ChatGPT for academic assignments: opportunities, challenges, and student perspectives in the digital classroom

The rise of ChatGPT as a powerful generative AI tool is steadily reshaping academic life, giving students new ways to improve their writing, deepen conceptual understanding, and access personalized feedback almost instantly. Studies reflect both enthusiasm and caution in this shift. Cornish and Larter (2024) and Moheno et al. (2024) report that students appreciate how ChatGPT helps them clarify ideas and receive quick guidance, yet Al-Sofi (2024) reminds us that concerns about plagiarism, accuracy, and responsible use remain very real. Faculty experiences also reveal this balance: Al Muhanna (2025) finds that many educators view ChatGPT positively but still feel the need for structured training, while Damaševičius (2024) stresses that human oversight is essential to ensure fairness in academic evaluations. At the same time, research by Leelavathi and Surendhranatha (2024) and Wang (2025) shows that, when used thoughtfully, ChatGPT can encourage critical thinking, creativity, communication, and self-directed learning skills that are central to meaningful academic development.

Building on this growing body of work, the present study focuses on three constructs digital literacy, ethical awareness, and perceived usefulness that naturally extend from established theoretical and empirical discussions. Digital literacy reflects students’ ability to evaluate and work with digital information, a skill increasingly necessary for making sense of AI-generated content. Ethical awareness emerges from long-standing conversations about academic integrity, responsible digital behavior, authorship, and data privacy, all of which have become more urgent in the age of AI. Perceived usefulness draws from technology acceptance research, explaining how students’ beliefs about the value of a tool influence their willingness to adopt it and use it to support learning. These constructs are therefore built on existing dimensions rather than being entirely new, representing a logical extension of what scholars have already established about technology and learning.

Even so, several gaps persist in current scholarship. Few studies examine how digital literacy, ethical awareness, and perceived usefulness interact to influence student performance, despite the growing dependence on AI in academic tasks. Much of the existing research also focuses mainly on traditional university students, leaving limited understanding of how nontraditional learners engage with AI tools (Yang et al., 2025). Emotional and motivational aspects of AI usage, highlighted by Hickman and Stoica (2025), remain underexplored as well. Additionally, institutional challenges including data privacy, infrastructure readiness, and policy clarity continue to shape how effectively AI can be integrated into learning environments. In response to these gaps, the present study examines how the three chosen constructs influence student performance when using ChatGPT for academic work. Grounding the inquiry in established literature while acknowledging the evolving realities of AI-driven education, this study provides a holistic foundation for understanding the multiple factors that shape learning outcomes in digital academic settings.

Generative AI has rapidly reshaped the educational landscape, yet much of the existing research presents a mixed picture of its potential and its limitations. Giannakos et al. (2025) highlights generative AI’s growing ability to personalize learning and expand creative possibilities, but he also emphasizes that its success depends heavily on teacher readiness and institutional support. This reflects a broader gap between technological possibility and practical implementation. Similarly, Wang et al. (2025) demonstrate the creative value of AI-driven art education through GANs and neural style transfer, while also raising concerns surrounding authorship, bias, and fairness issues that become equally significant when AI is used for academic assignments. From a broader perspective, Yusuf et al. (2024) systematic review maps emerging trends in educational AI but finds considerable fragmentation across institutions, particularly regarding ethical guidelines and integration strategies. Chugh et al. (2025) reinforces this tension by showing that although AI tools enhance creativity and personalization, they simultaneously introduce new concerns surrounding data privacy and ethical responsibility. Adding to this, Wang (2025) acknowledges the instructional benefits of AI but warns that without clear pedagogical frameworks, educators may struggle to align AI-generated assistance with learning outcomes. Collectively, these theoretical perspectives suggest that the educational value of generative AI remains contingent on human oversight, ethical clarity, and digital competence core elements that shape the foundations of this study.

Research focusing specifically on ChatGPT reflects a similar blend of opportunity and caution. Pradana et al. (2023) notes ChatGPT’s rapid emergence as an academic support tool but observes that its long-term impact on learning quality remains underexplored. Empirical studies emphasize that its benefits depend largely on how students engage with it. Levine et al. (2025) shows that ChatGPT strengthens students’ writing processes, though meaningful improvement occurs only when learners use feedback critically rather than duplicating outputs. Guarin et al. (2025) raise concerns about academic integrity, finding that while ChatGPT supports conceptual understanding, it may unintentionally encourage plagiarism when students are unaware of proper citation practices. Similarly, Thelwall (2025) demonstrates ChatGPT’s usefulness in evaluating complex academic texts but notes that overreliance may weaken students’ analytical reasoning. Abdullah et al. (2025) confirms students’ appreciation of ChatGPT as a writing assistant, yet highlights variability in their perceptions of its usefulness, suggesting that factors such as digital literacy and personal learning strategies significantly shape outcomes. Across these studies, a clear pattern emerges: the effectiveness of ChatGPT depends on students’ digital literacy, ethical awareness, and their ability to actively engage with AI-generated feedback all of which constitute the central variables of this study.

A recurring theme in recent literature is that AI adoption is moving faster than the ethical preparedness required to support it. Gouseti et al. (2025) identifies persistent concerns such as data privacy, algorithmic bias, and unequal access, all of which influence how fairly AI-supported learning environments operate. Barnes and Tour (2025) points out that educators often remain apprehensive due to unclear institutional policies around transparency and academic integrity. From the student side, Higgs and Stornaiuolo (2024) finds that many learners struggle with understanding authorship, originality, and proper use of AI-generated content, reinforcing the need for ethical literacy as part of AI-based learning. In K–12 contexts, Nadelson (2025) warn that premature or unregulated AI adoption may reinforce existing biases or increase student overdependence. Expanding on these concerns, Yan et al. (2024) notes that large language models can unintentionally spread misinformation, underscoring the necessity of strong ethical frameworks and critical digital engagement. Taken together, these studies show that effective AI integration requires more than access it demands digital competence, ethical awareness, and guided instructional support, all of which remain emerging areas within educational systems.

Although prior research acknowledges ChatGPT’s educational value, several important gaps still remain. Students consistently highlight how the tool enhances their writing and deepens their conceptual understanding, as noted by Cornish and Larter (2024), Moheno et al. (2024), and Al-Sofi (2024). At the same time, concerns continue to surface regarding the reliability of its responses, the risk of over-reliance, and the possibility of academic misconduct. Faculty perspectives add further nuance. Al Muhanna (2025) observes that many educators hold a generally positive outlook toward ChatGPT, yet they emphasize the need for targeted training to support meaningful adoption. Similarly, Damaševičius (2024) underscores that strong human oversight is essential to maintain fairness and uphold assessment integrity. In contrast, research by Leelavathi and Surendhranatha (2024) and Wang (2025) demonstrates that when used thoughtfully and responsibly, ChatGPT can foster higher-order thinking, creativity, and improved communication among learners. Despite these insights, current research seldom examines how digital literacy, ethical awareness, and perceived usefulness collectively shape student performance. This gap becomes even more evident when considering nontraditional learners or when exploring emotional and motivational dimensions of learning, as highlighted by Yang et al. (2025) and Hickman and Stoica (2025). These gaps provide the basis for the present study and guide its focus.

The conceptual model of this study draws on established educational and behavioral theories that explain how students adopt, evaluate, and effectively use digital tools such as ChatGPT to enhance academic performance. The three core constructs Digital Literacy, Ethical Awareness, and Perceived Usefulness are grounded in theoretical frameworks that describe technology adoption, responsible digital behavior, and learning performance in modern educational environments.

Digital literacy is foundational to effective learning in technology-rich settings. The Digital Literacy Framework by Eshet-Alkalai (2004) emphasizes that students require a blend of cognitive, socio-emotional, and technical abilities to navigate digital content meaningfully. These competencies shape how learners evaluate, interpret, and utilize AI tools for academic tasks. Complementing this, the Technology Skills and Competency Model proposed by Ng (2012) argues that higher levels of digital literacy support self-directed learning and strengthen academic performance by enabling learners to use digital tools more efficiently. Recent empirical studies (e.g. Holm, 2025; Budiman, 2023) validate these theoretical assumptions, showing that digitally literate students tend to achieve superior academic outcomes.

Ethical awareness is similarly rooted in established behavioral frameworks. The Moral Development Theory of Kohlberg (1981) suggests that individuals with higher ethical reasoning make more responsible choices an essential requirement in academic contexts where integrity and originality matter. When applied to digital environments, ethical awareness shapes how students use AI tools responsibly, helping them avoid plagiarism, misinformation, and academic misconduct. The Digital Ethics Framework by Floridi (2013) further stresses that ethical maturity is crucial for navigating algorithmic systems. These perspectives align with empirical findings by Prashar (2024), Rua (2024), and Waqas (2025), which show that students with heightened ethical awareness engage more critically and responsibly with AI tools like ChatGPT, resulting in improved learning quality and academic performance.

Perceived usefulness is grounded in the well-established Technology Acceptance Model (TAM) developed by Davis (1989), which posits that individuals are more likely to adopt a technology when they believe it enhances their performance. In academic contexts, this means students are more inclined to engage with ChatGPT when they view it as beneficial for writing, learning, and problem-solving. The Unified Theory of Acceptance and Use of Technology (Venkatesh et al., 2003) complements this view by identifying performance expectancy aligned with perceived usefulness as a strong predictor of technology adoption. Empirical research (Hussain and Anwar, 2025; Sustaningrum, 2025; Kim and Moon, 2025) reinforces these claims, showing that students who perceive ChatGPT as academically valuable demonstrate higher engagement and better academic outcomes.

Furthermore, the integration of ChatGPT in education can also be explained by broader motivational frameworks. Self-Determination Theory (Deci and Ryan, 2000) emphasizes autonomy, competence, and relatedness as core drivers of motivation. ChatGPT supports these dimensions by offering personalized guidance, interactive feedback, and a sense of academic support, thereby strengthening student engagement and skill development (as shown in Levine et al., 2025; Zhan and Yan, 2025). These theories complement TAM by explaining not only how students adopt ChatGPT but also how their internal motivation enhances learning quality. Taken together, these frameworks TAM, UTAUT, Digital Literacy Theory, Moral Development Theory, Digital Ethics, and SDT offer a strong theoretical foundation for understanding how Digital Literacy, Ethical Awareness, and Perceived Usefulness collectively shape Student Performance in AI-enhanced learning environments.

Digital literacy is increasingly recognized as a pivotal factor in enhancing student performance across diverse educational settings. Holm (2025) demonstrates that students with higher digital literacy and self-directed learning skills exhibit greater engagement, which translates into improved academic achievement, particularly in online courses. Similarly, Budiman (2023) highlights that first-year students possessing advanced digital literacy attain better academic outcomes, emphasizing the critical role of digital competencies in navigating academic environments and effectively utilizing digital resources. Zheng and Yan (2025) further reveals that English as a Foreign Language (EFL) students with stronger digital literacy display enhanced online learning capabilities, enabling more successful engagement in virtual learning contexts. In secondary education, Fayda-Kinik (2025) confirms through a meta-analysis a significant positive correlation between Information and Communication Technology (ICT) use and academic performance, underscoring the necessity of integrating digital tools into pedagogy. Pan (2024) notes that digital competence shapes students’ learning behaviors, fostering autonomous and efficient learning strategies that improve academic outcomes. Extending this perspective, Zhan and Yan (2025) finds that students with higher digital literacy effectively engage with ChatGPT feedback, enhancing writing performance. Bender (2024) emphasizes that proficiency with generative AI tools like ChatGPT further strengthens academic outcomes, and Yang et al. (2025) illustrates that digitally literate high school student’s leverage ChatGPT to improve learning in technical subjects.

Ethical awareness is a fundamental determinant of student performance across educational contexts. Prashar (2024) demonstrates that plagiarism awareness programs enhance students’ ethical sensitivity and critical-thinking skills, enabling sound ethical decision-making and promoting academic integrity. Similarly, Rua (2024) develops the Ethical Student Scale to measure students’ ethical awareness and behavior, revealing that higher ethical awareness correlates with improved academic outcomes. Cheng and Lee (2024) highlights that combining ethics instruction with internship experiences strengthens students’ ethical perceptions, fostering professional conduct and better performance. Kotluk and Tormey (2024) emphasizes that incorporating varying levels of compassion within ethics education further cultivates ethical awareness, positively impacting learning outcomes and personal growth. Extending this perspective to digital environments, Waqas (2025) proposes the “Mediated Digital Integrity Model,” demonstrating how institutional support, students’ attitudes toward AI, and moral disengagement interact to influence AI-facilitated academic misconduct. Collectively, these studies underscore the critical role of ethical awareness in enhancing both traditional and digital academic performance, informing the design of integrated ethics programs in higher education.

Perceived usefulness of ChatGPT has a significant positive impact on student performance in academic settings. Hussain and Anwar (2025) find that students who perceive ChatGPT as a valuable tool are more likely to adopt it for learning, resulting in greater engagement and improved academic outcomes. Durgungoz and Kharrufa (2025) emphasizes that students viewing ChatGPT as a study companion or instructional aid achieve better learning results, demonstrating its supportive role in academic tasks. Sustaningrum (2025) highlights that students who perceive ChatGPT as useful for enhancing interactivity and access to information experience deeper learning and higher performance levels. Similarly, Heine and König (2025) shows that perceived usefulness shapes students’ attitudes and intentions toward AI tools in teacher education, fostering enhanced academic achievement. Kim and Moon (2025) further identifies key determinants of ChatGPT’s perceived usefulness, revealing that students who recognize its educational value perform better academically. Collectively, these studies underscore the critical role of perceived usefulness in maximizing the educational benefits of ChatGPT.

Figure 1 illustrates the conceptual framework depicting the hypothesized relationships between Digital Literacy (H1), Ethical Awareness (H2), and Perceived Usefulness (H3) and Students’ Performance. The framework proposes that higher levels of digital literacy, stronger ethical awareness, and greater perceived usefulness of ChatGPT positively influence academic outcomes, thereby providing a theoretical basis for examining the direct effects of these factors on students’ learning achievements.

This study investigated the utilization of ChatGPT among undergraduate and postgraduate students in Karnataka, emphasizing its application in academic assignments, cost-benefit considerations, and its broader role in digital learning. The total population consisted of 1,944,670 students from the two participating universities. The final sample size of 304 students, representing an acceptable response rate in behavioral and social science studies where populations are very large. The sample size surpasses the minimum that Krejcie and Morgan’s (1970) sample size table recommended for large populations, which assumes sufficient representativeness with samples above 300. Methodological literature reports that samples over 200 are typically appropriate for multivariate analyses and structural modelling (Hair et al., 2019a, b). The final sample of 304 thus provides a sufficient level of precision and statistical power to support the objectives of this study. Simple random sampling has been adopted and data collection occurred between October 2024 and January 2025 using a structured questionnaire designed to capture students’ engagement with ChatGPT. The present research applied a self-administered questionnaire, which the respondents filled out independently without any interference from the researchers. In light of the possibility of bias, participants were guaranteed to remain fully anonymous, and no name or other personal information from them was collected. Instructions were clear to minimize misunderstanding and ensure responses were uniform. Moreover, informing respondents that none of the questions were right or wrong lowered social desirability tendencies among respondents. All these cumulative measures enhanced the accuracy and credibility of the data obtained. To enhance construct clarity, the authors indicate that the three variables adopted for this study, namely Digital Literacy, Ethical Awareness, and Perceived Usefulness, were all adapted from established constructs in prior literature. While these variables have roots in past empirical studies, an EFA was carried out to confirm their suitability and factor structure within the present context and population. The authors reviewed literature, a priori, in order to identify validated dimensions and item pools for the respective constructs before actual administration of the instrument. The selected items were further screened by a panel of three domain experts to ensure content validity and contextual relevance. A detailed listing of all the measurement items (see Appendix A) for each of the variables is then provided in the manuscript to allow transparency and support the robustness of the instrument design.

The instrument comprised two sections: the first gathered demographic details such as age, gender, academic level, and field of study, while the second examined ChatGPT usage for assignments, perceived cost advantages, and its influence on learning experiences. Descriptive statistics were applied to summarize participant demographics, followed by Chi-square tests to explore differences in usage patterns across groups. Structural Equation Modeling (SEM) was employed to analyze relationships between variables, offering insights into how assignment usage, cost-efficiency, and digital learning collectively shape students’ academic engagement with ChatGPT.

The data analysis followed a multi-stage approach to examine the impact of ChatGPT on student engagement. Initially, descriptive statistics were applied to summarize the demographic characteristics of the 304 participants, providing an overview of gender, age, academic level, and field of study. Subsequently, Chi-square tests were conducted to identify significant differences in ChatGPT usage across demographic groups, revealing variations in engagement patterns between undergraduates and postgraduates as well as across different academic disciplines. The primary analysis employed Structural Equation Modeling (SEM) to explore the relationships among key variables: ChatGPT usage for assignments, cost-benefit perceptions, and its broader application in digital learning. SEM allowed for the assessment of both direct and indirect effects of these factors on overall student engagement. This modeling approach offered a holistic understanding of how ChatGPT contributes to academic activities, highlighting the variables most strongly influencing engagement and providing insights into its role in supporting students’ academic performance and digital learning experiences.

The test results in Table 1 show that the rotated component matrix demonstrates a clear factor structure corresponding to the proposed constructs: Digital Literacy, Perceived Usefulness, Students’ Performance, and Ethical Awareness. Six Digital Literacy items (DL1–DL6) loaded strongly on the first component, with factor loadings ranging from 0.647 to 0.831, indicating high internal consistency and confirming the construct’s validity. The five Students’ Performance measures (SP1–SP5) loaded between 0.732 and 0.845 on the second component, reflecting a distinct and well-represented construct. Perceived Usefulness items (PU1–PU5) exhibited strong loadings from 0.686 to 0.811 on the third component, highlighting students’ perceptions of the utility of digital tools. Ethical Awareness items (EA1–EA6), with EA5 excluded due to low loading, loaded on the fourth component, ranging from 0.683 to 0.800, forming a coherent independent dimension. Principal Component Analysis with Varimax rotation facilitated clear factor differentiation, and convergence within six iterations confirms the stability of the solution. Overall, the results presented in Table 1 validate the measurement model and demonstrate satisfactory discriminant validity across all four constructs.

The test results in Table 2 indicate that the measurement model demonstrates satisfactory reliability and validity across all four constructs Digital Literacy (DL), Ethical Awareness (EA), Perceived Usefulness (PU), and Students’ Performance (SP). All constructs meet the recommended thresholds for internal consistency, with Cronbach’s alpha and composite reliability values exceeding 0.70 (Hair et al., 2019a, b). Convergent validity is also established, as the average variance extracted (AVE) for each construct is above the minimum criterion of 0.50, indicating that the indicators adequately capture the underlying constructs. Although a few item loadings (such as DL4 and EA4) are marginally below the preferred cutoff of 0.70, their inclusion is justified given the strong overall reliability and AVE values (Fornell and Larcker, 1981). In addition, variance inflation factor (VIF) values remain below 3, suggesting that multicollinearity does not pose a concern. Consistent with Kock and Lynn (2012) and Hair et al. (2019a, b), these VIF levels confirm stable estimates and distinct contributions of the independent variables to the model.

The test results in Table 3 show that the discriminant validity of the constructs was assessed using the Heterotrait–Monotrait (HTMT) ratio of correlations, which is considered a more robust criterion than traditional approaches such as the Fornell–Larcker test (Henseler et al., 2015). The HTMT values ranged from 0.510 to 0.691, indicating clear differentiation among the constructs. Specifically, the HTMT ratios were 0.689 between Digital Literacy (DL) and Ethical Awareness (EA), 0.533 between DL and Perceived Usefulness (PU), 0.557 between DL and Students’ Performance (SP), 0.691 between EA and PU, 0.545 between EA and SP, and 0.510 between PU and SP. According to established guidelines, HTMT values below 0.85 (Kline, 2011), or more conservatively below 0.90 (Henseler et al., 2015), indicate acceptable discriminant validity. As all HTMT values in this study fall well below the 0.85 threshold, the constructs are empirically distinct and free from multicollinearity and conceptual overlap. Overall, the results presented in Table 3 confirm that Digital Literacy, Ethical Awareness, Perceived Usefulness, and Students’ Performance represent four distinct constructs, thereby demonstrating strong discriminant validity within the measurement model.

Figure 2 presents the integrated measurement and structural model depicting the relationships among Digital Literacy (DL), Perceived Usefulness (PU), Ethical Awareness (EA), and Students’ Performance (SP). The measurement model indicates that the majority of indicator loadings surpass the recommended threshold, demonstrating that the observed variables reliably capture their respective latent constructs. Digital Literacy is adequately represented by items DL1–DL6, Perceived Usefulness by PU1–PU5, Ethical Awareness by EA1–EA6, and Students’ Performance by SP1–SP5, confirming satisfactory indicator reliability and construct representation. From the structural perspective, Digital Literacy shows a strong and positive influence on Perceived Usefulness, suggesting that higher levels of digital competence enhance students’ perceptions of ChatGPT’s utility in learning. Additionally, Digital Literacy exerts a substantial direct effect on Students’ Performance, underscoring its pivotal role in academic success. Perceived Usefulness also positively influences Students’ Performance, indicating that recognizing the value of ChatGPT contributes to improved learning outcomes. Conversely, the direct relationship between Ethical Awareness and Students’ Performance is relatively weak, implying that ethical considerations may not directly impact measurable academic performance in this context. Overall, the model demonstrates robustness and highlights Digital Literacy and Perceived Usefulness as key predictors of Students’ Performance.

The test results in Table 4 indicate that Confirmatory Factor Analysis (CFA) was conducted to validate the measurement model comprising four latent constructs: Digital Literacy (DL), Perceived Usefulness (PU), Ethical Awareness (EA), and Students’ Performance (SP). All standardized factor loadings exceeded the recommended threshold of 0.60, confirming that each indicator adequately represents its respective construct (Hair et al., 2019a, b). The model fit indices demonstrate an acceptable overall fit. The Chi-square/df ratio was 2.205, which falls within the ≤3 range suggested by Kline (2011). Incremental fit indices, including CFI (0.946) and TLI (0.924), exceeded the recommended cut-off value of 0.90, while SRMR (0.079) was within the acceptable ≤0.08 threshold, and AGFI (0.939) indicated strong model parsimony. Although the RMSEA value (0.103) was marginally above the recommended 0.08 threshold, the majority of the fit indices support the adequacy of the measurement model. Overall, the results presented in Table 4 confirm that the four-factor model demonstrates satisfactory construct validity and reliability, supporting its suitability for subsequent structural analysis, with minor model refinements potentially improving RMSEA.

Figure 3 presents the final structural equation model, illustrating how Digital Literacy (DL), Perceived Usefulness (PU), and Ethical Awareness (EA) directly relate to Students’ Performance (SP), along with the measurement quality of each construct. The model explains a meaningful proportion of variance in Students’ Performance (R² = 0.332), indicating that about one-third of students’ academic performance is jointly accounted for by these three factors. The measurement model demonstrates satisfactory indicator loadings, with most items exceeding recommended thresholds, suggesting that the constructs are reliably measured. Digital Literacy is well captured by items DL1–DL6, while Perceived Usefulness (PU1–PU5) and Students’ Performance (SP1–SP5) also show strong representation. Although a few Ethical Awareness items exhibit comparatively lower loadings, the construct remains adequately represented overall. Structurally, Digital Literacy shows a strong and positive effect on Students’ Performance, highlighting the importance of digital competence in achieving better academic outcomes. Perceived Usefulness also positively influences performance, indicating that students who recognize the value of tools like ChatGPT tend to perform better. In contrast, Ethical Awareness demonstrates a negligible direct effect, suggesting that ethical considerations alone may not immediately translate into measurable performance gains. Overall, the findings emphasize the central role of digital skills and perceived usefulness in enhancing student performance.

The test results in Table 5 illustrate the effects of Digital Literacy, Ethical Awareness, and Perceived Usefulness on Students’ Performance. Digital Literacy exhibited a positive and statistically significant influence on Students’ Performance (β = 0.393, p < 0.001), indicating that higher levels of digital competence enhance students’ academic outcomes, thereby supporting H1. This finding aligns with prior research highlighting digital skills as critical predictors of effective engagement and performance in technology-enhanced learning environments (Ng, 2012; Siddiq et al., 2016). Perceived Usefulness also demonstrated a positive and significant effect on Students’ Performance (β = 0.259, p < 0.001), suggesting that students’ recognition of the utility of digital tools contributes to improved learning outcomes, thus supporting H3. This result is consistent with the Technology Acceptance Model (TAM), which posits perceived usefulness as a key determinant of performance outcomes (Davis, 1989; Teo, 2011), underscoring the importance of favorable student attitudes toward digital technologies. In contrast, Ethical Awareness did not show a significant effect on Students’ Performance (β = −0.006, p = 0.924), indicating that while ethical values are important, they may not directly translate into measurable academic performance, thereby not supporting H2 (Arifin, 2018). Overall, the findings presented in Table 5 support H1 and H3, but not H2, demonstrating that Digital Literacy and Perceived Usefulness significantly enhance Students’ Performance, whereas Ethical Awareness does not exert a direct impact within the structural model.

The findings of this study offer a clearer understanding of how ChatGPT-assisted assignments shape student performance through the combined influence of digital literacy, perceived usefulness, and ethical awareness. The strong and significant effect of digital literacy on student performance reflects a pattern consistently reported in earlier work. Students with higher digital competence are better positioned to navigate, evaluate, and apply AI-generated information, which directly strengthens the quality of their academic work. This outcome aligns with the foundational arguments by Ng (2012) and the empirical evidence provided by Siddiq et al. (2016), both of which emphasize that digital skills are essential for effective participation in technology-enhanced learning environments. In this way, the present study extends existing scholarship by showing that digital literacy remains a decisive factor even when learning is mediated through advanced generative AI tools. The results also demonstrate the significant role of perceived usefulness, echoing the core propositions of the Technology Acceptance Model (Davis, 1989) and later extensions such as Teo (2011). Students who perceive ChatGPT as helpful whether through improved clarity, structured outputs, or timely feedback are more likely to use it strategically, leading to enhanced academic performance. Recent studies such as Hussain and Anwar (2025) and Durgungoz and Kharrufa (2025) similarly found that perceived usefulness shapes both attitudes toward AI tools and the quality of learning outcomes. The present study reinforces these earlier insights while contributing new evidence on how perceived usefulness operates specifically in the context of AI-assisted academic assignments.

In contrast, ethical awareness does not show a significant direct effect on performance. While ethical values remain an essential part of responsible digital behavior, the findings suggest that these values may not immediately translate to measurable academic gains. This is consistent with broader discussions, such as Arifin (2018), which note that ethical sensitivity often influences long-term decision-making, professional conduct, or responsible technology use rather than short-term academic outcomes. This observation highlights a subtle but important distinction: ethical awareness may not boost grades, but it contributes to integrity, trustworthiness, and responsible engagement elements that matter deeply for sustainable learning and professional development. Beyond the psychological variables, the study also reveals a practical issue often overlooked in AI research: cost. While educational institutions may offer limited access to digital tools, many students rely on personal subscriptions or premium features to maximize their use of AI platforms such as ChatGPT. This creates inequities in how students experience technology-enhanced learning. Recognizing the interplay between functional benefits and affordability helps broaden the discussion on AI integration, reminding educators and policymakers that accessibility must accompany innovation.

The measurement model further validates the conceptual structure of the study. The rotated component matrix clearly distinguishes the four constructs Digital Literacy, Students’ Performance, Perceived Usefulness, and Ethical Awareness with strong and stable factor loadings. Reliability results confirm that all constructs meet the recommended thresholds for internal consistency (Hair et al., 2019a, b), while VIF values below 3 (Kock and Lynn, 2012) rule out multicollinearity concerns. The HTMT results (Henseler et al., 2015) show clear discriminant validity, confirming that each construct captures a unique dimension of AI-mediated learning. CFA results further support model adequacy, with most indicators reflecting good fit (Kline, 2011), strengthening confidence in the structural relationships examined. Taken together, the structural model outcomes offer meaningful contributions to theory. The positive effects of digital literacy and perceived usefulness reinforce long-standing technology acceptance theories while demonstrating their relevance in the context of generative AI. At the same time, the non-significant effect of ethical awareness offers fresh insight, suggesting that ethical considerations may operate more as contextual or moderating factors rather than direct drivers of performance. This nuance adds depth to evolving conversations about responsible AI usage in education.

The integration of ChatGPT into academic work brings together long-standing discussions on digital literacy, technology acceptance, and ethical responsibility in higher education. Established research has consistently shown that digital competence forms the backbone of meaningful engagement with technological tools. Scholars such as Ng (2012) and Siddiq et al. (2016) emphasized that students who can skillfully evaluate, interpret, and manage digital information are better equipped to navigate modern learning environments. This study strengthens that understanding by illustrating how digital literacy continues to play a decisive role even in the era of generative AI. Technology acceptance theory also provides a strong foundation for explaining student behavior. According to Davis (1989) and Teo (2011), students are more likely to adopt and benefit from a tool when they perceive it to be genuinely useful. The present results affirm this principle within the context of ChatGPT, showing that when students find clear value such as better clarity, faster feedback, or organized responses they engage more actively and produce stronger academic work. Ethical awareness, although essential for responsible technology use, aligns with previous observations by Arifin (2018) in that it shapes attitudes and long-term behavior rather than immediate academic performance. Together, these theoretical connections demonstrate that established frameworks remain highly relevant and continue to explain how students interact with emerging AI technologies.

On the practical side, the study highlights several concrete steps that institutions can implement to support responsible and effective AI use. One immediate intervention is the introduction of AI-verification evidence for assignments created with the help of ChatGPT. Students can be required to include screenshots or reports from tools such as ZeroGPT or GPTZero to document how AI was used. This promotes transparency and reinforces ethical standards without discouraging legitimate academic support. Digital literacy development should also go beyond general training sessions. Embedding hands-on tasks such as evaluating AI-generated content, verifying accuracy, or revising outputs can help students cultivate deeper critical engagement. Demonstrating the practical usefulness of ChatGPT through guided activities in writing, outlining, and idea refinement can further encourage purposeful utilization. Ethics education, meanwhile, benefits from being situational rather than theoretical. Mini-projects or case-based tasks on bias, misinformation, or misuse can help students internalize responsible AI behavior. A further consideration is affordability: while institutions may offer basic access, advanced or premium features often come with additional costs. Ensuring that students have equitable access to essential AI tools is crucial for preventing disparities in performance.

Overall, strengthening digital skills, demonstrating practical value, integrating verification practices, and embedding ethical guidance can create a balanced environment where students use ChatGPT confidently and responsibly. These applied measures translate theoretical insights into actionable strategies that support both academic performance and long-term digital maturity.

This study offers valuable insights into how students use ChatGPT for academic work, but several limitations should be acknowledged. Since the data were self-reported, there is always a risk that students may have over- or under-stated their digital skills or ethical awareness. To reduce this concern, respondent anonymity was ensured so participants could answer honestly without pressure. After data cleaning, multicollinearity tests, including VIF checks, were also conducted to ensure that the variables were distinct and that the findings were not influenced by overlapping constructs. Another limitation is the relatively short data-collection period, which makes it difficult to capture long-term learning or behavioral changes associated with AI use. Although this could not be fully avoided, consistency checks were applied to maintain reliability within the available timeframe. The focus on a single student group and the exclusive use of ChatGPT also limits broader generalization across disciplines, institutions, and different AI tools. Additionally, ethical awareness was treated as one broad construct, which may have overshadowed more specific ethical dimensions such as academic honesty or data privacy. Looking ahead, future research can build a more comprehensive model by including mediating variables like motivation, self-regulation, or cognitive engagement to better explain how students’ skills and attitudes translate into performance. Moderating factors such as discipline, prior AI exposure, socioeconomic background, or institutional support could reveal when these relationships become stronger or weaker. Longitudinal designs, diverse student samples, multiple AI platforms, and a more detailed breakdown of ethical awareness would provide a richer and more holistic understanding of AI-supported learning.

All participants provided informed consent prior to their voluntary participation in the study, and their responses were kept anonymous and confidential in accordance with ethical research guidelines.

The supplementary material for this article can be found online.