore, the results of the heterogeneity study show significant disparities in AI’s impact on TCE, with particularly pronounced effects observed in eastern China and resource-constrained regions. These findings suggest that policymakers should adopt differentiated smart tourism strategies tailored to regional contexts. For instance, in economically prosperous eastern regions and areas with insufficient tourism resources, policymakers may choose to prioritize investments in AI technologies to maximize their benefits in enhancing TCE. In contrast, in regions where AI’s impact is less discernible, policymakers may allocate more attention to AI-related sustainable tourism initiatives, such as the continued promotion of intelligent low-carbon technologies in tourism, to establish foundational capabilities for future smart tourism development.

Finally, the notable positive spatial spillover effect of AI on TCE underscores the significance of regional cooperation and synergy in the sustainable development of tourism. Consequently, it is imperative to further strengthen inter-regional collaboration in smart technology and the integrated development of the tourism industry. For instance, policymakers should take the lead in establishing cross-regional tourism development alliances to foster a conducive environment for the integrated development of “artificial intelligence + tourism”. Tourism enterprises should promote networked cross-regional cooperation within the tourism industry value chain, thereby enhancing the spatial spillover level of AI on TCE.

4.3. Limitations

The study admits its limits despite its contributions. Primarily, owing to data constraints, the empirical analysis is restricted to provincial-level data in China, which curtails its broader applicability. As data accessibility enhances in the future, it is advisable for subsequent studies to delve into more granular datasets or international contexts. Secondly, anchored in the EKC theory, this paper solely examines three transmission mechanisms. Hence, forthcoming research endeavors may identify additional, potentially more efficacious mechanisms that merit investigation, such as alterations in tourist behavior, AI-powered emission monitoring, or the augmentation of energy efficiency via smart infrastructure.