Research on the Impact of Generative Artificial Intelligence Usage Behavior on the Learning Outcomes of Higher Vocational Students

Comments 1: [The manuscript deals with a timely and relevant topic on the relation between generative artificial intelligence use and learning outcomes among higher vocational students. However, the manuscript has major weaknesses conceptually, methodologically, analytically and presentationally. I do not regard the present version as fit for publication.]

Response 1: We sincerely appreciate the reviewer’s positive comments on the topic of our study. We are greatly pleased that this research, which explores the impacts of generative artificial intelligence on vocational college students’ learning outcomes, is recognized for its practical significance and timeliness. Our attempt to develop and validate two scales respectively for measuring generative AI usage behavior and vocational students’ learning outcomes is also acknowledged as innovative and practically relevant. These encouraging remarks demonstrate the considerable research potential of our work at the intersection of vocational education and artificial intelligence.

Meanwhile, we fully take on board the reviewer’s valuable concerns regarding conceptual clarity, methodological rigor, depth of data analysis and overall presentation. We have earnestly addressed all the comments and carried out substantial revisions to respond to every raised issue. We have strived to enhance the rigor and reproducibility of this paper in terms of theoretical construction, research design and empirical analysis to meet the standards of high-quality academic journals. We welcome any further comments and suggestions. This new section can be found on page 8, starting from paragraph 4, line 368-393; page 10, starting from paragraph 2, line 439-447;page 18, starting from paragraph 3, line 16-38;page 22, starting from paragraph 2, line 179-239.The revised parts are marked in red in the modified manuscript.

Comments 2: [Measurement of learning outcomes is very problematic. The study claims to evaluate students’ learning outcomes, but the outcomes are self-reported perceptions, not objective indicators such as grades, practical skills tests, teacher evaluations, internship performance or standardized assessments. This poses a serious risk of social desirability bias and common method bias.The dependent variable is thus not actual learning outcomes but perceived learning outcomes of the students. ## Title, Abstract, Discussion and Conclusion should be adjusted as appropriate.]

Response 2: We sincerely thank the reviewer for pointing out this critical methodological issue. We fully agree that measuring learning outcomes solely based on students’ self-reported perceptions carries risks of social desirability bias and common method bias, which may undermine the objectivity and reliability of the research findings. We address this concern from two perspectives: empirical testing and future research directions.

In terms of empirical testing, we have added a common method bias test in the revised manuscript. Specifically, we conducted an unrotated exploratory factor analysis on all measurement items using Harman’s single-factor test. The results show that the first factor accounts for 43.12% of the total variance, which is below the 50% threshold. This indicates that common method bias is not a serious issue in this study, and the self-reported data do not substantially compromise the reliability of the main conclusions. The relevant results are presented at the beginning of Section 4 in the revised paper.

From the perspective of future research, we explicitly acknowledge the inherent limitations of self-reported data in the section titled Limitations and Future Research. We also put forward targeted improvements. To assess students’ learning outcomes more accurately, follow-up studies are recommended to adopt multi-source data and objective evaluation indicators. On the basis of student self-reports, researchers may further incorporate course grades, practical training scores, teacher evaluations, internship performance, professional skill assessments and certification results, so as to effectively reduce subjective bias caused by relying merely on self-reported data. In addition, we have revised relevant statements in the abstract to clarify that the measured outcomes refer to self-perceived learning outcomes.

Nevertheless, we acknowledge that it is practically difficult to collect the aforementioned objective indicators for this large-scale questionnaire survey. For instance, academic records are scattered across different institutions, assessment criteria vary among majors, and anonymous surveys make it impossible to match participants with their individual grades. Therefore, the self-rating scale adopted in this study remains reasonable and feasible for an exploratory research. We have clearly explained this trade-off in the revised manuscript for readers’ prudent interpretation. This new section can be found on page 13, starting from paragraph 1, line 562-570; page 22, starting from paragraph 1, line 179-198. The revised parts are marked in red in the modified manuscript.

Comments 9: [The dimension of GenAI usage frequency confuses simple frequency with dependence and potential over-dependence. Things like “My learning efficiency would drop significantly without GenAI tools” are measuring dependence, not frequency. Perhaps this is part of the explanation for the negative regression coefficient.]

Response 9: We sincerely appreciate the reviewer’s perceptive observation. We fully agree that the items under the dimension of usage frequency involve minor conceptual confusion. We have accordingly proposed solutions for future research in this regard. As stated in Section 5, subsequent studies should separate usage frequency and dependence into two independent measurement dimensions. To be specific, usage frequency should focus on behavioral frequency, such as how many times the tools are used per day, while dependence should reflect cognitive and emotional reliance, for instance, whether students feel inconvenient without these tools. Such refined dimensional division can help more accurately examine their respective independent effects on learning outcomes.

Nevertheless, we would like to note that the negative coefficient of the usage frequency dimension derived from the regression results still has reference value for both theoretical and empirical research. It at least reveals that simply increasing usage frequency cannot improve learning outcomes and may even produce adverse effects. This finding also delivers meaningful implications for the practice of vocational education. We hope the above clarification fully addresses your concerns. This new section can be found on page 22, starting from paragraph 1, line 186-189. The revised parts are marked in red in the modified manuscript.

Comments 10: [The manuscript does not talk enough about academic integrity. Some items are using GenAI for homework, writing a thesis, writing code, and assignments so the paper should talk about plagiarism, unauthorized assistance, assessment validity, and AI policies of the institution.]

Response 10: We sincerely thank the reviewer for pointing out this important omission. We fully recognize that when generative artificial intelligence is used to assist students with assignments, thesis writing, coding and other tasks, it inevitably raises academic integrity issues including plagiarism, unauthorized assistance, assessment validity and institutional AI policies. The original manuscript indeed lacked adequate discussion on these topics.

We have added relevant content to the Future Research section in the revised paper. Specifically, we state at the end of Section 5 that this study has not fully explored academic integrity issues associated with the use of generative AI. Given that such tools can be applied to finish coursework, write papers, develop codes and complete various tasks, future research should delve into plagiarism, unauthorized help, assessment validity and relevant institutional AI policies. It is also necessary to further clarify the ethical boundaries and behavioral norms for students using generative AI. We hope these additions can fully address your concerns. This new section can be found on page 22, starting from paragraph 4, line 221-230. The revised parts are marked in red in the modified manuscript.

Comments 11: [The discussion is too optimistic and sometimes speculative. The authors jump too quickly from weak correlational evidence to sweeping recommendations for colleges, teachers, and students. While these recommendations may be reasonable, they are not well supported by the current empirical design.]

Response 11: We sincerely appreciate your impartial and constructive comments. We fully acknowledge that several parts of the original manuscript contained overinterpretation, and some universal recommendations were drawn directly from correlational data. Accordingly, we have systematically revised all suggestions throughout the paper to ensure that they are strictly grounded in our empirical findings and closely linked to the research results. The detailed revisions are presented as follows.

Based on the results, colleges, teachers and students need to work collaboratively with clear respective priorities to help vocational college students develop rational generative AI usage behaviors and improve their learning outcomes. For colleges, the focus should shift from simply encouraging AI use to standardizing usage scenarios and fostering good usage habits. This study reveals that generative AI usage contexts have a significantly positive correlation with learning outcomes. Therefore, rather than merely advocating the use of AI in vague terms, colleges should formulate clear behavioral guidelines for different learning scenarios. Relevant regulations can prevent the abuse of AI in inappropriate situations and maximize the positive effects of rational usage contexts on learning outcomes. Considering the notable differences in learning outcomes across various majors, colleges should avoid adopting a one-size-fits-all strategy for AI promotion. For knowledge-intensive majors, support can be prioritized for applying AI to knowledge integration and expansion. For skill-oriented majors, more attention should be paid to developing AI’s auxiliary functions in operational process simulation and case analysis. Meanwhile, major-specific prompt libraries and application templates can be established, and AI literacy can be appropriately integrated into the curriculum system (Hershkovitz et al., 2025).

For teachers, the emphasis needs to change from demonstrating tool functions to systematically guiding proper usage habits. Given the positive statistical correlation between usage habits and learning outcomes, vocational teachers, as learning facilitators, should help students understand the working mechanisms, strengths and limitations of generative AI, and cultivate their critical thinking and ethical awareness (Bower et al., 2024). Teachers can focus on key practices including prompt refinement, content verification, information screening, result comparison and tool selection. Combined with the practice-oriented features of vocational education, teachers can design targeted teaching activities based on real application scenarios of professional courses (Wang et al., 2024).

For students, it is essential to move from being able to use AI tools to using them proficiently. As direct users of generative AI, students should continuously improve their digital literacy. They need to recognize that simply increasing AI usage frequency may exert adverse impacts on learning outcomes, and refrain from cognitive offloading caused by over-reliance on generative AI for information retrieval and decision-making (Gerlich, 2025). In addition, students should enhance their ability to apply AI across diverse learning scenarios, and explore appropriate ways to utilize generative AI in knowledge acquisition, skill training, project practice and vocational simulation. Structured use of generative AI can ultimately help improve academic performance (Xu et al., 2025).

Thank you again for your valuable suggestions. We believe these revisions make the recommendations better aligned with our empirical evidence and avoid excessive speculation. This new section can be found on page 20, starting from paragraph 3, line 109-157. The revised parts are marked in red in the modified manuscript.

Comments 12: [The manuscript needs to be significantly shortened and reorganized. Some parts are too descriptive and the analytical contribution is somewhat limited.]

Response 12: We sincerely thank the reviewer for pointing out this structural issue. We have thoroughly revised the entire manuscript by streamlining descriptive content and adding in-depth theoretical analysis and elaboration. For instance, we have supplemented integrated explanations of the theoretical framework in Section 3.2.1 Development of the Generative AI Usage Behavior Scale. Specifically, this study systematically analyzes vocational college students’ generative AI usage behaviors by integrating Davis’s (1989) Technology Acceptance Model (TAM), Venkatesh et al.’s (2003) Unified Theory of Acceptance and Use of Technology (UTAUT), and existing assessment frameworks for AI literacy. This theoretical integration provides a more solid foundation for defining the three core dimensions, namely usage frequency, usage habits and usage contexts, and also strengthens the analytical contribution of the scale development.

With the above revisions, we have cut redundant descriptions and enhanced theoretical underpinnings and analytical depth throughout the paper. We hope these changes can optimize the overall structure, highlight the analytical value of this research, and fully address your concerns. This new section can be found on page 8, starting from paragraph 4, line 368-393. The revised parts are marked in red in the modified manuscript.

Comments 1: [Thank you for submitting. I was pleased and honoured to review your manuscript. Unlike some reviewers, I purely judge the paper on how it reads and what it says, rather than what I think it should be. This way, you are assured a fair review and any guidance I give will be to strengthen the paper, not my own opinions on how you could do something the way I would!!

Overall, I do agree this manuscript addresses a timely and important issue: the relationship between generative artificial intelligence usage and learning outcomes among higher vocational students. The focus on higher vocational education is a really clear strength, well done, as much of the existing literature on GenAI in education has concentrated on undergraduate or general higher education contexts. The paper usefully argues that higher vocational students differ from undergraduates in their learning characteristics, practical training needs, and employment-oriented learning outcomes. This gives the study a distinctive rationale and helps justify the development of context-specific measurement tools. I really enjoyed the conceptual ideas here.

The manuscript’s attempt to develop and validate two scales—the GenAI Usage Behavior Scale and the Higher Vocational Students’ Learning Outcomes Scale—is also valuable. The distinction between usage frequency, usage habits, and usage contexts is particularly promising, because it moves the discussion beyond whether students use GenAI and instead considers how they use it. The paper’s emphasis on usage habits, including prompt adjustment, verification of AI outputs, comparison across tools, and critical adoption of generated content, is especially relevant to current debates about responsible and effective GenAI use in education. As such, it has a strong foundation.

At the same time, respectfully, the manuscript requires revision before it is ready for publication. The central contribution is potentially strong, but the current version overstates what the design can demonstrate, underreports key methodological details, and needs clearer statistical interpretation. The paper would benefit from a more cautious framing, fuller transparency in scale validation, a better literature review, and more careful handling of the relationship between GenAI use and learning outcomes.]

Response 1: We sincerely appreciate the reviewer’s kind recognition and encouragement regarding our work. Thank you for acknowledging that this study focuses on vocational education, a field that has received relatively little attention in research on generative AI in education, and for affirming its practical significance. The emphasis on the "usage habits" dimension of generative AI usage behavior is also recognized as highly relevant to current discussions on the responsible and effective application of AI. These positive comments confirm that our research is built upon a promising and meaningful foundation.

Meanwhile, we fully take on board your important concerns covering four aspects: insufficient evidence to support certain arguments, incomplete reporting of methodological details for scale validation, lack of clarity and prudence in the interpretation of statistical results, and inadequate coverage of local contexts in the literature review. We have taken all these comments seriously and carried out substantial revisions to address every issue raised. We have strived to bring the revised manuscript up to the journal’s publication standards in terms of theoretical construction, methodological rigor and the clarity of statistical interpretation. We welcome any further feedback. This new section can be found on page 8, starting from paragraph 4, line 368-393; page 10, starting from paragraph 2, line 439-447;page 18, starting from paragraph 3, line 16-38;page 22, starting from paragraph 2, line 179-239.The revised parts are marked in red in the modified manuscript.

Comments 2: [Develop further Clarity on Research Design and Causal Framing

A major developmental area concerns the manuscript’s repeated use of causal language. The paper frequently refers to the “impact” of GenAI usage behavior, “influencing mechanisms,” and the “effects” of usage frequency, habits, and contexts on learning outcomes. However, your study is based on cross-sectional questionnaire data. Although correlation and multiple regression analyses can identify statistical associations, they cannot really establish causal relationships or causal direction.

This issue is particularly important because the manuscript interprets regression coefficients as if one-unit increases in GenAI usage dimensions directly lead to increases or decreases in learning outcomes. This is not, in my opinion, warranted by the research design. For example, students with stronger learning outcomes may develop better GenAI usage habits, rather than better habits necessarily producing stronger learning outcomes. Similarly, students who struggle academically may use GenAI more frequently, which could partly explain the negative regression coefficient for usage frequency. There’s a lot of scope there.

This isn’t difficult to fix. The manuscript should therefore revise its language throughout. Terms such as “impact,” “effect,” and “influence” should be replaced with more cautious expressions such as “is associated with,” “predicts in the regression model,” or “is statistically related to.” The title, abstract, discussion, and conclusion should all be checked for causal overclaiming. If the authors wish to make causal claims, they would need a stronger design, such as a longitudinal study, intervention design, quasi-experimental approach, or the inclusion of stronger controls and causal modelling assumptions. I understand you want your work to be picked up and cited, but in essence it’s more likely to if you neutrally frame your conclusions, as people will take it more seriously.]

Response 2: We sincerely thank the reviewer for pointing out this critical issue. We fully agree that a research design based on cross-sectional questionnaire data cannot establish causal relationships. The frequent use of causal terms such as impact, effect and influencing mechanism in the original manuscript indeed goes beyond what this study can substantiate. We have systematically revised the entire paper and replaced causal wording with more prudent expressions that describe statistical associations.

The detailed revisions are as follows. First, in the abstract, the original statement "The usage frequency of generative artificial intelligence exerts a significantly negative impact on learning outcomes, while usage habits and usage contexts show markedly positive effects, with usage habits demonstrating stronger explanatory power" has been revised to "Generative AI usage frequency has a significant negative statistical correlation with learning outcomes, whereas usage habits and usage contexts are significantly and positively correlated with learning outcomes, and usage habits present the strongest explanatory power".

Second, in the regression analysis section, the original text reads: "With other variables held constant, each one-unit increase in generative AI usage frequency corresponds to a 0.137-unit decrease in learning outcomes; each one-unit increase in usage habits leads to a 0.248-unit rise in learning outcomes; and each one-unit increase in usage contexts brings about a 0.205-unit growth in learning outcomes. In terms of the direction of effects, generative AI usage frequency has a significant negative effect on vocational college students’ learning outcomes, while usage habits and usage contexts both exert significant positive effects. In terms of effect magnitude, usage habits have the greatest influence on learning outcomes, followed by usage contexts, and usage frequency has the weakest influence."

This passage has been revised to: "In terms of correlation direction, generative AI usage frequency is significantly and negatively correlated with vocational college students’ learning outcomes, which is contrary to the positive prediction proposed in Hypothesis H1. Therefore, Hypothesis H1 is not supported. Nevertheless, this negative coefficient does not mean that usage frequency itself produces adverse effects on learning outcomes. Among vocational college students, high usage frequency is often accompanied by aimless use without clear learning objectives, over-reliance on AI-generated content, and the lack of active processing and in-depth thinking. These unfavorable habits fail to translate into desirable learning results, thus leading to a negative statistical correlation. By contrast, both usage habits and usage contexts serve as significant positive predictors of learning outcomes, which supports Hypotheses H2 and H3. The results indicate that in generative AI-assisted learning, how and in what contexts the tools are used are positively correlated with learning outcomes. In terms of correlation strength, usage habits show the strongest statistical association with learning outcomes, followed by usage contexts and then usage frequency. This finding is consistent with Hypothesis H2 which assumes that usage habits have the strongest positive predictive effect, so Hypothesis H4 is supported."

Third, in the discussion section, the original sentence "Finally, this study verifies the impacts of different dimensions of generative AI usage behavior on vocational college students’ learning outcomes. Specifically, usage frequency has a significant negative impact on learning outcomes, while usage habits and usage contexts both show significant positive impacts" has been revised to "Finally, this study examines the regression predictive effects of different dimensions of generative AI usage behavior on vocational college students’ learning outcomes. In the regression model, generative AI usage frequency acts as a significant negative predictor of learning outcomes, while usage habits and usage contexts are significant positive predictors." We have also replaced strongly causal words like impact with neutral phrases such as predicts in the regression model and is statistically correlated with.

We hope these comprehensive revisions fully address your concerns regarding causal language and make the overall writing more rigorous and prudent. We fully understand that presenting conclusions in a neutral and accurate manner is fundamental for the research to be taken seriously and widely cited. Thank you again for your valuable suggestions. This new section can be found on page 1, starting from paragraph 1, line 26-32; page 18, starting from paragraph 3, line 16-38;page 20, starting from paragraph 2, line 92-98.The revised parts are marked in red in the modified manuscript.

Comments 3: [Scale Development and Validation

The scale development work is really interesting, and in my view one of the most promising aspects of the paper, but it is currently underreported. The manuscript states that the two scales demonstrate good reliability and validity, and it reports high Cronbach’s alpha values, KMO values, and Bartlett test results. These are useful indicators, well done, but they are not sufficient on their own to establish strong scale validity. In my view, the paper may want to provide the full item wording for both scales, either in the main text or as an appendix. Readers need to see exactly how usage frequency, usage habits, usage contexts, knowledge mastery, skill application, and competency development were operationalized. The manuscript should also include factor loadings, cross-loadings, variance explained, item-total correlations, and the criteria used to remove items. At present, the reader is told that some items were removed because their factor loadings did not match theoretical expectations, but the evidence is not shown in enough detail. The avoidance of this seems a bit… odd, and might make the readers think something is not right. So, add greater clarity please?

A further issue concerns the pilot sample. The paper reports that a preliminary survey was conducted among 40 higher vocational students, I believe? However, it is not entirely clear whether the exploratory factor analysis was conducted on this small pilot sample or on the full formal sample? If factor analysis was performed on only 40 participants, the sample size is far too small for robust scale validation. If the full sample was used, the procedure should be clarified. Ideally, the authors should conduct exploratory factor analysis on one subsample and confirmatory factor analysis on another. This would provide much stronger evidence for the dimensional structure of the scales. In essence, explain your decision making in greater methodological detail.]

Response 3: We sincerely appreciate the reviewer’s recognition and encouragement regarding our scale development work. We are greatly pleased to learn that the Generative AI Usage Behavior Scale and the Vocational College Students’ Learning Outcomes Scale are regarded as one of the most promising core contributions of this paper. Such positive feedback confirms that our research has laid a solid foundation for the development of measurement tools. Meanwhile, we fully acknowledge your concerns about the completeness of the scale validation reporting and have made substantial revisions to address all the issues raised.

First, we have supplemented the reporting of statistical indicators including factor loadings and explained the reasons for item deletion in detail. We agree that merely presenting Cronbach’s α coefficients, KMO values and Bartlett’s test results is insufficient to fully validate the scale. Therefore, we have added the following content in Section 3.2.3 of the revised manuscript:

According to the rotated component matrix, Item C1 has factor loadings of 0.581 on Factor 1 and 0.528 on Factor 3, with a difference of only 0.053, which fails to meet the basic criteria for discriminant validity. In addition, the loading of C1 on its theoretically assigned Factor 3 is not significantly higher than that on Factor 1, and both loadings exceed the acceptable threshold of 0.5, indicating no distinct primary factor for this item. This is inconsistent with other items that show high loadings on a single factor, which impairs the parsimony and discriminant validity of the factor structure. Hence, Item C1 was removed, and factor analysis was reperformed on the remaining items to verify the stability of the revised structure. Ultimately, three factors were extracted. Factor 1 consists of Items B1 to B5 with factor loadings ranging from 0.799 to 0.837; Factor 2 comprises Items A1 to A5 with factor loadings between 0.658 and 0.862; Factor 3 includes Items C2 to C6 with factor loadings from 0.526 to 0.798.

For the Vocational College Students’ Learning Outcomes Scale, the KMO value is 0.961 and the p-value of Bartlett’s Test of Sphericity is less than 0.001, indicating that the data is suitable for factor analysis. The rotated component matrix shows that Item I1 has a loading of 0.496 on its theoretically assigned Factor 1, below the acceptable threshold of 0.50, and a loading of 0.595 on Factor 2. The difference between the two values is merely 0.099, which does not satisfy the requirements for discriminant validity and contradicts the theoretical classification. Accordingly, Item I1 was deleted. Three factors were finally retained. Factor 1 contains Items I2 to I6 with factor loadings ranging from 0.598 to 0.792; Factor 2 is composed of Items H1 to H3 with factor loadings between 0.750 and 0.811; Factor 3 includes Items J1 to J3 with factor loadings from 0.717 to 0.792.*

Second, we clarify and correct the information about the pilot sample size. We sincerely apologize for a critical typographical error in the original manuscript. The statement that "a small-scale pilot survey was conducted among 40 students from a vocational college" was incorrect. In fact, we distributed 400 questionnaires for the pilot survey rather than 40. This mistake has been fully corrected in the revised version.

To specify, a total of 400 questionnaires were distributed in the pilot phase, and 329 valid responses were retrieved. We conducted item analysis, exploratory factor analysis and reliability tests based on these 329 valid samples. For the formal survey, we collected another 977 independent samples for confirmatory factor analysis and subsequent hypothesis testing. Separating the pilot sample from the formal sample effectively avoids circular reasoning caused by conducting both exploratory and confirmatory analysis on the same dataset. Relevant explanations have been added in Section 3.2.3.

We believe these supplementary reports and clarifications fully address your concerns and enhance the transparency of the entire scale development process. Thank you for helping us improve the methodological rigor of this paper. This new section can be found on page 11, starting from paragraph 2, line 475-493;page 7, starting from paragraph 1, line 302-308; page 10, starting from paragraph 7, line 453-454.The revised parts are marked in red in the modified manuscript.

Comments 4: [Sampling, Representativeness, and Data Reporting

The manuscript benefits from a relatively large final sample of 977 valid responses. That’s really impressive. However, there are important reporting issues that need to be addressed. The methodology section, I believe, states that 1,000 questionnaires were distributed, while the data collection section, I think, later states that 1,329 questionnaires were collected and, confusingly, 977 valid responses were retained after excluding invalid responses. This inconsistency needs to be resolved. The authors should clarify whether 1,000 was the intended target, the minimum planned sample, or the number initially distributed. In essence, please ensure the numbers are right or double check. The paper should also provide a transparent sample flow. A short response-flow table would be useful, showing the number of questionnaires distributed or accessed, the number submitted, the number excluded, the exclusion criteria, and the final valid sample. This would improve the credibility of the data-cleaning process, and is an easy change to make. If you do not want to relabel all the tables, it’s also fine if you want to write this in a short paragraph of clarification instead – I do not want to add work to your paper.

Representativeness is another concern. The sample is heavily weighted toward first-year students, while third- and fourth-year students are underrepresented. This is fine and understandable if upper-year students were away on internships, or you just wanted to focus on the transition/starting point, but first, I might add some clarity around that – you wanted to capture them early on to understand the start points maybe. Likewise, it is a decision limits the generalizability of the findings. Your language needs to match- this is early on, vocational students for reason X and our conclusions, Y, are specific to this group. The distribution across major categories also appears uneven. Some categories seem to have very small sample sizes, which may affect the reliability of group comparisons. The authors should consider adding into the paper a report the number of students in each grade and major category and discuss the implications of unequal group sizes more explicitly.]

Response 4: We sincerely thank the reviewer for acknowledging the size of our sample. We also fully recognize the prominent issues regarding the transparency of sample reporting and sample representativeness.

First, we clarify the inconsistency in the reported sample figures. We have clearly distinguished and elaborated on the procedures of the pilot survey and formal survey in Section 3.3 Survey Data Collection of the revised manuscript. In the pilot phase, we distributed 400 questionnaires at a vocational college and obtained 329 valid responses, which were used for item analysis and exploratory factor analysis. For the formal survey, we collected data from a new independent sample. A total of 1,100 questionnaires were distributed and 1,000 were retrieved. After excluding responses with uniform answers, we finally obtained 977 valid questionnaires.

Second, we address the concerns about sample representativeness. We agree that the overrepresentation of first-year students and the severe underrepresentation of senior students have limited the external validity of the research findings. We have taken two measures in the revised manuscript. On the one hand, we present objective results and deliver prudent interpretations in the results section. Section 4.2 illustrates the sample distribution across different grades, and we explicitly state that the sample mainly reflects the learning status of first-year students. Thus, cautions should be exercised when generalizing the conclusions to senior students. We have also added a note in this section: In addition, several major categories including Resource, Environment and Safety, Public Security and Judicial, and Public Administration and Service have very small sample sizes (n ＜ 5) with a standard deviation of 0.00. The results of inter-group comparisons should therefore be interpreted with caution for readers’ reference.

On the other hand, we explicitly acknowledge this limitation in the section of Limitations and Future Research in Section 5. We point out the large disparity in sample size across grades, with a surplus of junior students and a shortage of senior ones, which may undermine the stability and generalizability of the findings. Besides, the small sample sizes of certain major categories may weaken the statistical power of inter-group comparisons. We suggest that future studies recruit participants from vocational colleges across diverse regions, tiers and institutional types, and strive to balance the proportion of students in each grade and major category. Increasing the number of senior students and participants from underrepresented majors will help improve sample representativeness and the external validity of conclusions. We also explain the practical difficulties in recruiting senior participants: most third- and fourth-year students are engaged in off-campus internships, making it far more challenging to distribute and collect online questionnaires compared with junior students who attend regular in-person courses on campus. This objective constraint is fully stated in the revised manuscript for readers’ careful interpretation.

All the above revisions are included in Sections 3.3, 4.2 and 5. We hope these clarifications and supplements fully address your concerns. This new section can be found on page 12, starting from paragraph 4, line 543-546;page 14, starting from paragraph 4, line 633-636; page 22, starting from paragraph 3, line 209-220.The revised parts are marked in red in the modified manuscript.

Comments 5: [Statistical Analysis and Interpretation

The statistical analysis is generally appropriate for an initial exploratory study, but several aspects require a bit more explanation. The correlation analysis shows positive relationships between GenAI usage dimensions and learning outcomes. However, the regression model shows that usage frequency becomes a negative predictor after usage habits and usage contexts are controlled. The manuscript interprets this as evidence that higher GenAI usage frequency negatively affects learning outcomes, but this interpretation is too strong.

A more cautious explanation is likely needed. The negative coefficient may indicate that frequent use without effective habits or meaningful contexts is associated with weaker outcomes. It may also reflect a suppression effect, since usage frequency is positively correlated with other GenAI usage dimensions. The authors could explore this possibility rather than presenting the result as straightforward evidence that frequent GenAI use is harmful – the data does not support that. Additional analyses could strengthen the paper, such as hierarchical regression, interaction effects between usage frequency and usage habits, nonlinear tests of frequency, or robustness checks using alternative scoring methods. However, if you do not want to do that (and I agree that its generally fine as is) then your conclusions need to be more modest here.

The differential analysis also needs refinement. The manuscript reports that grade-level differences are not statistically significant overall, but it discusses a near-significant result for skill application at p = .052 as if it were significant. This should be corrected, I believe? Unless the authors justify a different significance threshold in advance, p = .052 should not be treated as statistically significant, because it is a bit unclear why?! If post hoc comparisons are retained, the authors should explain why they are appropriate after a non-significant omnibus test. That would be a way forward, or simply clarify to a greater extent.

The major-category comparisons also require more careful treatment. Because some groups appear to have very small sample sizes, standard ANOVA assumptions may not hold. The authors should test assumptions, report effect sizes, and consider using Welch ANOVA or combining very small categories where appropriate.]

Response 5: We sincerely thank the reviewer for the detailed suggestions on the interpretation of statistical analyses. We fully agree that the explanations of regression coefficients and difference analysis results in the original manuscript were overly definitive and lacked due prudence.

Firstly, we have revised the interpretation of regression coefficients. As you pointed out, the negative coefficient of usage frequency after controlling for usage habits and usage contexts does not indicate that usage frequency itself exerts adverse effects on learning outcomes. We share this view and have thoroughly revised the relevant content. In Section 4.3.2, we replaced the original statement "usage frequency has a significant negative predictive effect on learning outcomes" with a more cautious explanation: "Nevertheless, this negative coefficient does not mean that usage frequency itself harms learning outcomes. Among vocational college students, high usage frequency is usually accompanied by unregulated use without clear learning objectives, over-reliance on AI-generated content, and the absence of active processing and in-depth thinking. These undesirable learning habits fail to produce positive learning results, thereby leading to a negative statistical correlation."

Secondly, we have corrected the presentation of the p-value in difference analysis. We acknowledge that a p-value of 0.052 should not be regarded as marginally significant. In Section 4.2, we revised the original phrase "The difference across grades is marginally significant (p=0.052)" to "The differences across grades are not statistically significant (p > 0.05)". Meanwhile, we have removed the post-hoc LSD comparisons and the corresponding claim that "First-year students score significantly higher than second-year students in skill application", so as to avoid overinterpreting non-significant findings.

Thirdly, we addressed the issues concerning small sample sizes in the comparisons across major categories. Given that these majors are conceptually distinct and cannot be combined, we retain the original classification covering 19 major categories. We added a note in Section 4.2: "In addition, several major categories including Resource, Environment and Safety, Public Security and Judicial, and Public Administration and Service have very small sample sizes (n ＜ 5) with a standard deviation of 0.00. The results of inter-group comparisons should therefore be interpreted with caution", allowing readers to make reasonable judgments. Furthermore, in the limitations section (Section 5), we noted that the extremely small samples (n ＜ 5) of the above-mentioned majors may violate the assumptions of standard ANOVA and undermine the stability of inter-group comparisons. We suggest that future studies collect more data from these majors to improve the reliability of research conclusions.

We hope these revisions fully address your concerns and enable more prudent and transparent interpretation of the statistical results. This new section can be found on page 18, starting from paragraph 3, line 16-38;page 14, starting from paragraph 3, line 617-621; page 14, starting from paragraph 4, line 633-636.The revised parts are marked in red in the modified manuscript.

Comments 6: [Measurement of Learning Outcomes

The paper’s conceptualization of learning outcomes is really well aligned with vocational education, especially the inclusion of knowledge mastery, skill application, and competency development. However, the measurement appears to rely entirely on students’ self-reported perceptions. This limits the strength of the conclusions. Self-reported learning outcomes may reflect confidence, motivation, or social desirability rather than actual achievement. This limitation should be made more prominent, and indeed a more detailed limitations section is needed overall, towards the end of the paper. The manuscript should refer to “perceived learning outcomes” unless objective indicators are added. To strengthen the study, the authors could incorporate course grades, practical training scores, teacher assessments, internship evaluations, professional skill tests, or certification outcomes. Even if such data cannot be included in the present study, the limitation should be clearly acknowledged and linked to future research directions.

The weighted composite score for learning outcomes also needs fuller justification. The manuscript states that the composite score uses Delphi-derived weights, with knowledge mastery weighted at 0.2 and skill application and competency development each weighted at 0.4. However, the Delphi process is not described in sufficient detail. The authors could instead report the number of experts, number of rounds, consensus criteria, and rationale for the final weights. They should also test whether the main findings remain stable under equal weighting.]

Response 6: We sincerely thank the reviewer for recognizing that our conceptualization of learning outcomes aligns well with the characteristics of vocational education. We also fully acknowledge the deficiencies regarding measurement approaches, weighting methods and the explanation of research limitations.

First, we address the limitations and improvements related to the measurement of learning outcomes. We agree that this study entirely adopts student self-report questionnaires to measure learning outcomes. Such measurements tend to reflect students’ self-confidence, learning motivation and social desirability bias rather than their actual knowledge mastery and skill improvement. Accordingly, we have revised the term "learning outcomes" to perceived learning outcomes in the abstract to accurately define the nature of the variable. We have also added detailed discussions in Section 5 Limitations and Future Research:

This study mainly relies on self-report questionnaires to measure learning outcomes, which is inherently subjective. The measured results may largely reflect students’ confidence, motivation and social desirability instead of their actual knowledge and skills. Future research shall update the constituent elements of generative AI usage behaviors in line with the latest developments and develop more systematic, scientific and context-adaptive assessment tools. Meanwhile, to measure students’ learning outcomes more accurately, follow-up studies should adopt multi-source data and objective evaluation indicators. On the basis of student self-reports, researchers may further incorporate course grades, practical training scores, teacher evaluations, internship assessments, professional skill tests and certification results, so as to effectively reduce subjective bias caused by single-source self-reported data.

Second, we clarify and supplement details about the weighting method. The reviewer pointed out that the original manuscript contained vague descriptions and an incorrect name for the weighting approach, where the Analytic Hierarchy Process was mistakenly referred to as the Delphi method. We have thoroughly revised this part. After re-verification, we confirm that the Analytic Hierarchy Process (AHP) was applied in this research. We have supplemented the complete operating procedures in Section 3.3 Survey Data Collection:

We adopted the Analytic Hierarchy Process to calculate indicator weights for the overall level of vocational college students’ learning outcomes. Fifteen experts in vocational education with over five years of teaching or research experience were invited to conduct pairwise comparisons among three dimensions including knowledge mastery, skill application and competency development using the Saaty 1–9 scale to assess their relative importance. The consistency ratio (CR) of all expert judgment matrices was below 0.1, indicating acceptable consistency. We integrated all judgment matrices via the geometric mean method and obtained the weight vector of each indicator through normalization. After rounding, the final weights for knowledge mastery, skill application and competency development were 0.2, 0.4 and 0.4 respectively.

Third, we conducted a robustness test with equal weights. To examine whether the regression results are sensitive to different weight settings, we performed regression analysis under the equal-weight scheme, where knowledge mastery, skill application and competency development each account for one-third of the total score. The results, as presented in Table 1, are largely consistent with those derived from the original weighted scheme with no substantial changes. Usage frequency still shows a negative correlation, while usage habits and usage contexts remain positively correlated. There are no shifts in the direction or significance level of coefficients, which proves that the core findings of this study are robust against different weight assignments.

We hope these comprehensive revisions fully address your concerns and enhance the transparency and rigor of the procedures for measuring learning outcomes and determining indicator weights. This new section can be found on page 1, starting from paragraph 1, line 26-32; page 22, starting from paragraph 1, line 179-198;page 12, starting from paragraph 3, line 531-540.The revised parts are marked in red in the modified manuscript.

Table 1. Regression Analysis Results of Higher Vocational Students’ GenAI Usage Behavior on Learning Outcomes.（Equal weights）

Comments 7: [Review, Writing, Structure, and Presentation

The manuscript is generally well structured, but it requires careful language editing. There are recurring grammatical errors, spacing issues, awkward phrases, and occasional duplicated words. Examples include subject-verb agreement problems, punctuation errors, and phrases with multiple works in it (See: “This study.”).

These issues do not undermine the value of the research, but they do affect readability and should be addressed before publication. The abstract should be revised after the main methodological and interpretive changes are made. It currently presents the findings clearly but uses causal wording that is stronger than the design supports. The discussion section should also be tightened. Some claims are repeated, and the practical recommendations would be stronger if they were more directly linked to the study’s actual findings. The tables are useful, but the presentation should be improved. The authors should report exact p-values where appropriate, replace “p = 0.000” with “p < .001,” include effect sizes, and ensure that all table labels are clear and consistently formatted. Formula formatting should also be corrected, particularly in the regression section.

I’m also not convinced the Literature Review works as is, and I can agree that most of the statistical elements need just refining to be clearer to the reader, but the review is in my opinion too generic.

Presently, you defer to what appears to be the most common papers I’ve seen around many submissions, and often I believe to be the ones that AI itself has indexed and recommending, based on the years and clustering. Several DOIs do not work, for example the paper by Kim. What I found less helpful in reading your review is that it is in particular is lacking is a grounding in the literature landscape of Chinese HE/FE/student/university engagement with AI, which seems directly relevant to your Chinese vocational/FE angle of discussion.

Consistent with our journal policy, I do not recommend specific papers or named authors. As a researcher in the field, however, I know there’s a lot of really interesting stuff coming out of Chinese HE on AI at the moment, and I would have thought it would be logical to incorporate a dedicated subsection to move from the generic AI landscape stuff you write about, to next Chinese HE research on students/faculty and AI, and then to AI on vocational learning within the context of China, in a structured way? Otherwise the reader immediately thinks “why is this so generic” if they are read on research in these China specific areas.

What seems very popular at the moment is far ranging, but papers of late include topics as interesting as Chinese student perspectives on AI and guanxi in HE, ChatGPT in learning design for Chinese learners, utilising generative AI for Chinese student IELTS and EAL learning, Chinese learners and TAM, AI socio-emotional learning amongst Chinese students, determining factors of Chinese faculty, teachers and educators readiness to adopt generative AI, TPACK and AI in Chinese universities, Chinese student academic integrity and AI. Any/all/none of these could be considered, or an entirely specific vocational AI strand, though I am less sure of the areas of current publishing, you may want to expand further.

These are topics that may offer some direction to structure a subsection about Chinese learners more specifically around, as without this it looks like you have not reviewed the landscape, rather reviewed the topic in more general terms. I might therefore take a few days to add in a few paragraphs on the landscape of Gen AI research in China specifically, expanding in more detail, perhaps with the structure of general>HE China>HE vocational.]

Response 7: We sincerely thank the reviewer for the detailed suggestions on language refinement, standardization of statistical reporting and localization of the literature review. We fully recognize the existing shortcomings in these aspects and have carried out systematic revisions accordingly.

First, regarding language polishing and formatting standardization. We have gone through the entire manuscript and thoroughly corrected recurring grammatical errors, subject-verb disagreement, punctuation mistakes, redundant spaces and duplicated words. In addition, we have invited peers with rich experience in academic English writing to polish the text, so as to ensure fluent and standardized expression.

Second, in terms of the formatting of statistical results. All instances of p = 0.000 in tables and the main text have been uniformly revised to p < 0.001. We have also fixed formatting errors in the formulas of regression equations to guarantee that mathematical symbols and corresponding variables are clear and legible.

Third, the structural optimization of the literature review. As pointed out, the original literature review was too generalized and lacked a systematic overview of AI research within China’s higher education and vocational education sectors. We fully accept this comment and have added content focusing on the local context of China in Section 2, which reads as follows:

Against the backdrop of China’s higher vocational education, the rapid advancement of generative artificial intelligence has drawn growing academic attention to its applications in vocational colleges. Researchers have begun to systematically explore the current status and influencing factors of vocational college students’ AI literacy. Existing studies reveal that although Chinese vocational students generally hold positive attitudes towards generative AI, their practical capabilities vary greatly. In particular, they show deficiencies in key aspects such as prompt design, result evaluation and ethical judgment (Song et al., 2026). Some studies also indicate that vocational students tend to use generative AI in a superficial manner, regarding it merely as a tool to finish specific tasks while lacking systematic and critical awareness of its use (Zou et al., 2025). Overall, research on AI literacy in China’s vocational education is developing rapidly, yet there are no authoritative standard measurement tools for assessing generative AI usage behaviors. Therefore, it is urgent to establish analytical frameworks and measurement instruments tailored to the local settings of China’s vocational and higher education, so as to meet the new requirements for talent cultivation amid digital transformation.

Fourth, the rectification of invalid DOIs. We have noticed that the DOIs of several references including the paper by Kim et al. were inaccessible. We have checked the DOI links of all references one by one, updated the invalid links and supplemented missing publication details.

We hope these comprehensive revisions fully address your concerns and bring the manuscript up to the journal’s publication standards in terms of language use, statistical norms and the quality of the literature review. This new section can be found on page 15, Table 4; page 18, starting from paragraph 2, line 9-10;page 11, starting from paragraph 2, line 478-479;page 4, starting from paragraph 2, line 154-170.The revised parts are marked in red in the modified manuscript.

Comments 8: [Conclusion and Recommendation

Overall, this is a promising manuscript with a relevant topic, a meaningful vocational-education focus, and a potentially useful measurement framework. Its main contribution lies in showing that the quality and context of GenAI use may matter more than simple frequency of use. This is an important message for educators and institutions seeking to integrate GenAI into vocational learning environments.

However, the paper requires major revision. The authors should substantially revise the causal framing, clarify the sample reporting inconsistency, provide fuller scale validation evidence, treat self-reported learning outcomes more cautiously, and strengthen the statistical interpretation. With these improvements, the manuscript could make a valuable contribution to research on GenAI use in vocational education.]

Response 8: We sincerely appreciate the reviewer’s full recognition and encouragement regarding our work. Thank you for acknowledging the core findings of this study. The development of the Generative AI Usage Behavior Scale and the Vocational College Students’ Perceived Learning Outcomes Scale, as well as our focus on vocational education — an area that has been relatively understudied in existing literature — are recognized for their clear practical implications and solid theoretical foundations. These positive remarks confirm that this research is built upon a promising and valuable basis.

Meanwhile, we fully take on board the key concerns raised, which cover the causal framing, inconsistent sample descriptions, insufficient evidence for scale validation, the need for a more prudent interpretation of self-reported learning outcomes, and greater clarity in explaining statistical results. We have taken all comments seriously and conducted comprehensive and substantial revisions to address every issue. Furthermore, we have polished the language and standardized the formatting throughout the manuscript, correcting grammatical errors, formula layouts and table presentations.

We have strived to bring the revised paper in line with the journal’s publication standards in terms of theoretical construction, methodological rigor and the clarity of statistical interpretation. All revisions are presented in the updated manuscript for your review. We welcome any further feedback. Once again, thank you for your time and professional guidance that have helped improve the quality of this paper.

Comments 1: [Thank you for the opportunity to review your manuscript "Research on the Impact of Generative Artificial Intelligence Usage Behavior on the Learning Outcomes of Higher Vocational Students". The article deals with an important topic, especially in light of the rapid expansion of the use of Generative Artificial Intelligence (GenAI) tools in vocational education and training settings. The study focuses on a population that does not receive sufficient attention in the research literature, students in vocational training institutions, and thus adds an important layer to the existing knowledge. In addition, the development of two dedicated measurement tools is an important attempt to expand the toolbox of emerging research in the field. The article is based on a large and diverse sample, presents a well-organized research procedure, and produces interesting findings with potential for research and application contributions. However, there are several essential aspects that require methodological improvement and strengthening before publishing the article.]

Response 1: We sincerely thank the reviewer for the in-depth comments on the significance and features of this study. We are greatly encouraged by your recognition of our three core strengths. Firstly, this research addresses the important topic of the rapid adoption of generative AI tools in vocational education and training. Secondly, focusing on vocational college students, a group that has not received adequate attention in current literature, adds valuable insights to the existing body of knowledge. Thirdly, the development of two dedicated measurement scales is regarded as a useful supplement to the toolset for emerging research in this field. These positive evaluations demonstrate that our research is built on a promising and meaningful foundation.

Meanwhile, we fully acknowledge the key methodological issues that need to be revised and improved. We have taken all your comments seriously and completed comprehensive and substantial revisions to the manuscript. We welcome any further feedback. Thank you again for your time and professional guidance, which have greatly helped enhance the quality of this paper.

Comments 2: [The abstract is clearly written and presents the gist of the article. The literature review is comprehensive, but mostly descriptive. It is advisable to add a discussion of contradictory findings in previous studies and a clearer justification for developing the model in the review, this will strengthen the research justification.]

Response 2: We sincerely appreciate the reviewer's compliments on the clarity of the abstract and the comprehensiveness of the literature review. We fully agree that the original literature review was largely descriptive, lacked in-depth discussion on inconsistent findings in prior studies, and failed to elaborate clearly on the theoretical basis for model development. We have made substantial revisions to Section 2 Literature Review by incorporating findings from studies with both positive and negative perspectives. The detailed revisions are as follows.

First, we added a systematic discussion on conflicting results. A new paragraph is included in Section 2.3 to sort out contradictory conclusions regarding the relationship between generative AI use and learning outcomes in existing literature:

Nevertheless, divergent views also exist in relevant research. Some scholars raise concerns about academic integrity, over-reliance on AI, as well as data privacy and security issues amid students' use of generative AI (Maxwell et al., 2025). Meanwhile, the application of generative AI may lead to cognitive offloading and mental laziness, and impair students' abilities of independent problem-solving and critical thinking (Li et al., 2026; Choudhuri et al., 2026). For instance, when generative AI is adopted for writing tasks, it improves superficial fluency yet weakens students' understanding of in-depth text structures and independent reasoning skills (Rahman et al., 2025). Furthermore, generative AI does not exert a uniformly positive effect on learning motivation. Such tools usually fail to remind users of uncertain answers, which may cause students to overestimate their own capabilities, adopt a passive learning attitude, and gradually lose intellectual curiosity and critical thinking, thereby undermining their intrinsic motivation (Abdelghani et al., 2023). For students with low self-efficacy, completing complex tasks with generative AI without independent thinking and knowledge organization may bring about favorable short-term learning performance. However, their academic performance will decline markedly without AI support and their long-term knowledge transfer ability will be impaired. This indicates that the use of AI may give rise to adaptive dependence (Li et al., 2025).

Second, we transformed the literature review from a descriptive style to a critical one. We revised the simple enumeration structure such as "Researcher A proposed... Researcher B found..." into a critical narrative framed as "Scholars holding positive views argue that... However, other studies present conflicting findings...", which enhances the analytical depth of this section.

We hope these revisions turn the literature review from descriptive to analytical, better identify research gaps and consolidate the theoretical foundation of this study. All relevant updates have been incorporated into Section 2 of the revised manuscript. This new section can be found on page 5, starting from paragraph 2, line 222-277.The revised parts are marked in red in the modified manuscript.

Comments 3: [Although the article presents an interesting empirical contribution, the study does not rely on an established theoretical framework such as: TAM, UTAUT, SRL, etc. It is recommended to incorporate a clear theoretical framework that explains why different patterns of use of GenAI are expected to affect learning outcomes and how the study variables are related to each other.]

Response 3: We sincerely thank the reviewer for pointing out this important theoretical deficiency. We fully agree that the original manuscript lacked a clear elaboration of the theoretical framework and failed to explain why different patterns of usage behavior were expected to exert differentiated impacts on learning outcomes. Accordingly, we have integrated classic theoretical frameworks in Section 3.2.1 Development of the Generative AI Usage Behavior Scale to strengthen the theoretical foundation of this study. The detailed revisions are presented as follows:

The Technology Acceptance Model (TAM) and the Unified Theory of Acceptance and Use of Technology (UTAUT) are classic theoretical frameworks for explaining users’ adoption and usage of information systems, which have been widely validated and applied in the field of educational technology (Daruwala, 2026). This study systematically analyzes vocational college students’ generative AI usage behaviors by combining Davis’s (1989) Technology Acceptance Model (TAM), Venkatesh et al.’s (2003) Unified Theory of Acceptance and Use of Technology (UTAUT), and existing AI literacy assessment frameworks. An evaluation framework for generative AI usage behavior is thus proposed, consisting of three dimensions: usage frequency, usage habits and usage contexts.

The framework covers the above three core dimensions. For the dimension of usage frequency, guided by the core logic of TAM that perceived usefulness drives the intensity of usage behavior and drawing on measurement approaches of existing scales, we incorporate two indicators: behavioral dependence and task-oriented usage frequency. This dimension mainly assesses how frequently students use generative AI and their degree of reliance on it in study and daily life.

The dimension of usage habits is primarily built upon the procedural competencies defined in AI literacy frameworks. It focuses on evaluating users’ behaviors during dynamic interaction with generative AI, including interaction optimization, information screening and critical adoption, so as to measure vocational college students’ practical ability to use generative AI efficiently and critically.

For the dimension of usage contexts, we adopt the construct of facilitating conditions from UTAUT as the main theoretical basis. In UTAUT, facilitating conditions are designed to measure the availability of environmental resources, technical support and system compatibility, and essentially serve as a core variable for capturing contextual differences. Considering that practical teaching runs through the entire talent cultivation process of vocational college students, this dimension divides usage contexts into learning, daily life and personal development scenarios. It aims to examine differences in usage behavior under varying levels of facilitating conditions and assess the breadth of students’ generative AI application across diverse situations.

With the above theoretical integration, we can explain from the classic perspectives of TAM and UTAUT why usage frequency, usage habits and usage contexts show differentiated correlations with learning outcomes respectively. The complete theoretical framework has been fully presented in Section 3.2.1 of the revised manuscript. We hope these revisions clarify and solidify the theoretical underpinnings of this research and fully address your concerns regarding the theoretical framework. This new section can be found on page 8, starting from paragraph 4, line 368-393.The revised parts are marked in red in the modified manuscript.

Comments 4: [The development of the measurement tool is a central part of the article. The authors did conduct EFA, reliability tests, and basic validity tests. However, it is also very worthwhile to conduct a Confirmatory Factor Analysis (CFA). Without CFA, it is difficult to determine that the proposed theoretical structure is indeed valid.]

Response 4: We sincerely thank the reviewer for this valuable methodological suggestion. We fully agree that while exploratory factor analysis (EFA) can preliminarily reveal the dimensional structure of a scale, it cannot fully verify the validity of the theoretical structure. Confirmatory factor analysis (CFA) plays an irreplaceable role in verifying whether the three-factor structure proposed in this study fits the empirical data well. We have addressed this concern in two aspects in the revised manuscript.

First, we explicitly acknowledge the absence of CFA in the research limitations. We have added the following statement in Section 5 Research Limitations:

Although exploratory factor analysis, reliability tests and basic validity tests have been conducted for the Generative AI Usage Behavior Scale and the Vocational College Students’ Perceived Learning Outcomes Scale, confirmatory factor analysis has not been adopted to cross-verify the robustness of the theoretical structure. To a certain extent, this weakens the evidence for the construct validity of the scales.

Second, we specify the application of CFA and structural equation modeling in the future research directions. We further note in the Future Research section of the same chapter:

Given the complex relationships among the variables in this study, structural equation modeling can be adopted for in-depth mechanism analysis in follow-up research. This approach can better address multiple causal relationships and examine potential mediating or moderating effects.

We hope this candid discussion of limitations and targeted future research directions fully addresses your concerns. All relevant revisions are included in Section 5 of the revised manuscript. This new section can be found on page 22, starting from paragraph 2, line 179-239.The revised parts are marked in red in the modified manuscript.

Comments 5: [The article presents only four research questions. However, given an extensive literature review and a large sample, it would have been expected to present 4-5 explicit research hypotheses. For example: Intelligent use of GenAI will positively affect learning outcomes. Developing hypotheses will strengthen the research structure and anchor the findings in existing literature.]

Response 5: We sincerely thank the reviewer for this important suggestion. We fully agree that formulating explicit research hypotheses on the basis of an extensive literature review and large sample data can greatly improve the rigor of the research framework and better link the findings to existing studies. We have added a dedicated paragraph for research hypotheses in Section 3.2.2 Development of the Vocational College Students’ Learning Outcomes Scale, which reads as follows:

Based on previous studies on generative AI usage behavior and learning outcomes, we put forward the research hypotheses of this study.

H1: Usage frequency has a significant positive predictive effect on learning outcomes.

H2: Usage habits have a significant positive predictive effect on learning outcomes.

H3: Usage contexts have a significant positive predictive effect on learning outcomes.

H4: Among all dimensions of generative AI usage behavior, usage habits exert the strongest positive predictive effect on learning outcomes.

The above hypotheses are presented in Section 3.2.2 of the revised manuscript. We believe this addition clarifies the research structure and provides clear theoretical guidance for subsequent empirical tests. This new section can be found on page 10, starting from paragraph 2, line 439-447.The revised parts are marked in red in the modified manuscript.

Comments 6: [Regarding the statistical analysis. Very large sample, extensive reliability and validity tests, very high Cronbach's Alpha, exploratory factor analysis (EFA), and more. It is recommended to consider using SEM (Structural Equation Modeling) to examine the model more comprehensively. Also, despite the sample size, structural equation analysis (SEM) was not performed. The results are clear and well presented, but the regression results explain only about 20% of the variance in learning outcomes (R² = 0.204), with many significant variables that were not measured. Therefore, it is recommended to discuss in more detail additional variables that may affect learning outcomes, such as: motivation, engagement in learning, self-efficacy, etc.]

Response 6: We sincerely appreciate the reviewer's recognition of our sample size, reliability and validity tests, and statistical analysis. We fully acknowledge that the explanatory power of the current regression model is limited, and structural equation modeling (SEM) was not adopted for comprehensive mechanism analysis, which constitutes a methodological limitation of this study. Accordingly, we have supplemented prospects for the application of SEM in Section 5 Research Limitations and Future Research Directions of the revised manuscript:

Secondly, in terms of model construction and variable control, the regression model adopted in this study is relatively simplistic and fails to incorporate other factors that may affect learning outcomes. Future research should thoroughly discuss and introduce key control variables such as prior academic performance, socioeconomic status, institutional background, major, grade level, teacher support, learning motivation, learning engagement and self-efficacy. This will help explore the relationship between generative AI use and learning outcomes in a more rigorous manner. In addition, given the intricate relationships among the variables in this study, structural equation modeling can be employed for in-depth mechanism analysis. It is conducive to addressing complex causal relationships and examining potential mediating and moderating effects.

We hope this frank elaboration of limitations and specific future research directions fully addresses your concerns. All relevant revisions have been incorporated into Section 5 of the revised manuscript. This new section can be found on page 22, starting from paragraph 2, line 199-208. The revised parts are marked in red in the modified manuscript.

Comments 7: [The findings are presented in a clear and orderly manner. It seems that the most interesting finding in the article is: high frequency of use of GenAI was found to be negatively related to learning outcomes, while usage habits and usage contexts were found to be positively related. This is a very important finding, but the explanation for this finding is relatively brief. Therefore, it is recommended to develop a more in-depth discussion of this finding.]

Response 7: We sincerely thank the reviewer for commending the clear presentation of our research findings. You also pointed out that the negative correlation between usage frequency and learning outcomes, as well as the positive correlations of usage habits and usage contexts with learning outcomes, are the most intriguing results of this paper. We fully agree that the interpretation of this core finding was overly brief and requires further elaboration. We have expanded the discussion on this result in Sections 4.3.2 and 5 of the revised manuscript, with the supplementary content as follows.

In the regression results section (Section 4.3.2), we added:

Nevertheless, this negative coefficient does not indicate that usage frequency itself exerts adverse impacts on learning outcomes. Among vocational college students, high usage frequency is often accompanied by aimless use without clear learning objectives, over-reliance on AI-generated content, and a lack of active processing and in-depth thinking. These poor learning habits fail to translate into positive academic performance, thus resulting in a negative statistical correlation.

In the discussion section (Section 5), we supplemented:

The predictive effect of generative AI usage frequency on vocational college students’ learning outcomes does not follow the assumption that more frequent use leads to better learning performance. Unstructured use or excessive reliance on generative AI may help students finish learning tasks efficiently, yet deprive them of the ability to think independently and build personal knowledge systems, which in turn hinders the improvement of learning outcomes.

In the recommendations part of the discussion, we put forward relevant suggestions based on this finding:

For students, it is essential to move from merely being able to use AI tools to using them proficiently. As direct users of generative AI, students should continuously improve their digital literacy. They need to recognize that simply increasing AI usage frequency may impair learning outcomes and avoid cognitive offloading caused by over-reliance on generative AI for information retrieval and decision-making (Gerlich, 2025).

We hope these in-depth discussions fully address your concerns and offer a more thorough interpretation of the theoretical significance and practical implications of the core findings. All revisions have been included in Sections 4.3.2 and 5 of the revised manuscript. This new section can be found on page 18, starting from paragraph 3, line 16-38;page 21, starting from paragraph 2, line 147-157. The revised parts are marked in red in the modified manuscript.

Comments 8: [Some of the limitations require a broader consideration: use of the questionnaire in self-report only, possibility of social desirability bias, limitations of generalization outside of China, underrepresentation of students in advanced years.]

Response 8: We sincerely thank the reviewer for this valuable suggestion. We fully recognize that this study has multiple limitations regarding measurement instruments, sample representativeness and external validity, which require more comprehensive discussion in the future research section. We have systematically supplemented relevant content in Section 5 Research Limitations and Future Research Directions of the revised manuscript, as detailed below.

First, with regard to measurement instruments. As technology evolves rapidly, generative AI tools keep expanding in categories, functions and application scenarios, which further enriches the connotation of AI usage behavior. In addition, this study mainly adopts self-report questionnaires to measure learning outcomes. This method is inherently subjective, and the results may largely reflect students’ self-confidence, motivation and social desirability rather than their actual knowledge mastery and skill improvement. Future research shall update the components of generative AI usage behavior in line with the latest developments and develop more systematic, scientific and context-adaptive assessment tools. Meanwhile, to measure learning outcomes more accurately, subsequent studies should adopt multi-source data and objective indicators. On the basis of student self-reports, researchers may further incorporate course scores, practical training results, teacher evaluations, internship assessments, professional skill tests and certification outcomes to effectively reduce biases arising from single-source self-reported data.

Second, regarding sample selection and representativeness. There is a notable imbalance in sample size across grades, with an overrepresentation of junior students and insufficient participants from senior grades. This may affect the stability and generalizability of the research findings. Furthermore, all samples are recruited from vocational colleges in China, so whether the conclusions can be applied to educational contexts outside China remains to be verified. Future research shall select participants from vocational colleges of different regions, tiers and institutional types on a larger scale. Particular attention should be paid to balancing the sample proportion across grades and recruiting more senior students, so as to improve sample representativeness and the external validity of conclusions. Cross-cultural comparative studies are also encouraged to examine the robustness of the findings across different contexts.

We hope these supplements make the discussion on research limitations and future directions more comprehensive and transparent. All revisions have been included in Section 5 of the revised manuscript. This new section can be found on page 22, starting from paragraph 1, line 179-208. The revised parts are marked in red in the modified manuscript.

Comments 9: [The article does not include a visual illustration (figures and graphic models) of the research model. It is recommended to add a conceptual model and a research diagram. Such an addition will improve the clarity of the article.]

Response 9: We sincerely thank the reviewer for this constructive suggestion. We fully agree that adding visual diagrams for the research design can greatly enhance the clarity and readability of the manuscript, enabling readers to intuitively understand the research framework. We have inserted relevant diagrams in Section 3.1 Research Design of the revised manuscript. This new section can be found on page 7, Figure 1. Research Design Flowchart.

Comments 10: [In general, the article is written in good English, but several linguistic and stylistic problems are evident: grammatical errors, inconsistent use of tenses, long and cumbersome sentences. For example: This This, Test results was, GenAI behavior are defined. Before publication, it is recommended to have a professional language editing by a linguistic editor. The bibliography is extensive and up-to-date, but it contains few sources from the leading literature on the subject of GenAI specifically in higher education. It is suggested to add a number of more studies directly related to higher education in order to strengthen the connection to the literature:

Carmi, G. (2025). Learning with Generative AI: An Empirical Study of Students in Higher Education. Education Sciences, 15(12), 1696. https://doi.org/10.3390/educsci15121696

Dabis, A., & Csáki, C. (2024). AI and ethics: Investigating the first policy responses of higher education institutions to the challenge of generative AI. Humanities and Social Sciences Communications, 11(1), 1–13. https://doi.org/10.1057/s41599-024-03526-z

Maxwell, D., Oyarzun, B., Kim, S., & Bong, J. Y. (2025). Generative AI in higher education: Demographic differences in student perceived readiness, benefits, and challenges. TechTrends, 69(6), 1248-1259. https://doi.org/10.1007/s11528-025-01109-6

Sergeeva, O. V., Zheltukhina, M. R., Shoustikova, T., Tukhvatullina, L. R., Dobrokhotov, D. A., & Kondrashev, S. V. (2025). Understanding higher education students’ adoption of generative AI technologies: An empirical investigation using UTAUT2. Contemporary Educational Technology, 17(2), ep571. https://doi.org/10.30935/cedtech/16039]

Response 10: We sincerely appreciate the reviewer’s valuable comments on the writing style, references and academic standards. We have conducted comprehensive revisions accordingly.

First, we have corrected all identified grammatical errors, tense inconsistencies and problematic complex sentences one by one. In addition, professionals with extensive experience in academic English writing have polished the full manuscript to ensure clear, professional expressions that meet the journal’s publication requirements.

Second, we have fully adopted your suggestions regarding references. We have incorporated the four key papers you recommended, namely Carmi (2025), Dabis & Csáki (2024), Maxwell et al. (2025) and Sergeeva et al. (2025), which focus on the application of generative AI in higher education. These works are cited and discussed at appropriate places in the main text. This update greatly strengthens the links between the literature review and cutting-edge research in higher education, and further consolidates the theoretical foundation of this study.

The supplemented contents are as follows:

In Section 2.3: Among relevant factors, students’ positive attitudes, satisfaction and experience in using generative AI all play a facilitating role in this process (Carmi, 2025). Meanwhile, users’ usage habits, performance expectancy, social influence and personal innovativeness are critical drivers for the adoption of generative AI (Sergeeva et al., 2025; Lin & Jiang, 2025).

Also in Section 2.3: Some studies have pointed out that students have concerns over academic integrity, excessive reliance on AI, data privacy and information security when using generative AI (Maxwell et al., 2025).

In Section 5: Meanwhile, to improve transparency, teachers should clearly specify the scope of AI use, permitted and prohibited practices, and relevant academic integrity requirements in course syllabi (Dabis & Csáki, 2024).

We believe that the targeted revisions on language and references have substantially improved the overall quality and academic rigor of the manuscript. Thank you again for your time and professional guidance. This new section can be found on page 5, starting from paragraph 2, line 232-236;page 6, starting from paragraph 1, line 246-248;page 21, starting from paragraph 1, line 143-146.The revised parts are marked in red in the modified manuscript.

Comments 11: [In conclusion, this is an article with high relevance, a broad data base, and a certain applied contribution. However, there are a number of theoretical and methodological weaknesses that limit the strength of its scientific contribution and require corrections before acceptance for publication.]

Response 11: We sincerely thank the reviewer for the high praise on the practical relevance, data foundation and application value of this study. We fully acknowledge that the manuscript still has several deficiencies in theoretical construction and methodology, which limit its academic contributions. We have carried out comprehensive and systematic revisions to address all the concerns raised. We have strived to bring the revised paper up to the journal’s publication standards in terms of theoretical depth, methodological rigor and academic contribution. All revisions are marked in the updated version. We welcome any further feedback. Thank you again for your time and professional guidance that have helped improve this paper.

Comments 1: [The authors have made several revisions and have responded carefully to the reviewer’s comments. The revised manuscript appears to be more transparent regarding its limitations and the exploratory nature of the study. The topic remains timely and relevant, particularly given the increasing use of generative AI in vocational education. The focus on GenAI usage behavior and students’ learning outcomes has practical and academic interest.]

Response 1: We sincerely appreciate the reviewer’s positive feedback on our revisions. We are particularly gratified that the revised manuscript has been recognized for its transparent discussion of research limitations and clear positioning as an exploratory study, and that the research topic closely aligns with practical demands regarding the application of generative AI in higher vocational education. These encouraging remarks affirm both the practical and academic value of this study.We thank the reviewer for devoting time and providing professional guidance to improve the quality of this paper, and we look forward to further support from the journal for this research. The revised parts are marked in red in the modified manuscript.

Comments 2: [However, several major methodological concerns remain insufficiently resolved. The dependent variable still relies on self-reported perceived learning outcomes rather than objective indicators such as grades, teacher evaluations, practical skill assessments, internship performance, or standardized tests. The authors acknowledge this limitation, but the issue remains central to the validity of the study.]

Response 2: We sincerely thank the reviewer for once again pointing out this core methodological issue. We fully acknowledge that the reliance of the dependent variable on self-reported perceived learning outcomes instead of objective indicators constitutes a critical limitation to the validity of this study. We address this concern from two perspectives as follows.

First, we elaborate on the practical obstacles to accessing objective indicators. Objective data such as students’ course grades, teacher evaluations, and skill assessment results are classified as students’ private personal information in Chinese higher vocational colleges, which researchers cannot obtain through anonymous questionnaire surveys. Requesting real-name information or retrieving academic records from educational institutions would contradict the principles of anonymous research and requirements for student privacy protection. This practical constraint is the primary reason why this study adopts self-reported perceived learning outcomes, and it represents a common challenge confronting large-scale questionnaire research within China’s vocational education sector.

Second, we explain how we have addressed this limitation within the manuscript. In the research design section, we explicitly clarify that this study measures "perceived learning outcomes" rather than objective learning outcomes. We consistently adopt this terminology throughout the abstract, main text, discussion, and conclusion to prevent conceptual confusion and overgeneralization of the findings. Furthermore, we have explicitly acknowledged this shortcoming in the section titled "Limitations and Future Research Directions", where we detail specific avenues for future research to incorporate multi-source objective data, including course grades, practical training scores, teacher assessments, internship evaluations, and professional skill examinations or certification records.

We hope the above clarification fully addresses the reviewer’s concerns. We fully recognize that this limitation weakens the robustness of our conclusions. Nevertheless, given the types of data accessible under the current practical constraints, this exploratory study on the effects of generative AI usage in vocational education still delivers exploratory value and informative implications for relevant scholarship. We once again appreciate the reviewer’s comments, which have greatly improved the methodological transparency and rigor of this manuscript. The revised parts are marked in red in the modified manuscript.

Comments 9: [The discussion of academic integrity has been improved, but it still seems mostly placed in future research rather than integrated into the core discussion of GenAI use in assignments, thesis writing, coding, and assessment validity.]

Response 9: We sincerely thank the reviewer for pointing out this critical issue. We fully agree that discussions on academic integrity should not be confined merely to the future research section; instead, they ought to be organically integrated into the core analytical sections of the manuscript. We have made substantial revisions to relevant discussions in the revised version, closely linking the theme of academic integrity to core GenAI application scenarios including coursework completion, thesis writing, programming, and assessment validity. This topic now runs through several key chapters such as the literature review, discussion, and practical implications, rather than serving as a peripheral remark limited to future research directions. Detailed revisions are outlined as follows:

In the literature review (Section 2.3), when sorting out inconsistent findings from existing studies, we supplemented targeted discussions on concrete academic integrity risks with the following text:

“Nevertheless, contradictory arguments have been raised in prior scholarship. Several studies identify prevalent concerns regarding academic integrity, excessive AI reliance, data privacy and security among students using generative AI (Maxwell et al., 2025). Specifically, when students directly submit AI-generated content as their own work without proper attribution for assignments, thesis writing, coding and other tasks, such conduct may amount to plagiarism or academic misconduct. Yet many students lack clear awareness of the ethical boundaries of such practices. In the absence of institutional guidelines governing AI utilization, this ambiguous grey area of AI application tends to proliferate. It not only undermines the diagnostic capacity of traditional assessments to reflect students’ true competencies, but also intensifies threats to the academic integrity framework.”

In the discussion chapter (Section 5), while analyzing how AI usage habits shape learning outcomes, we further established an intrinsic linkage between usage habits and academic integrity by adding the following passage:

“In fact, there exists an intimate correlation between AI usage habits and academic integrity. Students with sound generative AI usage habits are more inclined to question, verify, and critically examine AI-generated outputs, which enables them to maintain heightened awareness and self-discipline in compliance with academic norms when citing or adopting AI-produced content. In contrast, students lacking healthy usage habits are prone to uncritically adopt outputs generated by AI. Such habitual over-reliance may inadvertently lead to plagiarism and other forms of academic misconduct. In the absence of clear institutional regulations governing AI utilization, this ambiguous grey area of AI application tends to become prevalent. It not only intensifies threats to the academic integrity framework but also diminishes the diagnostic value of assessments for evaluating students’ genuine competencies. In other words, academic integrity and metacognitive habits are two sides of the same coin. Sound generative AI usage habits encompass not only technical competencies such as optimized human-AI interaction and information screening, but also vigilance over the legitimate boundaries of AI content as well as sound ethical judgment.”

In the practical implications subsection (Section 5), we put forward targeted measures tied to this theme with the addition of the following sentence:

“Meanwhile, to enhance transparency, instructors should explicitly articulate permitted and prohibited AI usage scenarios alongside corresponding academic integrity requirements within course syllabi (Dabis & Csáki, 2024).”

All the above revisions are fully incorporated into Sections 2.3 and 5 of the revised manuscript. We hope this integration of academic integrity considerations throughout the literature review and core discussion sections adequately addresses the reviewer’s concern. The revised parts are marked in red in the modified manuscript.

Comments 10: [Overall, the manuscript has improved, but several central weaknesses remain. Many concerns are acknowledged as limitations rather than being addressed through additional analyses or stronger methodological corrections.]

Response 10: We sincerely appreciate the reviewer’s recognition of the improvements made in the revised manuscript, and we fully understand the core concern raised: several issues have only been addressed by acknowledging limitations rather than through substantive corrections, which indeed weakens the methodological rigor of this paper. To tackle this, we have comprehensively revised the manuscript via substantive improvements including supplementary data collection, model optimization, and the addition of control variables.

We respectfully ask the reviewer to recognize the inherent constraints of this exploratory study imposed by practical conditions. The value of this research lies in establishing a referential foundation for developing measurement scales and exploring preliminary correlations regarding generative AI usage behaviors in vocational education. Its primary contribution is to offer theoretical hypotheses, measurement instruments, and empirical insights for follow-up research, rather than drawing definitive causal conclusions. We anticipate that future studies can adopt more rigorous research designs to further verify and extend our findings. Should the reviewer deem the current revisions insufficient to meet the journal’s publication criteria, we are ready to receive specific feedback and make every effort to further refine the manuscript. The revised parts are marked in red in the modified manuscript.

Comments 1: [The authors have improved the conceptual framing by clarifying that the study focuses on students’ self-reported or perceived learning outcomes rather than objective learning outcomes. This distinction is important and should be maintained consistently throughout the manuscript, including the title, abstract, discussion, and conclusion.]

Response 1: We sincerely appreciate the reviewer’s reminder regarding this vital conceptual distinction. We fully acknowledge that clearly anchoring this study to perceived learning outcomes rather than generic learning outcomes constitutes a core conceptual boundary of our research. This differentiation must be consistently maintained throughout the manuscript to prevent readers from overinterpreting our empirical findings. Accordingly, we have systematically standardized relevant terminology across the entire paper, with detailed revisions outlined below:

First, we revised the abstract. All references to the dependent variable in the abstract, covering the research background, methodology, analytical results and concluding remarks, have been uniformly revised to "perceived learning outcomes".

Second, we standardized terminology across all main chapters. We meticulously reviewed and revised more than 150 instances of the phrase "learning outcomes" distributed across the Introduction, Literature Review, Methodology, Results, Discussion and Conclusion sections, replacing each with "perceived learning outcomes". Corresponding adjustments have also been made to the keywords.

Third, we updated captions and notes in all tables and the appendix. Table headers and explanatory notes for Tables 3 to 6, as well as descriptions of learning outcome items in the questionnaire appendix, have all been revised to uniformly read "perceived learning outcomes".

Fourth, we refined wording to eliminate ambiguous shorthand. Throughout the revision process, we took special care to avoid abbreviating "perceived learning outcomes" as simply "learning outcomes", completely separating the two terms to remove any potential ambiguity.

Following these comprehensive revisions, the terminology for the dependent variable is fully consistent throughout the manuscript, which fundamentally clarifies the scope of applicability of our research conclusions. We hope these thorough adjustments adequately address the reviewer’s concern about consistent conceptual wording.

The revised parts are marked in red in the modified manuscript.

Comments 2: [The additional data collection is a positive improvement. The larger sample strengthens the empirical basis of the study and partially addresses the previous concern regarding sample imbalance. However, the authors should carefully check the manuscript for consistency, as some sections still seem to describe grade and major imbalance as a major limitation.]

Response 2: We are truly grateful to the reviewer for recognizing our efforts in supplementary data collection. We fully agree that the substantial addition of junior-year student samples has substantially alleviated the original grade imbalance issue, which necessitates corresponding updates to the limitations section to reflect the updated sample composition. In light of this, we have implemented the following revisions in the manuscript:

First, we removed limitation statements that no longer align with the updated sample data. The original text in the limitations section stated that certain broad major categories (resources, environment and safety; public security and justice; public administration and services) contained very small sample sizes (n<5), which might weaken the statistical power of ANOVA tests and the reliability of cross-group comparisons, thereby limiting the generalizability of our findings. After supplementary sampling, each of these major groups now includes over 10 respondents, rendering the original description obsolete for the current dataset.

Second, all text addressing grade imbalance has been eliminated. We collected an additional 160 valid questionnaires from junior students, which considerably raised the proportion of junior participants and greatly balanced the grade distribution. Accordingly, we deleted all prior remarks regarding skewed grade sampling.

Third, we streamlined the discussion of sample representativeness within the future research directions. The revised version now reads: “All participants of this study were recruited from vocational colleges in China, and the generalizability of the findings to educational contexts outside China remains to be further verified.” This wording retains our cautious stance on external validity while eliminating inaccurate claims inconsistent with the updated sample structure.

All the above modifications are incorporated into Section 5 “Limitations and Future Research Directions” of the revised manuscript. We hope these adjustments ensure that our discussion of study limitations accurately matches the characteristics of the updated sample. The revised parts are marked in red in the modified manuscript.