How Digital Well-Being and Generative AI Achieve Student Academic Success and Sustainable Development Goals

Summary

This study finds that digital well-being and generative AI boost English as a Foreign language learning engagement, leading to better academic performance. Higher achievement increases awareness of sustainable development goals, linking educational technology to both scholastic and sustainability outcomes.

Abstract

This study examines how digital well-being and generative AI adoption contribute to English as a Foreign language (EFL) students' academic success and engagement with Sustainable Development Goals (SDGs). Using a quantitative cross-sectional design, data were collected from 440 participants through validated scales measuring digital well-being, AI usage, learning engagement, academic performance, and SDG awareness. Partial least squares structural equation modeling (PLS-SEM) was employed to analyze the relationships between these constructs while assessing the model's predictive validity and robustness. The findings reveal that both digital well-being and generative AI significantly enhance EFL learning engagement, which in turn drives academic success. Furthermore, academic achievement positively influences students' SDG awareness, highlighting the interconnectedness of digital education and sustainability outcomes. The study identifies learning engagement as a critical mediator between technology-enhanced learning and academic performance. These results underscore the importance of integrating digital wellness strategies and AI literacy into educational frameworks to foster both scholastic achievement and global citizenship. These are achieved by designing curricula and pedagogies that simultaneously teach students to manage their digital lives for effective learning, and equip them to leverage and critique advanced technologies to understand and act upon the world's most pressing challenges.

Introduction

In an era marked by rapid technological advancements, the integration of digital well-being and generative artificial intelligence (GAI) into education has emerged as a transformative force with far-reaching implications for student academic achievement and the United Nations' Sustainable Development Goals (SDGs)¹. While digital well-being focuses on promoting healthy technology usage to enhance cognitive and emotional balance, GAI introduces unprecedented opportunities for personalized learning², accessibility, and educational efficiency³. Nevertheless, as much as there is a developing line of interest in these areas across the world, the existing body of knowledge is scattered, with little focus on the future promise in their synergies to developing academic success and promoting global sustainability milestones⁴. This research helps to fill this much-needed gap by examining specifically how responsible use of GAI, when informed by digital well-being principles, can maximize learning outcomes but simultaneously support key SDGs, especially SDG 4 (Quality Education), SDG 3 (Good Health and Well-Being), and SDG 10 (Reduced Inequalities).

A review of current literature reveals several underexplored dimensions. First, while studies on digital well-being emphasize the risks of excessive screen time, digital distraction, and mental health concerns among students⁵, few have examined how structured digital mindfulness practices can coexist with AI-driven learning tools⁶. Second, research on GAI in education predominantly highlights its efficiency in automating assessments, generating content, and enabling adaptive learning⁷, yet overlooks its psychological and behavioral impacts on students' well-being⁸. Third, although the SDGs framework advocates for inclusive and equitable education, minimal attention has been paid to how AI-enhanced learning environments, when designed with well-being in mind, can reduce educational disparities and promote sustainable learning practices⁹. These gaps underscore the need for an integrative approach that harmonizes technological innovation with human-centric well-being strategies.

This study addresses these limitations by proposing a novel framework that aligns digital well-being principles with GAI applications in education. In contrast to a previous study¹⁰, which analyzed these notions independently, our study considers their dynamics as such, thus presenting empirical evidence of how such AI can be morally implemented to support, rather than erode, student focus, motivation, and long-term academic achievement. In this way, the present manuscript will add to the further discussions in the spheres of educational technology, psychology, and sustainability science, providing a current variant that will interest a large number of readers, researchers, policymakers, and educators interested in finding evidence-based approaches to striking the right balance between focusing on the integration of technologies and sustainable, holistic development of learners and students.

The theoretical contributions of this research are threefold. First, it advances the discourse on digital well-being by introducing a new dimension of AI-mediated well-being, which explores how generative AI can be calibrated to minimize cognitive overload and promote mindful engagement¹¹. Second, it builds on the literature concerning AI in education, as it integrates considerations of well-being as an essential design feature, disrupting the existing efficiency-driven discourses¹². Third, it enhances the quality of the SDGs scholarship by showing how updated education systems through AI can become drivers of sustainable development, especially in closing digital divides and ensuring that learners become mentally resilient. These innovations of the theorist put our research among the avant-garde of interdisciplinary studies, providing a perspective on the current interdependence of technology, education, and sustainability on a global scale. Academically, the study gives a holistic framework of how it is possible to assess the dual outcome of GAI and digital well-being on learning results through empirical data¹³. Among the policymakers, it presents them with practical recommendations on how the use of AI in education can be governed in an ethical way, protecting the safety of students¹⁴. To educators, it shows evidence-based methods of how to implement the AI tool that can support, but not violate, the effectiveness of pedagogy¹⁵. In addition, the connection between these results and the SDGs gives a bigger picture of the overall implications of technologically advanced education in society, hence the reasons why digital pioneering should be aligned with sustainable human development.

The emergence of rapidly spreading digital technologies and generative artificial intelligence (AI) in education triggered a massive intellectual interest in their combined effect on the student academic performance level and the attainment of the whole host of societal objectives, including Sustainable Development Goal¹⁶. Though these technological developments hold a revolutionary potential, the applications of such technologies into the pedagogical environments are a rich field of polarities and controversial logistics, where existing studies should be critically synthesized to define the insights and the limitations of the upcoming studies¹⁷. This literature review will review the insights of three related fields: (1) digital well-being and its educational implications, (2) generative AI and its influence on educational practice, and (3) the compatibility of the given technological interventions with SDGs, specifically those that refer to quality education, health, and equity. Digital well-being has emerged as a critical area of study in increasingly technology-immersed learning environments¹⁸. While research highlights risks such as cognitive overload, digital distraction, and passive consumption associated with screen overuse, there is growing recognition of the potential to mitigate these effects¹⁹. Strategic interventions such as digital mindfulness, intentional technology use, and metacognitive reflection can foster meaningful engagement and protect mental and emotional health. Importantly, digital well-being extends beyond minimizing screen time; it involves cultivating a balanced, self-determined relationship with technology²⁰. This perspective is especially relevant in AI-enhanced learning contexts, where ethical challenges like algorithmic bias and over-reliance on automation can directly impact students' sense of competence, autonomy²¹, and relatedness, core psychological needs essential for motivation and well-being²². Bridging this gap requires a framework that not only addresses the risks of digital interaction but also proactively integrates well-being principles into the design and use of generative AI tools in education.

With the newness of generative AI, which creates human-like text, photos, and games that respond to actions, a new realm of possibilities of personalized and adaptive learning has opened up²³. Research has confirmed that it is effective in the automation of administrative processes, creation of personalized learning resources, giving real-time feedback, and enhancing both efficiency and accessibility of instructions²⁴. As an example, AI-based tools such as Intelligent Tutoring Systems and ChatGPT have been demonstrated to facilitate differentiated instruction²⁵. That being said, there also exist ethical and psychological implications of the pedagogical use of generative AI, such as the problem of academic integrity, bias in the algorithm, and the possibility of students losing their critical thinking faculties when relying too heavily on AI-generated works²⁶. Although the current research provides vast amounts of information on the functioning abilities of AI in education, it neglects to explore the psychological and behavioral consequences on students, especially when it comes to digital well-being²⁷. This lapse points to the necessity of a closer consideration of the ways generative AI might be utilized to benefit cognitive and emotional wellness in students as opposed to harming them²⁸. The combination of digital well-being and generative AI also acquires new importance when filtered through the prism of the SDGs, through which the world has developed a global calling of equitable and sustainable development. On its part, SDG 4 (Quality Education) requires equitable and inclusive access to learning opportunities²⁹. In the same manner, SDG 3 (Good Health and Well-Being) and SDG 10 (Reduced Inequalities) stress the need to take care of mental health and promote social equity, which are both related to digital experiences of students³⁰. AI and SDGs, however, are currently discussed in a techno-positivistic way, praising the scalability of AI interventions and ignoring psychosocial aspects of technology utilization³¹. To illustrate, although AI can personalize the learning experience of underserved populations, its implementation lacks well-being protection, which can increase digital fatigue and increase the inequality gap between students who have different access to technology and self-regulation abilities³². The tension between innovation and well-being adds to the fact that this gap in literature is vital, the existence of a consistent, coherent framework to marry AI-driven educational progress to the humanistic ideals of the SDGs³³.

Educational psychology, human-computer interaction, and sustainability studies represent some of the theoretical approaches that can provide useful lenses with which to bridge these two different strands of research. An example of such a theory is Self-Determination theory (SDT), according to which healthy learning and well-being can be achieved when people experience the satisfaction of autonomy, competence, and relatedness³⁴. When applied to digital environments, SDT implies that the generative AI might increase both autonomy via tailored learning programs and competence due to the adaptability of feedback, but must also be designed to support relatedness by transcending social displacement³⁵. Likewise, the notion of digital balance, the condition in which the use of technology matches both individual and collective wellness, offers a normative basis regarding assessing the role of AI in education³⁶. Nonetheless, these theoretical constructs have been rarely embodied into empirical analysis of AI-enabled learning environments, which leaves questions with regards to how it is possible to engineer AI systems which are inherently oriented towards digital well-being³⁷. A critical review of the literature shows that there are a few unresolved tensions and areas of future development³⁸. To begin with, although the matter of digital well-being smartly focuses on restraint and mindfulness, generative AI is the issue of engagement and immersion, the paradox of which needs empirical solutions³⁹. Second, the ethical discussion around AI in education has been very much based on data privacy and algorithmic transparency without giving much thought to the psychological implications of the interaction with AI⁴⁰. Third, the focus on sustainability and equity provided by the SDGs does not comprehensively relate to the technical and pedagogical aspects of the actual implementation of AI. To overcome this disability, it is important to crosscut disciplines and use the insights of cognitive science, behavioral economics, and educational policy to come up with frameworks that are not only technologically stable but also humane. Figure 1 shows the conceptual framework of the present study, which includes all variables and their path relationship with one another. In this model hypothetical relationship among study variables is given in the conceptual framework.

Protocol

This study did not require formal ethical approval from an Institutional Review Board (IRB) as it involved anonymous surveys and/or interviews with participants who provided informed consent. No personal or sensitive data was collected, and all responses were treated confidentially for research purposes only. Participation was voluntary, and respondents were informed of the study's objectives before providing their input.

The study employed a quantitative, cross-sectional design to investigate the relationships among digital well-being, generative AI usage, and academic outcomes among English as a Foreign Language (EFL) students. Data were collected via an online survey from a purposively sampled population of 440 undergraduate EFL students across multiple Chinese universities who actively used generative AI tools in their learning. The structured questionnaire utilized validated scales, including a 12-item measure for digital well-being⁴¹, an 8-item scale for generative AI usage, a self-perceived academic success scale⁴², and measures for SDG awareness and learning engagement, all employing 5-point Likert responses. Data analysis followed a two-phase protocol using SPSS v.27 for descriptive statistics and reliability checks, and SmartPLS v.4.0 for Partial Least Squares Structural Equation Modeling (PLS-SEM) to evaluate the measurement and structural models, with bootstrapping applied to test path significance.

Research design
The research design utilized in this study is quantitative and cross-sectional to determine how digital well-being, generative AI, student academic success, and SDGs relate to each other among EFL students. A cross-sectional design enables the researcher to gather data at one point in time and gives a picture of the variables being studied. The quantitative research design makes the results grounded on measurable and statistically analyzable information to make the results more reliable and generalized. This pattern is especially applicable to analyze the direct and indirect effects of digital well-being and generative AI on academic achievement and SDGs because it will enable the employment of sophisticated statistical procedures like structural equation modelling.

Population and sampling
The target population for this study consisted of undergraduate students enrolled in English as a Foreign Language (EFL) programs across higher education institutions in China. To participate, students had to meet the following inclusion criteria: being currently enrolled as an undergraduate student, actively using generative AI tools (such as ChatGPT, Copilot, or similar platforms for writing assistance, research, or content generation) as a part of their regular academic learning process, and providing informed consent. Exclusion criteria were not using generative AI for academic purposes, being a postgraduate or non-degree student, and providing incomplete survey responses.

A purposive sampling technique was employed to recruit participants who met these criteria, ensuring the meaningful measurement of the core AI usage variable. The final sample comprised N = 440 participants, a size deemed adequate for the statistical robustness required for Partial Least Squares Structural Equation Modeling (PLS-SEM). To enhance representativeness and generalizability, recruitment spanned multiple universities and diverse academic disciplines (Humanities, Social Sciences, STEM). Demographic data for the sample are as follows: Age ranged from 18 to 24 years (M = 20.4, SD = 1.6); Sex was distributed as 58% female (n = 255) and 42% male (n = 185); Year of Study included first-year (22%, n = 97), second-year (31%, n = 136), third-year (28%, n = 123), and fourth-year (19%, n = 84) students. Information on weight was not collected as it was not relevant to the study's constructs.

Data collection procedure
Data collection was conducted via an anonymous online survey hosted on a secure platform (e.g., Qualtrics). Potential participants across the targeted universities were invited through institutional email lists, announcements on Learning Management Systems (e.g., Moodle, Blackboard), and relevant student social media groups. The invitation clearly outlined the study's purpose, voluntary nature, estimated completion time (15-20 minutes), and confidentiality assurances. The first page of the survey presented a detailed informed consent form; only participants who provided digital consent proceeded to the questionnaire. To mitigate order bias, the survey sections were presented in a randomized sequence for each respondent. The data collection window was open for two weeks. A single reminder was sent via the original channels one week after the initial invitation to encourage participation and improve response rates. Following the collection period, the dataset was screened for completeness, with removal of responses that were substantially incomplete (e.g., >20% missing data). Subsequent checks for univariate and multivariate outliers were performed prior to final analysis to ensure data quality.

Instrumentation and measures
The study utilized a structured questionnaire to measure key variables, including digital well-being, generative AI usage, student academic success, SDGs awareness, and EFL students' learning engagement. Digital well-being was measured using Arslankara et al.⁴¹, 12-item scale, including subscales for responsibility (4 items), satisfaction (4 items), and wellness (4 items). Generative AI usage was operationalized through an 8-item scale⁴³, focusing on ChatGPT adoption frequency. In the present study, academic performance was evaluated using a self-perception scale adapted from Yu et al.⁴², as cited in Mehrvarz et al.⁴⁴. This instrument comprises four items. SDGs awareness was measured using a validated scale adapted from Atmaca et al.⁴⁵, comprising three key dimensions: Economic sustainability (12 items), social sustainability (9 items), and environmental sustainability (14 items), by students' awareness and participation in activities. Finally, EFL language learning engagement was evaluated using a 9-item scale from Eerdemutu et al.⁴⁶. All scales employed 5-point Likert responses, with pilot testing (n=30) confirming reliability (Cronbach's α > 0.82) and confirmatory factor analysis validating construct distinctiveness.

Data analysis plan
The collected data were analyzed using a combination of SPSS (v.27) and SmartPLS (v.4.0) through a two-phase analytical approach. First, SPSS was employed for preliminary analyses, including descriptive statistics (means, standard deviations), reliability testing (Cronbach's alpha), and correlation matrices to examine baseline relationships between variables. Second, SmartPLS, a partial least squares structural equation modelling (PLS-SEM) software, was utilized to test the hypothesized model. PLS-SEM was chosen for its ability to handle complex relationships among latent variables and its robustness with non-normally distributed data. The analysis included assessing measurement model validity (convergent and discriminant validity) and structural model significance (path coefficients, R² values, and predictive relevance). Bootstrapping was performed to test the statistical significance of the paths.

Results

The results provide robust empirical support for the study's core hypotheses, confirming a significant sequential pathway where digital well-being and generative AI usage positively influence EFL learning engagement, which in turn drives academic success and subsequently fosters greater engagement with Sustainable Development Goals (SDGs). Specifically, both digital well-being (β=0.32, p<0.001) and generative AI usage (β=0.28, p<0.001) had strong direct effects on learning engagement, explaining 41% of its variance...

Discussion

Prior research has largely examined digital well-being, AI in education, and Education for Sustainable Development (ESD) in isolation⁴⁷. A significant gap existed in understanding how personal digital competencies intersect with emerging technologies to drive not just scholastic outcomes, but also global citizenship. This study uniquely bridges these domains. We demonstrate that digital well-being (β = 0.32, p<0.001) and GAI adoption (β = 0.28, p<0.001) are significant, complemen...

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
Academic Success Scale	Adapted from Yu et al. (2010)	N/A	Survey Instrument
Blindfolding Procedure (for Q2)	SmartPLS	Internal Routine	Analytical Procedure
Bootstrapping Procedure	SmartPLS	Internal Routine	Analytical Procedure
Composite Reliability Assessment	SmartPLS	Internal Routine	Analytical Procedure
Correlation Matrix Analysis	IBM SPSS Statistics	Internal Routine	Analytical Procedure
Cronbach's Alpha Reliability Analysis	IBM SPSS Statistics	Internal Routine	Analytical Procedure
Data Storage System	Institutional IT	Secure Server	Equipment & Material
Descriptive Statistics Analysis	IBM SPSS Statistics	Internal Routine	Analytical Procedure
Digital Well-Being Scale	Arslankara et al. (2022)	N/A	Survey Instrument
EFL Learning Engagement Scale	Eerdemutu et al. (2024)	N/A	Survey Instrument
Fornell-Larcker Criterion Assessment	SmartPLS	Internal Routine	Analytical Procedure
Generative AI Usage Scale	Adapted from Abbas et al. (2024)	N/A	Survey Instrument
Heterotrait-Monotrait (HTMT) Ratio Assessment	SmartPLS	Internal Routine	Analytical Procedure
IBM SPSS Statistics	IBM	Version 27	Software
Informed Consent Documentation	Research Team	Electronic Form	Equipment & Material
Online Questionnaire	Research Team	Self-Administered	Equipment & Material
PLS Algorithm	SmartPLS	Internal Routine	Analytical Procedure
PLS-SEM Software	SmartPLS	Version 4.0	Software
SDG Awareness Scale	Adapted from Atmaca et al. (2019)	N/A	Survey Instrument
SmartPLS	SmartPLS GmbH	Version 4.0	Software