Student Perceptions of AI-Assisted Writing and Academic Integrity: Ethical Concerns, Academic Misconduct, and Use of Generative AI in Higher Education

Academic Editor: Savvas A. Chatzichristofis Received: 16 July 2025 Revised: 21 August 2025 Accepted: 28 August 2025 Published: 2 September 2025 Citation: Lund, B., Mannuru, N. R., Teel, Z. A., Lee, T. H., Ortega, N. J., Simmons, S., & Ward, E. (2026). Student Perceptions of AI-Assisted Writing and Academic Integrity: Ethical Concerns, Academic Misconduct, and Use of Generative AI in Higher Education. AI in Education, 1(1), 2. https://doi.org/10.3390/ aieduc1010002 Copyright: © 2025 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/ licenses/by/4.0/). Article Student Perceptions of AI-Assisted Writing and Academic Integrity: Ethical Concerns, Academic Misconduct, and Use of Generative AI in Higher Education Brady Lund * , Nishith Reddy Mannuru , Zoë Abbie Teel, Tae Hee Lee, Nathanlie Jugan Ortega, Sara Simmons and Evelyn Ward College of Information, University of North Texas, Denton, TX 76201, USA; [email protected] (Z.A.T.) *Correspondence: brady[email protected] Abstract The rise of generative AI in higher education has disrupted our traditional understandings of academic integrity, moving our focus from clear-cut infractions to evolving ethical judgment. In this study, a survey of 401 students from major U.S. universities provides insight into how beliefs, behaviors, and policy awareness intersect in shaping how students interact with AI-assisted writing. The findings indicate that students’ ethical beliefs—not institutional policies—are the strongest predictors of perceived misconduct and actual AI use in writing. Policy awareness was found to have no significant effect on ethical judgments or behavior. Instead, students who believe AI writing is cheating were found to be substantially less likely to view it as ethical or engage with it. These findings suggest that many students do not treat AI use in learning activities as an extension of conventional cheating (e.g., plagiarism), but rather as a distinct category of academic conduct/misconduct. Rather than using punitive models to attempt to punish students for using AI, this study suggests that education about AI ethics and the risk of AI overreliance may prove more successful for curbing unethical AI use in higher education. Keywords: academic integrity; AI-assisted writing; plagiarism; AI ethics; higher education; generative artificial intelligence 1. Introduction For students in higher education, the boundaries of academic misconduct have traditionally seemed relatively fixed—cheating, plagiarism, and collusion were clearly defined. But the rise of generative artificial intelligence, particularly large language models (LLMs), has begun to unsettle the map of academic integrity. Unlike plagiarism, which implies a straightforward act of copying without attribution, LLMs produce text that is technically original but procedurally opaque, shaped by probabilistic associations rather than authorship (White et al.,2023). As a result, the outputs of these systems sit uneasily between generative tool (e.g., ChatGPT) and coauthor, innovation and infraction—an ambiguous zone where institutional norms have yet to catch up. Research by Lund et al. (2025) suggests that students are navigating this ambiguity without a clear ethical compass, often uncertain where acceptable use of these tools ends and misconduct begins. This uncertainty is not simply a matter of policy lagging behind technology. It likely reflects a deeper epistemological shift among students in what it means to write, to know, and to learn. For students, LLMs offer powerful new affordances: instant feedback, tailored content, frictionless summaries of complex material (Williamson & Murray,2024). AI Educ. 2026,1, 2 https://doi.org/10.3390/aieduc1010002 AI Educ. 2026,1, 2 2 of 13 They expand access to knowledge and scaffold the learning process in ways traditional instruction may not. But with these benefits come ethical frictions—over-reliance, epistemic passivity, and blurred authorship (Zhai et al.,2024). More than just technical tools, LLMs are shaping how students conceive of academic labor, creativity, and intellectual ownership (Darban,2025). While institutions scramble to update their honor codes and draft AI integrity policies, students are already making choices in real time, guided as much by perceived fairness and personal values as by formal rules. Understanding these choices—how students interpret the ethics of AI use, what they believe constitutes “help” versus “cheating,” and how these beliefs vary across contexts— is essential for designing pedagogies and policies that are not only effective, but just. These questions do not have easy answers. They require grappling with the tension between technological possibility and academic responsibility, between efficiency and effort, between the right to use tools and the obligation to learn (Yue et al.,2025). In this study, we examine how students position themselves within this evolving landscape, and what their perspectives reveal about the future of academic integrity in an AI-powered world. Our findings highlight the importance of ethical beliefs rather than institutional policies in ensuring appropriate student conduct with AI tools, suggesting the universities invest in a course of ethics instruction to combat AI-powered misconduct. 2. Literature Review 2.1. Generative AI in Higher Education Across university campuses, Generative Artificial Intelligence (GenAI) is no longer an emerging novelty. It is here, embedded in the margins of syllabi, lingering in student browsers, shaping drafts before they are even written. Whether higher education institutions choose to formally embrace it or not, the tide of AI adoption is already rising. As Bearman et al. (2022) suggest, GenAI’s presence is becoming less a matter of policy choice and more an inevitability that demands thoughtful navigation. McDonald et al. (2025) describe this moment not as an inflection point but as an integration point, one where strategy, ethics, pedagogy, and imagination must converge. The literature paints an academic landscape that is marked by both promise and peril. On one side, scholars consistently note GenAI’s usefulness as a flexible, scalable, and often responsive assistant for teaching, learning, and administration. Chiu et al. (2023), Compton and Burke (2023), and Kurtz et al. (2024) all highlight its growing role in tutoring, feedback provision, and even lesson planning. These tools are increasingly integrated into how instructors assess student understanding and where administrators forecast enrollment patterns, refine communications, and streamline operational planning (Zawacki-Richter et al.,2019). For students—particularly at the undergraduate level—GenAI has been described as a kind of always-on tutor: one that explains, rewrites, rephrases, and even reassures (Shahzad et al.,2025). Yet, as Yusuf et al. (2024) observe, the growing body of research pays comparatively less attention to graduate students, whose needs and challenges may differ substantially. Despite the enthusiasm around its functionality, the literature repeatedly sounds a cautionary note. Ethical dilemmas prompted by AI are neither small nor theoretical. Scholars such as Batista et al. (2024) and Perera and Lankathilaka (2023) point to the blurring line between help and dishonesty, with GenAI enabling new forms of academic misconduct that are difficult to detect. The generation of fabricated citations, a known flaw in some GenAI models, adds another layer of complexity, undermining academic integrity and muddying the waters of authorship and originality (H. Wang et al.,2024). Kurtz et al. (2024) echo these concerns, noting that cheating facilitated by GenAI poses a serious threat to educational integrity. AI Educ. 2026,1, 2 3 of 13 Beyond ethical concerns lies a deeper pedagogical question: what happens to student learning when artificial machines become too helpful? Cordero et al. (2025), D. Lee et al. (2024), and Zhang and Xu (2025) explore the risk of intellectual atrophy, warning that students may become overly dependent on GenAI tools, sacrificing opportunities to develop creativity, reasoning, and problem-solving skills. Batista et al. (2024) reinforce this perspective, while also suggesting that without guided instruction, GenAI may ultimately short-circuit meaningful engagement with course content. And yet, across much of this critical discourse, a student’s own perception of GenAI—their understanding, curiosity, trust, or skepticism—is rarely addressed as a factor shaping its misuse. There is consensus, however, that GenAI’s role in higher education is no longer optional or something that can be prevented. As the literature repeatedly emphasizes, the path forward is not resistance, but regulation—anchored by institutional policies, ethical guidelines, and shared expectations (Batista et al.,2024;Cordero et al.,2025;Kurtz et al., 2024;D. Lee et al.,2024;Perera & Lankathilaka,2023;H. Wang et al.,2024;Zhang & Xu,2025). Still, any effective policy must reflect the diversity of higher education institutions and their learners. Yusuf et al. (2024) caution that cultural variability across global institutions makes universal policy frameworks unrealistic. Instead, they argue for context-specific strategies that are sensitive to distinct educational values and ethical frameworks. Central to this adaptation is GenAI literacy—not only for students but for faculty and staff as well. Farrelly and Baker (2023) emphasize that GenAI cannot be responsibly integrated without equipping educators with the tools to navigate it critically and creatively. They, along with McGrath et al. (2023), highlight frameworks like those developed by Ng et al. and Hillier, which provide structured approaches to building GenAI literacy in academic environments. Zhang and Xu (2025) similarly underline that institutions must invest in upskilling their workforce, not merely to keep pace with technology but to shape its use in pedagogically sound and ethically aligned ways. 2.2. Academic Integrity in the Age of AI Within higher education, academic integrity is a principle that is often considered a core tenet of an institution’s values. An academic institution must be able to uphold academic integrity within its programs; otherwise, the credibility and quality of the programs offered by the institution become questionable (Balalle & Pannilage,2025). The International Center for Academic Integrity [ICAI] (2021) currently describes academic integrity as being the commitment to the values of honesty, trust, fairness, respect, responsibility, and courage within academic communities. Thus, dishonest practices which threaten academic integrity, such as plagiarism and cheating, are largely frowned upon within higher education, especially in higher degree levels (S ,ercan & Voicu,2022). Several studies have examined the ethical behavior of higher education students. While some of these studies point to specific attributes like gender and peer networks in influencing students’ ethical behavior (Joseph & Berry,2010), many more studies point to the importance of instruction and discussion of ethical principles as critical in developing ethical behavior (Armstrong et al.,2003;Koerber et al.,2005). Notably, Susilowati et al. (2021) found that courses on ethical behavior considerably enhanced ethical behavior, whereas a general ethical or policy climate did not influence behavior in meaningful ways. While threats to the core values of academic integrity are far from being a novel sighting, the rise of generative AI writing tools, such as ChatGPT, have brought reason for concern within academic institutions (Benke & Sz˝oke,2024;Laflamme & Bruneault,2025; Sabzalieva & Valentini,2023). With LLMs being able to easily provide detailed writing, they have become a prime tool for cheating in academic settings (Ward et al.,2024). AI Educ. 2026,1, 2 4 of 13 When discussing students’ use of LLMs and academic integrity, one of the main concerns presented is that of the development of knowledge and critical thinking skills in students. (Cong-Lem et al.,2024;Khatri & Karki,2023;Balalle & Pannilage,2025;Salehi et al.,2025;Yeo,2023). If a student is overly reliant on AI tools to complete assignments, then there is no guarantee that the student is actually reaping any knowledge from the activity (Gupta,2024;Khatri & Karki,2023). This brings concern not only in the context of an academic institution but also in the context of the fields in which these students will work in in the future; this is especially a concern for subject fields in which assessment is heavily based in the quality of a student’s written work, such as the social sciences, arts, and humanities (Gupta,2024). This concern over the development of critical thinking is also in line with the concern regarding the originality of students’ work. When discussing the use of generative AI for writing content, debates over what constitutes authorship and plagiarism arise (Lund & Naheem,2024). According to Yeo (2023), there are no universally accepted definitions for authorship or for plagiarism; however, there is a general agreement that “authorship” requires that the work be the person’s own original work, and that “plagiarism” involves using someone or something else’s content without credit. While usage of generative AI for essays may not constitute plagiarism, it can reasonably be determined to be “false authorship,” since the student did not write the essay themselves. Functionally, it is almost the same as having a peer write the essay in place of the student. Additional concerns over the usage of LLMs in education persist as well. The topic of student responsibility is a significant one, especially with it being a core tenet of the idea of academic integrity itself (Alio˘gulları et al.,2025;International Center for Academic Integrity [ICAI],2021). The accuracy of LLMs is also a topic of discussion, since LLMs can generate unintentional biases and “hallucinations,” or false facts, and present them as truth (Alio˘gulları et al.,2025;Salehi et al.,2025). Irresponsibility in students can greatly impact the credibility of an academic program if occurring on a large enough scale, as can the decrease of critical thinking skills and accurate knowledge within a student population. While the rise of AI-generated writing in academia is of great concern, there are potential uses of AI that would still fit within the ideals of academic integrity. Yeo (2023) discusses the usage of writing assistants, or what might be called “revision tools,” such as Wordtune, which do not generate their own content and instead make subtle changes to text that the user has already provided. If discussing the usage of such tools and the legitimacy of a one’s authorship, it could be reasonably argued that using a tool such as Wordtune would still allow for one to be considered the original author, since they provided the original idea and text simply used Wordtune to revise their work (Yeo,2023). These arguments could apply to other tools such as Grammarly, another revision tool that provides revision and editing services but does not fully generate a written work. These ongoing debates about authorship, responsibility and the misuse of generative AI in higher education contexts can be further understood from the perspective of classical ethical theory. From a utilitarian perspective, we can view AI use in relation to its consequences for student learning, institutional credibility, and societal outcomes. Because overreliance on AI in academic preparation may result in the erosion of trust in academic standards and limit intellectual growth, the utilitarian perspective suggests that academic integrity policy should limit use of generative AI tools for composing manuscripts. But students are unlikely to use this same utilitarian calculus in making decisions whether to use these tools. AI Educ. 2026,1, 2 5 of 13 2.3. Student Perceptions of AI Use and AI Ethics As AI applications become increasingly prevalent in educational settings, there has been substantial academic interest in students’ perspectives on their purposes and ethical implications. As these technologies become a routine part of educational practice, learners recognize both the benefits and challenges they present, which informs efforts to promote responsible use and maintain academic integrity. This tension between AI’s educational potential and the complexities it introduces provides a critical foundation for developing informed policies, practices, and pedagogies. One study by J. E. Lee and Maeng (2023) explored AI systems perception among 30 high school students in South Korea. They found that students valued chatbots for their convenience and efficiency, meaning a rating of 4.33. Students liked that they could find information without temporal or spatial bounds, and they believed that the chatbot was user-friendly, rating it at 3.87 out of 5 for usability. The study also showcased some ethical concerns students had about the chatbot systems, particularly around plagiarism and copyright, with a mean rating of 3.80 for concerns about originality and copyright issues. Students expressed concerns about their personal data being breached, which demonstrates some understanding of privacy risks. Students surveyed with no previous experience using a chatbot were more skeptical about ethical issues and educational concerns, such as the potential for over-dependence on chatbots undermining exploratory learning than students who had previously used chatbots for English language learning. J. E. Lee and Maeng (2023), recommend that “teachers should provide educational guidance for students to take a critical approach to information provided by the chatbots” (J. E. Lee & Maeng,2023, p. 69). This represented the AI’s duality as a positive educational resource and a source of ethical challenges, especially for younger students. Another study by Xiao et al. (2023) showed a mixed method to analyze its effect on academic integrity on learners. Their findings showed that students valued AI as a tool, such as ChatGPT to assist drafting, but had concerns regarding plagiarism and diminished authentic learning experiences. It points out, “some students struggle to put their moral views into words, which reach a path in using assisted writing tools” (Xiao et al.,2023, p. 45). They reported that 60% of participants had used these services for academic work, and eventually, 72% responded in the affirmative to ethical matters, for cheating. On one hand, this echoes the complexity and duality of situations when embracing AI, with students seeing its promise as a learning aid. This deep study recommends educational policy to promote responsible ethical use of tools so that organizations develop guidelines that clarify ethical risks while tapping into the benefits of AI. In addition, other studies analyze student perceptions of AI tools across diverse educational and cultural contexts. They found that cultural context and institutional factors affect attitudes toward AI ethics. For example, participants were generally more concerned about plagiarism and academic dishonesty in a Western context, as opposed to Asian contexts, where privacy ultimately was a bigger concern when deciding to engage with the technology. Tlili et al. (2023) stated, “Students in Asia expressed more trepidation about any implications regarding data privacy while acknowledging risks, 68% expressed risk of personal information leakage” (Tlili et al.,2023, p. 12). The study suggested that globally, 55% of students used AI tools for academic purposes with perceived cognitive advantages of improved writing and problem-solving skills. Yet, ethical concerns were ubiquitous regardless of location, with 70% of participants having diminished critical thinking. Essentially, these showcase the applicability of ethical frameworks of students’ perceptions addressed to responsible engagement with AI technologies. Moreover, a study by Yu (2023) examined the longer-term ethical use in education by surveying 200 undergraduate students using AI-assisted writing. A significant finding of AI Educ. 2026,1, 2 6 of 13 the study reported that 62% of students ascribed to the use of AI for academic purposes, but 80% said they were concerned that students would lose their sense of independent thought due to overuse or becoming too reliant on AI. Yu (2023) states, “Over reliance on AI might elicit ethical and potentially privacy issues that may lessen authorial ownership” (Yu,2023, p. 5). Students raise concerns about biased outputs and the presence of prejudiced content in AI systems. Despite these issues, many students recognize the value of AI tools for their effectiveness and efficiency, especially given the risk of potential overreliance in educational contexts. Moreover, the study found that three-quarters of students reported that AI improved the quality of writing assignments. Educators of future school leaders will need to employ critical thinking and ethical consideration in their learning as a counterbalance to the negative impacts of AI. Overall, this study stressed the importance of practitioners examining the implications of AI on education in the short term, while also developing ethical implications and academic integrity. 2.4. Research Problem and Questions The rapid adoption of AI tools within higher education has prompted concerns about their ethical use, particularly when used in student writing (T. Wang et al.,2023). While many institutions have responded by developing new academic integrity policies related to student AI use, it remains unclear whether students’ awareness of these policies influences their ethical perceptions or their actual use of AI tools. Additionally, the relationship between students’ personal ethical beliefs and their behavior with AI-assisted writing tools is not well understood in the current literature. There is a significant need to explore the ways students interpret and act upon academic integrity standards in the context of AI, and what drives the acceptance or rejection of AI-assisted writing as an ethical academic practice. This problem informs the following research questions: 1. How does students’ awareness of academic integrity policies influence their perceptions of AI-assisted writing? 2. Do students’ perceptions of AI-assisted writing predict their actual AI tool usage? 3. How do policy awareness and ethical beliefs influence students’ perceptions of the severity of academic misconduct involving AI-assisted writing? 4. What are the strongest predictors of whether students perceive AI-assisted writing as an ethically acceptable academic practice? 3. Methods: A Survey of AI and Academic Integrity This study utilizes an electronic survey/questionnaire method. The questionnaire was created using an online survey platform (Qualtrics) and distributed electronically to participants. Questions were developed by the authors based on the existing literature and then refined through pilot testing with a small group of students at their institution. Eligible participants for the survey were students enrolled in higher education institutions who were at least 18 years old at the time of survey completion. The survey was disseminated via email to students at several major universities across the United States, with recipients encouraged to share the survey link with others to expand participation. It remained open from 1 April to 1 May 2024. During this period, 521 responses were collected, of which 401 were valid and complete. These responses were exported from the survey platform into a CSV file for subsequent statistical analysis using the Python package statsmodels 0.14.4. The questionnaire consisted of ten multi-part questions. This includes five questions relating to student demographics—educational status (undergraduate, masters, doctoral), major, residency status (domestic/international), and gender. These questions are followed by five question categories comprising a total of 22 multiple choice and Likert items related to AI use and perceptions of academic integrity. Among the AI and academic integrity AI Educ. 2026,1, 2 7 of 13 questions, students were asked to rate various activities based on their perceived level of academic misconduct (on a scale from “not academic misconduct at all” to “major academic misconduct”), the seriousness of AI use for various academic activities, and a series of statements related to AI rated on a five-point scale from “strongly disagree” to “strongly agree,” including, “Cheating on assignments is unethical,” “Cheating is okay as long as I don’t get caught,” and “Using AI to help write papers is cheating.” Survey data was transferred to a spreadsheet for further analysis. A series of regression analyses were performed to ascertain any significant relationships among the variables, according to the four research questions posed for this study. For the first analysis, we looked at the relationship of responses to questions pertaining to AI policy awareness and views of AI use on assignments as ethical. For the second analysis, we examined the relationship between students’ ethical beliefs about AI use and their self-reported engagement in AI-assisted writing. For the third analysis, we assessed how students’ ethical views, beliefs about cheating, and educational level predicted their perception of the seriousness of AI-assisted writing as academic misconduct. For the fourth analysis, we investigated which factors—including views on cheating, perceived misconduct seriousness, and policy awareness—best predicted students’ ethical acceptance of AI-assisted writing. 4. Results 4.1. The Role of Policy Awareness in Shaping Ethical Views of AI-Assisted Writing Universities have issued guidelines to address the ethical use of artificial intelligence (AI) in academic work, yet the effectiveness of these policies remains an open question. Specifically, does knowing the rules actually influence what students believe? To address this, we asked: RQ1: How does students’ awareness of academic integrity policies influence their perceptions of AI-assisted writing? Because the dependent variable—ai_use_ethical—is ordinal (ranging from 1 to 6 on a Likert scale), we used Ordinal Logistic Regression. Results are shown in Table 1. Table 1. Ordinal Logistic Regression Results for Perceived Ethicality of AI Use (Dependent Variable: ai_use_ethical). Predictor Variable Coefficient (β) Std. Error z-Value p-Value 95% Confidence Interval policy_aware 0.464 0.528 0.879 0.379 [−0.570, 1.498] ai_writing_cheating −0.262 0.061 −4.304 <0.001 [−0.382, −0.143] cheating_assignments_unethical −0.071 0.074 −0.969 0.333 [−0.216, 0.073] edu_status −1.056 0.552 −1.915 0.056 [−2.138, 0.025] Note: Statistically significant p-values (p< 0.05) are bolded. The model estimates ordinal logistic regression coefficients, where negative values indicate a lower perceived ethicality of AI use. The regression reveals no significant effect of policy awareness on students’ ethical judgments of AI use (p= 0.379). In other words, knowing the rules doesn’t necessarily shape what students believe is right or wrong. This finding challenges the assumption that institutional policies alone can meaningfully govern students’ ethical perspectives. Instead, the strongest predictor is the belief that AI writing is cheating. Students who equate AI-assisted writing with academic dishonesty are substantially less likely to view it as ethical ( β = − 0.262, p< 0.001). The odds ratio (0.77) suggests they are 23% less likely to deem it acceptable. Importantly, this indicates that ethical attitudes are more strongly shaped by internalized beliefs about cheating than by policy exposure. Interestingly, a student’s broader stance on academic dishonesty—whether they believe cheating is wrong—did not significantly predict how they viewed AI use (p= 0.333). This suggests that students may treat AI writing as a distinct category, not merely as an extension of conventional cheating. It may require its own ethical vocabulary. AI Educ. 2026,1, 2 8 of 13 Academic level also plays a role. The edu_status coefficient is marginally significant (p= 0.056), with graduate students less likely than undergraduates to see AI-assisted writing as ethical. The odds ratio (0.35) implies they are about 65% less likely to approve. This difference may reflect higher standards of originality and rigor in graduate education. 4.2. Do Ethical Beliefs Predict Behavior? Building on RQ1, we then asked: RQ2: Do students’ ethical perceptions of AI-assisted writing predict their actual use of AI tools? To measure behavior, we combined three related ordinal variables—ai_writing_full_paper, ai_writing_section, and ai_revision—into a composite measure: ai_assisted_writing. Each item was scored on a 6-point Likert scale. We calculated the mean across all three behaviors and rounded to the nearest whole number, preserving its ordinal nature. Results of these analyses are shown in Table 2. Table 2. Ordinal Logistic Regression Results for AI-Assisted Writing Perceptions (Dependent Variable: ai_writing_cheating). Predictor Variable Coefficient (β) Std. Error z-Value p-Value 95% Confidence Interval ai_use_ethical −0.309 0.055 −5.608 <0.001 [−0.417, −0.201] ai_writing_cheating 0.339 0.071 4.786 <0.001 [0.200, 0.477] policy_aware 0.367 0.612 0.599 0.549 [−0.833, 1.566] grammarly_pro_revise 0.645 0.061 10.487 <0.001 [0.524, 0.765] Note: Statistically significant p-values (p< 0.05) are bolded. Students who view AI use as ethical are significantly more likely to engage in it. The negative coefficient for ai_use_ethical ( β = − 0.309, p< 0.001) indicates that students who perceive AI as ethically acceptable are less likely to frame its use as misconduct. Likewise, those who believe AI writing is cheating are significantly more likely to see AI-assisted writing as problematic. Again, policy awareness is not a significant predictor (p= 0.549), reinforcing our earlier finding: knowing the rules doesn’t necessarily shape action. Interestingly, students who use Grammarly Pro are nearly twice as likely to use AI tools in writing (OR = 1.91). Familiarity with AI-powered tools appears to reduce resistance to more advanced writing assistants like ChatGPT. 4.3. What Predicts the Perceived Severity of AI Misconduct? We next asked: RQ3: How do students’ ethical beliefs and policy awareness influence their perception of the severity of academic misconduct involving AI writing? The same linear regression analyses were used with this new dependent variable. Findings are shown in Table 3. Table 3. Linear Regression Results for Perceived Seriousness (Dependent Variable: Perceived_Seriousness). Predictor Variable Coefficient (β) Std. Error t-Value p-Value 95% Confidence Interval Intercept (const) 0.277 0.747 0.371 0.711 [−1.191, 1.746] policy_aware 0.194 0.274 0.709 0.479 [−0.344, 0.732] ai_use_ethical −0.164 0.024 −6.938 <0.001 [−0.211, −0.118] ai_writing_cheating 0.242 0.031 7.791 <0.001 [0.181, 0.304] cheating_assignments_unethical 0.097 0.037 2.595 0.010 [0.024, 0.171] cheating_assignments_hurts_others 0.168 0.035 4.781 <0.001 [0.099, 0.237] edu_status 1.049 0.295 3.552 <0.001 [0.469, 1.629] Note: Statistically significant p-values (p< 0.05) are bolded. AI Educ. 2026,1, 2 9 of 13 The results reinforce a key theme: ethical beliefs—not policy awareness—are the strongest predictors. Students who see AI-assisted writing as ethical are less likely to view it as serious misconduct ( β = − 0.164, p< 0.001). Conversely, students who view AI writing as cheating take it more seriously (β= 0.242, p< 0.001). Broader moral beliefs also matter. Students who view cheating as unethical or harmful to others are more likely to classify AI use as serious misconduct. Once again, policy awareness does not predict perceived seriousness (p= 0.479). Educational level, however, does: graduate students are significantly more likely to view AI use as serious misconduct ( β = 1.049, p< 0.001), reflecting perhaps a heightened sensitivity to academic standards at the graduate level. 4.4. What Drives Ethical Acceptance of AI Writing? Finally, we employed an ordinal linear regression to address RQ4: What are the strongest predictors of whether students view AI-assisted writing as ethically acceptable? These findings are shown in Table 4. Table 4. Ordinal Logistic Regression Results for Predictors of Students’ Ethical Acceptance of AIAssisted Writing (Dependent Variable: ai_use_ethical). Predictor Variable Coefficient (β) Std. Error z-Value p-Value 95% Confidence Interval cheating_assignments_unethical 0.142 0.092 1.544 0.123 [−0.038, 0.322] ai_writing_cheating −0.048 0.074 −0.648 0.517 [−0.192, 0.097] perceived_seriousness −0.669 0.155 −4.319 <0.001 [−0.973, −0.365] cheating_ok_if_not_caught 0.629 0.072 8.728 <0.001 [0.488, 0.770] grammarly_pro_revise −0.034 0.089 −0.379 0.704 [−0.209, 0.141] edu_status −0.093 0.555 −0.168 0.866 [−1.180, 0.994] Note: Statistically significant p-values (p< 0.05) are bolded. Two variables emerge as the strongest predictors of ethical acceptance: perceived seriousness and cheating leniency. Students who see AI writing as a serious form of misconduct are far less likely to view it as ethical ( β = − 0.669, p< 0.001). Those who believe cheating is acceptable if not caught are far more likely to endorse AI writing as ethical (β= 0.629, p< 0.001). Surprisingly, believing that AI writing is cheating does not significantly predict ethical acceptance (p= 0.517), nor does simply thinking that cheating is unethical (p= 0.123). These results suggest that students may separate abstract ethical ideals from practical judgments about AI. Additionally, use of Grammarly Pro and educational status do not significantly affect views on AI ethics. Familiarity with AI tools does not automatically imply approval, and graduate students are not categorically stricter—at least not when it comes to ethical acceptance of AI writing. 5. Discussion Patterns that have emerged from this study draw a more nuanced picture of student engagement with generative AI than existing policy documents or institutional statements might suggest. In classrooms where the presence of AI tools like ChatGPT or Grammarly Pro is increasingly commonplace, it is not just the policies that matter—it is how students interpret them through their own ethical lenses. Contrary to what some institutional leaders might assume, mere awareness of a university’s AI policy does little to sway student views on whether using AI is ethical or not. The numbers in this study stress that formal guidance often trails behind the informal norms students create for themselves.