Abstract
Keywords
Introduction
Climate change and environmental protection have garnered significant global attention, making environmental education (EE) an urgent priority. The rise of Artificial Intelligence-Generated Content (AIGC) presents opportunities to enhance the effectiveness of EE. January 2024 was the hottest month on record, and WMO reports increasing droughts, floods, heatwaves, and mounting economic costs from extreme events (Bateman, 2024; Uralovich et al., 2023), underscoring the urgent need for environmental educational responses. Yet, despite heightened awareness, a critical gap remains: how can translate environmental knowledge and intent into sustained pro-environmental behavior?
Artificial Intelligence has a special advantage in tackling this challenge. Unlike the usual ‘one-size-fits-all’ teaching method, AI can provide adaptive and personalized learning paths to match learners’ prior knowledge and learning pace. For example, AI can monitor students’ learning progress in real time and provide instant formative feedback, as well as generate interactive scenario simulations to help students experience the consequences of environmental choices, which is more beneficial for enhancing environmental education.
Using AI in ecological education can provide engaging and personalized learning experiences, which in turn encourage students to adopt environmentally friendly behaviors. However, challenges such as data privacy, student autonomy, and ethical concerns in AI decision-making must be addressed. Whether AI can be more effective in instilling environmental concepts in students is also a question.
The current research on environmental education (EE) is extensive and varied, focusing on different aspects of it. Higher education in environmental studies now uses diverse teaching methods (van de Wetering et al., 2022). Curricula now cover sustainability, ecology, and environmental policy factors(Bautista-Puig & Sanz-Casado, 2021; Hashinaga et al., 2023; Kohl et al., 2022; Nugmanovna & Gennadievna, 2022).
Teaching methods have become more creative, focusing on hands-on learning and problem-solving for environmental care (Ajaps, 2023). Furthermore, there’s a move to include sustainability in every part of campus life, like how the school is run, how resources are used, in research, and in working with the community (Ali et al., 2020; Sustainable Stanford, 2024). This helps students learn and take active roles in solving environmental problems (Bloyd Null et al., 2023; Null & Asirvatham, 2023).
The research on environmental education (EE) still has not provided precise details (Ardoin et al., 2020). Educators in this field need tailored strategies to understand and teach the subject effectively (Bijlhout et al., 2023). The study of the effect of AI technology on environmental education is still insufficient (Wang et al., 2022).
To address this gap, this study proposes a framework that combines green behavior, adaptable education, and AI to better understand how education influences green behavior. The AI-Driven Green Behavior Model (AIGB) is a new framework that include AI along with factors from psychology and education to show how students turn eco-friendly intentions into real actions. The model emphasizes a complex interplay of internal factors—such as attitudes, subjective norms, perceived behavioral control, and anticipated emotions—with external influences, all converging on the key mediator: desire. This catalyst drives the shift from intention to tangible behavior, highlighting the critical role of a flexible, adaptive educational environment.
There are growing concerns about the use of AI in communications, particularly in education, and whether it could eventually replace the role of humans. This research explores the role of AI in education and finds that while AI can provide vast amounts of information and generate diverse communication styles, the human connection between teachers and students remains irreplaceable and essential for fostering deep and lasting behavioral change.
Literature Review and Conceptual Model
Literature Review
Environmental education (EE) is well-established in promoting awareness, values, and knowledge of sustainability issues through traditional methods such as lectures, field trips, problem-based learning, and campaign-style work. These methods are effective in raising concern and shaping attitudes but often lack personalization, real-time feedback, and clear evidence of sustained behavioral change. Environmental education (EE) is well-established in promoting awareness, values, and knowledge of sustainability issues through traditional methods such as lectures, field trips, problem-based learning, and campaign-style work. These methods are effective in raising concern and shaping attitudes but often lack personalization, real-time feedback, and clear evidence of sustained behavioral change. Ajaps (2023) also highlights that structural and epistemic constraints in higher education can limit the effectiveness of sustainability teaching, emphasizing the need for justice-based approaches.
AI-enhanced and immersive digital approaches are increasingly used in environmental education. Virtual and augmented reality (VR/AR) tools have been shown to enhance environmental empathy, skill transfer, attitudes, and engagement compared to traditional media (T. Xie et al., 2025). A recent systematic review on digital tools for fostering sustainability awareness found that such tools, including VR/AR and mobile apps, positively impact student sustainability awareness, particularly in short interventions (Hajj-Hassan et al., 2024).
Meta-reviews on artificial intelligence in education (AIED) indicate that most research focuses on AI to assist teachers or automate instruction, whereas studies explicitly measuring behavioral change in environmental sustainability over time remain scarce (Mustafa et al., 2024). For instance, immersive field trip–style VR studies in higher education demonstrate improved knowledge retention and positive environmental attitudes, although evidence on sustained behavioral change or scalability is limited (Hamilton et al., 2025).
Integrating pedagogy and behavioral theory points to two under-studied questions in the literature: (1) How does AI-enabled personalization interact with psychological drivers (attitude, subjective norm, perceived behavioral control, anticipated emotions) to form desire and intention? (2) How do adaptive class designs (pedagogical adaptability) and AI tools together improve the translation from intention to sustained pro-environmental behavior?
Theoretical Background
The Model of Goal-Directed Behavior
The Model of Goal-Directed Behavior (MGDB) is a theoretical model that describes the processes leading individuals to engage in a particular behavior. The MGDB expands on the Theory of Planned Behavior (TPB) by introducing additional factors such as past behavior, anticipated emotions, and desires. The model posits that human behavior is the result of a sequential chain of cognitive events, including beliefs, intentions, and desires (Ogren et al., 2020; Perugini & Bagozzi, 2001). It provides a comprehensive approach to understanding why individuals choose to engage in behaviors that are beneficial for the environment. By using MGDB, researchers and policymakers can gain a deeper understanding of the factors that motivate environmentally friendly behavior (Ferreira & Liu, 2023; Meng & Han, 2016).
Adaptability of Teaching
Adaptability in college environmental education encompasses the dynamic and responsive evolution of curricula, teaching methods, and technological integration to address changing environmental issues, student needs, and advancements in various fields (Margolis, 2018; Radó, 2020). This approach involves updating the curriculum with recent sustainability research, applying diverse teaching methods like project-based learning, and adapting to societal and job market trends (Krishnannair & Krishnannair, 2021). This adaptability ensures that students are thoroughly prepared with the necessary knowledge, skills, and attitudes to tackle contemporary and future environmental challenges effectively (Urquiza Gómez et al., 2015).
When connecting the Model of Goal-Directed Behavior and adaptability education to a context like students’ environmentally friendly behavior, the conceptional model can be applied.
The Role and Adoption of Artificial Intelligence in Teaching
Current research explore how AI is opening new possibilities for governance and instruction, aiming to enhance educational quality and access. The focus is on how AI can personalize teaching and learning, facilitate better decision-making, and handle increased educational demands due to the expansion of higher education (Ayeni et al., 2024).
Various studies have been conducted to understand the adoption of AI, using models like the Unified Theory of Acceptance and Use of Technology (UTAUT) and methodologies like the Analytic Hierarchy Process (AHP) to assess factors influencing its use by educators (Andrews et al., 2021; Xue et al., 2024). The findings suggest that AI could significantly aid in the adoption of innovative teaching methods and improve administrative efficiencies (Onesi-Ozigagun et al., 2024). However, challenges such as data privacy, algorithmic bias, and the need for comprehensive teacher training remain critical issues that need to be addressed to fully harness AI’s potential in education (Bukar et al., 2024; Ghimire et al., 2024).
Hypothesis Development and Conceptual Model
Hypothesis Development
The MGDB framework explains that a person’s motivation to do something—called ‘desire’—is shaped by their attitudes, social pressure, expected emotions, and sense of control. This desire then influences what they plan to do, which in turn affects what they actually do (Chiu & Cho, 2022).
The hypothesis of attitude and desire are described as below:
The hypothesis of subjective norm and desire are described as below:
Positive anticipated emotions: The Model of Goal-Directed Behavior (MGDB) emphasizes the role of anticipated emotions as central determinants of desire, which in turn drives behavioral intentions. Among these, positive anticipated emotions are particularly influential because they capture students’ expectations of rewarding feelings associated with engaging in pro-environmental behaviors. For college students, such emotions include optimism about making a difference, empowerment to effect meaningful change, pride in their contributions, hope for a better environmental future, inspiration drawn from sustainability success stories, and gratitude for nature and conservation efforts (Bagozzi et al., 2016; Hagenauer et al., 2018; Odou & Schill, 2020).
In the MGDB framework, these positive emotional anticipations act as motivational drivers that energize and sustain students’ involvement in environmental initiatives. By envisioning the emotional rewards of their actions, students are more likely to translate their attitudes and social influences into a stronger desire to act sustainably.
So the hypothesis can be depicted as this:
Conceptual Model
The conceptual model based on the MGDB (Model of Goal-Directed Behavior) educational adaptability; it also considered the adoption of AI in teaching. In this conceptual model, attitude, subjective norms, perceived behavioral control, anticipated emotions, class adaptability and the adoption of AI in teaching serve as the input factors influencing the central mediating variable, ‘desire’. This desire then influences the dependent variables: students’ eco-friendly intentions and ultimately their actual eco-friendly behaviors. The model suggests a flow of influence from internal and external factors to desire, then to intention, highlighting the complexity of factors that motivate students to act in environmentally responsible ways. This model reflects the incorporation of educational adaptability factors with the psychological constructs of the MGDB to understand and predict eco-friendly behaviors in the higher education context. The model is named as AI-Driven Green Behavior Model (AIGB) on Figure 1.
AI-driven green behavior model.
Methodology
Study Context
Chinese universities have introduced environmental courses to raise students’ awareness and encourage eco-friendly practices. These programs combine theory, practice, and research to support environmental protection and sustainable development. Studying students’ interactions and comments in this context allows us to capture authentic attitudes and behaviors toward environmental issues, providing a real-world basis for modeling environmental learning outcomes.
China University MOOC is an online education platform co-launched by NetEase and the Higher Education Press, offering Massive Open Online Courses (MOOCs) from prestigious Chinese universities to the public (Chinese university MOOC, 2024). Currently, the platform primarily serves Chinese university students and teachers. In fact, almost all Chinese universities integrate MOOC resources into their curricula, making it a standard component of higher education. Therefore, students’ comments on the MOOC platform reflect a broad and diverse range of university learners, which increases the external validity of using these data. Numerous universities have also started offering courses on ecological civilization and environmental protection through this MOOC platform, expanding the scope and impact of environmental education.
This study conducted a course search on China University MOOC using ‘environment’ and ‘ecology’ as keywords, resulting in 128 courses. These courses are either currently being taught or have already been completed. MOOC courses can be repeatedly offered according to the academic term’s schedule, and instructors could update and modify their courses as needed.
The course titles searched on the MOOC, focusing on environmental issues and related topics, like: Ecological Civilization Introduction to Environmental Protection Environmental Monitoring Current Environmental Hot Issues Ecology Environmental Ecology Industrial Ecology Environmental Geology ……
Based on these course titles, a network graph (Figure 2) can be created to visually represent the emphasis and frequency of terms associated with these environmental and ecological courses. The words ‘Environmental’ and ‘Ecology’ appear prominently. This graph visually illustrates the focus of the courses on environmental issues and their relevance to ecological and industrial topics.
Network graph of ecology & environmental courses of MOOC.
Data Collection
Data Acquisition
This study uses crawler technology to search for courses containing the keywords ‘environment’ and ‘ecology’ on the MOOC Platform. Crawler technology was chosen because it enables efficient and comprehensive data collection at scale, which would be infeasible through manual collection. The selected keywords ensure that the dataset is relevant to environmental education topics, directly addressing the research focus. The collected data include online comment content, number of likes, course name, subject classification and title of the main instructor for 128 courses, including ‘Current Hot Environmental Issues’ and ‘Environmental Monitoring’. The study used Python and the constructed thesaurus to analyze the frequency of subject terms in each comment. To ensure the accuracy of the thesaurus, a professor and two undergraduate students proofread and revised the subject terms. A total of 16,001 comments were collected.
Data cleaning is the first step in the analysis process as it eliminates errors such as duplicate and invalid values that can affect subsequent analyses. This research utilizes Excel software’s classification and filtering mechanism to remove irrelevant comments, such as those containing the phrases ‘1111’, ‘don’t want to evaluate’. We also exclude courses that contain the keywords ‘environment’ or ‘ecology’ but are not related to the current study. The deleted invalid comments are categorized into three types, as shown in Table 1.
Examples of Invalid Comments.
The data was cleaned and resulted in 13,583 valid comments with a validity rate of 84.89%, which met the criteria for the empirical test.
The frequency of occurrence of keywords is displayed in the word cloud on Figure 3, with larger words indicating higher frequency and greater representativeness of students’ concerns. Figure 3 displays the word cloud of the comment text data after data cleaning, demonstrating that the cleaned data accurately represents students’ evaluations of the ecological environment course.
The word cloud of the comment text.
Categorizing Vocabulary With Behavioral Variables
During text mining, the study cleaned all collected comments by removing common but non-functional words using a combined stopword list from HIT, SCU, and Baidu (Stopwords, 2022). Researchers refined this list in Excel, creating an enhanced version with 3,961 words, which was then used to filter the comments for cleaner word segmentation.
To test the conceptual model, we quantified the variables involved in the model through text mining. The variables are measured using a word dictionary established for each variable. The frequency of past behavior was measured by the number of times the cause was taught.
395 words with the highest frequency were selected as the source of words in this study. The construction and validation of the word dictionary followed a rigorous, multi-stage process to ensure its validity and reliability. Firstly, each variable, such as attitude or perceived behavioral control, is meticulously defined. Secondly, the researchers carefully categorized each word into the respective variables. This initial categorization, however, faced challenges due to the inherent ambiguity of natural language. For instance, words like ‘change’ could be interpreted in multiple contexts. To address this, a third and crucial step was implemented: having experts review this list and test it on a text sample, refining iteratively based on feedback and performance (Ojeda-Hernández et al., 2023). The criteria for categorizing each vocabulary into different variables were in Table 2.
Vocabulary That Match Each Variable.
Quantifying Behavioral Variables
The frequency of students’ cause-related comments was measured directly, while eight variables were quantified through word frequency analysis. A Python script was used to analyze the occurrence of keywords for each variable across all student comments. This automated process, however, presented specific technical challenges. A primary challenge was ensuring the script accurately accounted for different word forms (e.g., ‘protect’, ‘protects’, ‘protecting’, ‘protection’) to avoid under-counting. To address this, the script employed the Natural Language Toolkit (NLTK) library to perform lemmatization, which reduces words to their base or dictionary form. Furthermore, to validate the accuracy of the automated quantification, a manual spot-check was conducted. A random sample of 2% of the comments was independently coded by two researchers using the final dictionary. The high inter-coder agreement (Cohen’s kappa > .85) confirmed the reliability of the automated method. This method offers a systematic and data-driven way to assess the variables, as illustrated in Figure 4.
Quantifying process of text.
Empirical Analysis
Partial Least Square-Structural Equation Modeling
This study uses PLS-SEM to test factors shaping students’ green behavior because it is prediction-oriented, tolerates modest samples and non-normal data, and handles multicollinearity and complex path models. PLS-SEM was chosen because it is particularly suitable for testing complex theoretical models with multiple latent variables, such as the AI-Driven Green Behavior Model (AIGB), and allows the simultaneous assessment of relationships between psychological, social, and educational factors. This method provides a robust and data-driven foundation for the subsequent qualitative investigation.
Model Test
Approximate fit: SRMR = 0.052 (<0.08), indicating acceptable global fit. Predictive relevance:
Hypothesis Tests.
Results
In the context of the conceptual model, the test results specifically indicate the following:
The Explanatory Study of Influencing Factors
Based on the quantitative results, three key hypotheses that had not yet been tested were identified. To delve deeper into these hypotheses, student interviews were conducted. One area of investigation was how social environmental norms affect students’ motivation to get involved in environmental issues. Another topic explored was whether worries about environmental degradation inspire students to take action. Additionally, the impact of AI on students’ environmentally friendly behaviors was examined, along with the reason why it doesn’t directly boost their inclination to engage with environmental concerns. The purpose of these interviews was to provide deeper insights and supplement the initial quantitative results.
Economic social researchers prefer grounded theory due to its distinct methodological processes, the capability to form unique categories, and a more ‘objective’ approach to qualitative research (Matteucci & Gnoth, 2017). Prior theories on MGDB contributed to the creation of codes that were instrumental in organizing the data. Adhering to key aspects of grounded theory, this research employed an additional coding analysis using NVivo 12. This was done to examine the connections between various factors influencing students’ eco-friendly behavior.
Data Collection
Semi-Structured Interviews Design
This study’s data were derived from semi-structured, in-depth interviews. These interviews allowed for a concentrated exploration of participants’ perspectives, fostering an environment where open and engaging dialog could flourish between the interviewer and interviewee. The interviews were conducted by the first and second authors, who are college educators and students. Their dual roles enable them to view the subject matter from both teacher and student viewpoints. Semi-structured interviews were chosen because they provide both structure and flexibility—ensuring comparability across participants while allowing deeper exploration of individual perspectives.
To maintain the integrity and efficacy of the interviews, this study took measures to mitigate potential biases during the interview stages, focusing on both the interview design and specific process controls. The interview protocol comprises three primary categories of questions. The initial set focuses on students’ general awareness and views regarding environmental protection. The second segment delves into the students’ feels about the subject norm on environmental protection. Lastly, the third group of questions investigates the AI related factors influencing their ecofriendly behaviors. The interview protocol is in Appendix 1.
Sources of Data
The researchers’ college has enough students available for interviews. However, to ensure a diverse range of perspectives aligned with the study objectives, purposive sampling was employed. Invitations for the interviews were distributed across student QQ groups from different universities. The students’ names and key personal identifiers were kept confidential. Thirty students accepted the invitation, participants were considered qualified if they were currently enrolled university students, had taken at least one course related to environmental education or sustainability, and had prior experience with online learning platforms or exposure to AI-assisted educational tools. Based on these criteria, 25 students were deemed qualified for the interview.
In total, 24 students were interviewed for this study. Each interview lasted approximately 8 to 15 min and took place between February 9th and 21st, 2024. After obtaining consent from the interviewees, the entire interview process was recorded. The researchers listened to the audio recordings and used software to transcribe them into text. Upon reviewing the interview transcripts, six were excluded because the interviews were too short (less than 2 min) or the content deviated significantly from the research topic.. Ultimately, 18 interview transcripts, comprising over 45,000 words, were considered valid for the study. The number of valid transcripts reached a level of data saturation, as no new themes or insights emerged from the later interviews. The demographic characteristics of interviewees are on Table 4.
Demographic Characteristics of Interviewees (
Qualitative Data Analysis
Grounded theory, originally designed for systematic qualitative data analysis and theory building, has gained widespread use in social science research for various purposes. Its applications now include exploring and understanding complex phenomena, such as causality, highlighting its versatility and effectiveness in studying social dynamics.
Nvivo 12.0 software was utilized to examine students’ perceptions of sustainable causes and their attitudes toward eco-friendly activities. Interview records was imported into the software, conducting a thorough review and coding items that matched the Adaptive Environmental Intentions Model (AEIM). These codes were based on the model’s factors and the data from the interviews, enabling an organized and systematic analysis. The study involved using Nvivo’s query function to explore relationships among different codes and to identify connections with the conceptual model and the interview data. The coding was regularly reviewed and revised for accuracy and consistency. The specific codes from the semi-structured in-depth interviews are provided in Appendix 2.
Results of Study 2
The Negative Influence of Subjective Norms
Students might be negatively influenced by environmental terms if they perceive them as too abstract or distant, leading to a lack of emotional connection due to their apparent lack of relevance to everyday life. Additionally, the vast and complex information surrounding environmental issues can be overwhelming, potentially making students feel powerless to effect positive change. This sense of overwhelm can be compounded by a negativity bias, where the terms are frequently associated with adverse outcomes like pollution, habitat destruction, or climate change, which could foster feelings of hopelessness or a fatalistic outlook toward environmental protection efforts. As one interviewee depicted: When we first started implementing garbage classification, although there were many voices of social support, I also saw many people around me with skeptical attitudes and resistant behaviors. In the absence of transparent information and feasible methods, although I agree with green behavior, I will remain skeptical about specific measures.
The Complex Effects of Emotions
Negative emotions can indeed have complex effects on motivation and behavior. Anxiety and anger about industrial pollution can complexly affect people’s desire to protect the environment. When overwhelmed by issues like pollution, individuals might experience ‘eco-anxiety’, a sense of helplessness and fear that can lead to feelings of despair or powerlessness, decreasing their motivation to act. Anger might spur action but can also cause burnout or a focus on blaming rather than engaging in personal actions. Moreover, some people might avoid the issue altogether to manage their emotional state. However, these emotions can also drive urgency and awareness, potentially increasing engagement in environmental protection. Effective environmental movements can address these emotional responses by providing emotional support and practical engagement methods, helping individuals feel empowered and capable of contributing to meaningful solutions. As on interviewee depicted: There are reports of factories emitting polluted and even poisonous gases into the atmosphere, which is appalling. At times, I find myself improperly disposing of large amounts of food. although it brings a sense of guilt, I also experience a sense of happiness. Are you referring to the impact of attending environmental protection classes on me? Perhaps it would help me view those negative impacts more rationally.
Enhancing Sophistication in AI to Foster Eco-Friendly Behaviors in Education
The adoption of artificial intelligence (AI) in teaching environmental causes does not necessarily guarantee an increase in eco-friendly behaviors among students. While AI can personalize learning and handle extensive data efficiently, the impact on fostering pro-environmental attitudes depends on the compelling nature of the content, how it addresses behavioral change factors beyond mere knowledge, and integrates social norms and perceived control over outcomes. Furthermore, AI lacks the human element often crucial in inspiring students, such as a teacher’s passion and commitment. Additionally, ethical and practical issues like data privacy and the digital divide may affect the effective implementation and reception of AI in educational settings. Thus, for AI to truly foster environmentally conscious attitudes, it must be part of a broader strategy that includes addressing motivational, social, and practical aspects of behavior change. One student said: Teachers may utilize AI-generated videos and images in class, and sometimes they discuss the spatial temporal characteristics of climate change using big data analysis. While these technologies fascinate me, I believe they are more reflective of technological advancement than they are of actual contributions to environmental protection.
Mapping of Interview Themes to SEM Paths
Table 5 presents a systematic mapping of specific, unexpected findings from the Structural Equation Modeling (SEM) analysis onto the qualitative themes derived from interviews. This integration provides explanatory insights into the causal mechanisms behind the quantitative results.
Mapping of Unexpected SEM Findings to Qualitative Themes.
Discussion
This study introduced the AI-Driven Green Behavior Model (AIGB), extending the Model of Goal-Directed Behavior (MGDB) by integrating adaptability in environmental education and AI adoption. The empirical results confirm that desire is the strongest predictor of eco-friendly intention, reinforcing MGDB’s emphasis on desire as the motivational core.
The findings show that attitude, subjective norm, perceived behavioral control, and anticipated emotions significantly influenced desire, aligning with prior MGDB research and indicating that students’ beliefs, social context, and emotions collectively shape their motivation for sustainable action. Notably, positive anticipated emotions exerted a stronger effect than negative anticipated emotions, suggesting that fostering hope, pride, and optimism is more effective than relying on guilt or fear appeals in environmental education.
The results also indicate that AI adoption positively moderates the relationship between adaptability in education and desire. This suggests that AI-based tools, when integrated into flexible teaching approaches, enhance students’ motivation by providing personalization and immediate feedback. However, AI adoption did not directly influence behavioral intention, underscoring its role as a facilitator rather than a primary motivational driver. This supports the argument that while AI enriches the learning environment, it cannot replace the emotional and inspirational impact of human educators.
Finally, instructional adaptability itself had a significant effect on desire and intention, highlighting the importance of curricula that respond to changing environmental challenges and student needs. This reinforces that dynamic, responsive teaching—augmented but not replaced by AI—creates the most fertile ground for cultivating eco-friendly behavior.
Conclusion
Theoretical Implications
The theoretical contributions of this research are multifaceted, primarily introducing and elaborating on the AI-Driven Green Behavior Model (AIGB). This new framework integrates psychological and educational factors, offering a comprehensive understanding of how education influences students’ desires and actions toward eco-friendly behaviors. Key theoretical contributions include:
Firstly, integration of AI with Goal-Directed Behavior and Class Adaptability. The AIGB is grounded in established theories, including the Model of Goal-Directed Behavior and theories of class adaptability. It introduces the adoption of AI into study. It uniquely combines these perspectives to examine the complex interactions between various factors that drive environmental behaviors. It emphasizes the central role of education in promoting student desire, facilitating the transition from intentions to actions.
Secondly, enhancing understanding of AI and human roles in environmental education. The study clarifies how AI and educators complement each other in fostering eco-conscious behaviors, advocating for a balanced and supportive integration of technology in education.
Thirdly, tested adaptability in educational practices in environmental sector. A significant theoretical contribution is the emphasis on adaptability in environmental education. AIGB highlights the need for flexible curricula, teaching methods, and technology use to meet evolving environmental challenges, promoting engagement, critical thinking, and long-term environmental stewardship.
Our results support the Model of Goal-Directed Behavior, showing that attitudes, norms, control, and emotions predict desire (Chao, 2022). Curriculum adaptability further strengthens desire and intention in environmental education (Du & Ashraf, 2025). Importantly, while AI enhances the effect of instructional adaptability on desire, it does not directly drive intention—highlighting the unique motivational role of human educators (Pang et al., 2025).
Practical Implications
In environmental education practice, this research outlines several practical implications:
Firstly, employ holistic educational strategies. Educators should address students’ emotional, social, and cognitive factors to cultivate eco-friendly intentions and actions. In addition, AI tools can support these strategies—for example, using intelligent learning platforms to track students’ engagement with sustainability content, adaptive quizzes to reinforce eco-friendly behaviors, or virtual simulations to demonstrate environmental consequences. These applications help make learning more interactive, personalized, and directly linked to students’ environmental actions.
Secondly, continuously update educational content and methods. The study advocates for continuously updating educational content and teaching methods to include the latest sustainability research and innovative strategies like personalized AI adoption. Specific classroom applications could include AI-driven recommendation systems that suggest eco-friendly projects based on student interests, chatbots that guide students through environmental problem-solving activities, and AI analytics to monitor behavior changes over time. This ensures that education remains relevant and engaging, thereby enhancing student engagement and fostering critical thinking and innovation.
Thirdly, respond to evolving environmental and societal needs. The AIGB underscores the necessity for education systems to be adaptable, personalized, and responding to the evolving environmental challenges, societal shifts. This adaptability is crucial for preparing students to effectively tackle current and future environmental problems.
Moreover, our findings highlight that positive anticipated emotion exerts the strongest effect on students’ desire, while classroom adaptability strongly influences intention. Therefore, educators and policymakers should prioritize interventions that enhance students’ positive emotions—for instance, through gamified tasks, real-world success stories, or immersive simulations—and ensure high adaptability in teaching by using AI tools to personalize content and adjust learning pace. These targeted strategies can more effectively translate intentions into sustained eco-friendly behaviors.
This study aligns with prior research emphasizing the role of emotions in eco-friendly behaviors, while extending the literature by introducing AI as a supportive tool in environmental education.
Limitations and Future Research
The limitations of this research include:
Lack of Teachers’ Perspectives. The study primarily relies on students’ viewpoints, utilizing text analysis and interviews from the student standpoint. Teachers’ insights, which could provide valuable context and depth to the findings, were not included. As instructors are the designers of instructional activities and the implementers of AI tools, the lack of their insights limits a comprehensive understanding of the practical challenges in AI adoption, educator acceptance, and human-AI collaborative teaching models. Second, the data were gathered within the higher education context of a single country (China), where unique cultural values, educational policies, and focal environmental concerns may affect the generalizability of the findings to other educational and cultural settings.
Insufficient Exploration of College-Specific Social Norms. While social norms play a crucial role in shaping eco-friendly behaviors, this research does not thoroughly investigate these norms within the specific context of colleges. A deeper analysis could reveal important influences on students’ environmental attitudes and actions.
Further study is needed to explore the repetitive teaching. The research mentions the effects of teachers delivering the same environmental content to different student groups, yet this warrants deeper investigation. A more comprehensive understanding of how repetitive teaching by teachers affects various cohorts’ engagement and learning outcomes is crucial for refining educational strategies. Future research should also incorporate teachers’ perspectives to explore the challenges and strategies educators encounter when integrating AI into teaching, thereby enabling a more holistic assessment of AI’s role in education. Furthermore, cross-regional comparative studies in more diverse cultural and educational contexts are recommended to test the robustness of the AIGB model and investigate how socio-cultural factors moderate the influence of AI on eco-friendly behavioral intentions.
Footnotes
ORCID iD
