Response to Reviewers

Dear Editors and Reviewers,

We would like to express our sincere gratitude to the Academic Editor and the Reviewer for their rigorous evaluation of our manuscript (PONE-D-26-08042) and their highly constructive feedback. We have carefully considered all the comments and have revised the manuscript accordingly. We believe that these revisions have substantially improved the theoretical foundation, methodological rigor, and clarity of our paper.

Below, we provide a point-by-point response to each comment. The original comments are presented in bold, followed by our responses and the corresponding modifications made in the manuscript.

Part I: Responses to Journal Requirements (Academic Editor)

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming.

●Response: Thank you for this reminder. We have carefully checked the revised submission package and ensured that the manuscript has been revised to comply with PLOS ONE’s style requirements, including the formatting of the manuscript and the naming of the uploaded files. We will submit the revised files using the journal’s required file labels and formats, including “Response to Reviewers,” “Revised Manuscript with Track Changes,” and “Manuscript.”

2. Please include your full ethics statement in the ‘Methods’ section of your manuscript file. In your statement, please include the full name of the IRB or ethics committee who approved or waived your study, as well as whether or not you obtained informed written or verbal consent. If consent was waived for your study, please include this information in your statement as well.

●Response: Thank you for the reminder. We have incorporated the full ethics statement into the 'Methods' section as requested. Specifically, the details regarding the ethical waiver from the Ethics Committee of Beihang University and the informed written consent procedures have been added to the end of Section 4.1 (Survey samples).

3. Thank you for stating the following financial disclosure:

This research was supported by the Fundamental Research Funds for the Central Universities, Key Teaching Reform Project of Beihang University, and Research and Development Fund Project for Postgraduate Education of Beihang University (JG2024030).

Please state what role the funders took in the study. If the funders had no role, please state: "The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript."

If this statement is not correct you must amend it as needed.

Please include this amended Role of Funder statement in your cover letter; we will change the online submission form on your behalf.

●Response: Thank you for your guidance. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. We have added this required statement to the "Funding" section at the end of the revised manuscript. Additionally, we have explicitly included this amended Role of Funder statement in the new cover letter accompanying this revised submission. Please kindly review it.

4. Please upload a copy of Figure 1, 2, to which you refer in your text on page 10, 15. If the figure is no longer to be included as part of the submission please remove all reference to it within the text.

●Response: Thank you for this reminder. We apologize for the confusion caused by the initial file naming. Following PLOS ONE’s guidelines, we have submitted the figures as separate files rather than embedding the images in the manuscript. The files previously labeled as "S1_Fig.tif" and "S2_Fig.tif" are indeed the "Figure 1" and "Figure 2" referred to in the text. To ensure clarity and consistency, we have now renamed these image files to Fig1.tif and Fig2.tif and re-uploaded them to the submission system as "Figure" items.

5. Please include a separate caption for each figure in your manuscript.

●Response: Thank you for this reminder. In accordance with the journal's formatting guidelines, a separate caption has been added for each figure in the revised manuscript. Specifically, we have moved the captions from the end of the manuscript directly into the main text, inserting them immediately after their first citation (in Section 3.4 and Section 5.4, respectively), and deleted the "(TIF)" text at the end of the captions. Please kindly review.

6. If the reviewer comments include a recommendation to cite specific previously published works, please review and evaluate these publications to determine whether they are relevant and should be cited. There is no requirement to cite these works unless the editor has indicated otherwise.

●Response: Thank you for your guidance. In our case, the reviewers did not recommend any specific publications for citation.

Part II: Responses to Reviewer #1

A. Responses to Specific Comments

1. The title is too long and overly descriptive, reducing clarity and impact. Phrase “Are there differences across dimensions of engagement?” appears exploratory rather than theory-driven.

●Response: Thank you very much for this helpful comment. We agree that the original title is somewhat lengthy and that the interrogative phrase at the end may make the study appear more exploratory than theoretically focused. To improve clarity and concision, we have revised the title. Revised title: “In-class AI usage and university students' attitudes: The mediating role of learning engagement”

2. Convenience sampling from a single research-intensive university (H University) in China severely limits generalizability. The sample (N=287) is relatively small for the complexity of the structural equation model tested.

●Response: We acknowledge the limitations regarding sampling and sample size. Using convenience sampling at a single institution indeed restricts the generalizability of our results. We have revised the "6.3 Research limitations and future directions" section to reflect this more accurately. Modification: In Section 6.3, we explicitly stated: "First, the participants in the research sample were limited to students from a single research-intensive university in China, and the sample size (N=287) was relatively small. This may limit the generalizability of the findings to students in other types of institutions or different cultural contexts."

3. The sample disproportionately represents science and engineering students (202 vs. 85 humanities/social sciences), yet no analysis examines whether findings differ by discipline. This is a significant omission given that AI usage patterns likely vary substantially across fields.

●Response: Thank you for your comment. Your insight is very reasonable—AI usage patterns may indeed differ among students from different academic disciplines. We conducted differential analyses on five core variables across disciplinary fields and found that three showed no significant differences, while two did show significant differences, while both with small effect sizes. This indicates that although the disciplinary differences are statistically significant, their practical impact is very small. Due to the unbalanced sample composition (overrepresentation of science and engineering students), we did not conduct direct group comparisons. (Added to Section 5.3)

4. No power analysis is reported to justify that N=287 is sufficient for the SEM with multiple mediators and Bayesian estimation.

●Response: Thank you for this important methodological point. We would like to clarify that traditional frequentist power analysis is not directly applicable to Bayesian estimation, because Bayesian inference does not rely on null hypothesis significance testing or the concept of statistical power in the Neyman–Pearson sense. Instead, the adequacy of sample size in Bayesian SEM is evaluated through MCMC convergence diagnostics and the precision of posterior estimates.

In our study, with N = 287 and 82 free parameters, the following evidence supports that the sample size is sufficient for stable estimation:

Convergence: All MCMC chains converged well, with PSR (Potential Scale Reduction Factor) ≤ 1.001, well below the recommended threshold of 1.05. Trace plots and autocorrelation plots showed no abnormal trends.

Effective Parameters: The pD (effective number of parameters) was 81.6, which is very close to the total number of free parameters (82), indicating that nearly all parameters were identified by the data rather than solely influenced by priors.

Estimation Precision: The 95% credible intervals for all parameters were reasonably narrow, providing sufficient precision for substantive interpretation.

Thus, while traditional power analysis was not performed, the reported Bayesian convergence diagnostics (added to Section 5.4) fully justify that N = 287 is adequate for the current model complexity.

5. All variables (CU, CE, BE, EE, AT) were measured using self-report questionnaires from the same respondents at a single time point. This introduces substantial risk of common method variance, which the authors do not adequately address. The Harman's single-factor test or more rigorous approaches (e.g., marker variable technique) are absent.

●Response: Thank you for this important methodological observation. In this study, Harman's single-factor test was used to examine common method bias. An unrotated principal component analysis on all measurement items revealed five factors with eigenvalues greater than 1.0, with the first factor accounting for 38.114% of the total variance, which is below the recommended threshold of 50% (Podsakoff et al., 2003). This indicates that common method bias is not a serious concern in this study and is unlikely to substantially distort the relationships among the variables. (Added to Section 4.3)

6. The study uses cross-sectional data but makes causal claims about mediation (e.g., "AI usage affected students' attitudes both directly and indirectly"). Without longitudinal or experimental design, the direction of causality is ambiguous. Reverse causation (positive attitudes → more AI usage) is equally plausible.

●Response: Thank you for this critical methodological observation. The reviewer is correct that cross-sectional data cannot support strong causal claims about mediation. The original wording has been revised throughout the manuscript to avoid causal language. The following actions have been taken:

1.Revision of causal language. All instances of causal claims (e.g., "affected," "influenced," "impacted") have been replaced with non-causal, associative language appropriate for cross-sectional data (e.g., "was associated with," "predicted", "was related to,"). The abstract, results, and discussion sections have all been revised accordingly.

2.Limitations statement. A clear limitation has been added to the Discussion section (Section 6.3) explicitly stating that the cross-sectional design precludes causal inference and that longitudinal or experimental designs are needed to establish temporal precedence and causality.

7. The Attitudes scale (3 items) was "compiled independently by the researchers" with no reported validation beyond this study. The authors should provide evidence of prior validation or at minimum report item wording and pilot testing results.

●Response: We appreciate the reviewer's vigilance. The three attitude items were reviewed by three subject matter experts. The complete wording of the three attitude items is provided in the Appendix (S2 File). In this study, the Cronbach's α for the Attitudes scale was 0.884. Exploratory factor analysis (EFA) extracted a single factor with an eigenvalue of 2.443, accounting for 81.431% of the total variance. The standardized factor loadings of the three items ranged from 0.890 to 0.920, with no evidence of double loading. Together, these indices consistently demonstrate that this newly developed Attitudes scale has good reliability and validity, indicating that the instrument is suitable for use in subsequent research (Added to Section 4.2). Nevertheless, we acknowledge the lack of prior validation as a limitation and have noted this in the Limitations section (Section 6.3) of the revised manuscript.

8. The In-Class AI Usage Behavior Scale asks about frequency ("Never" to "Very frequently") but does not distinguish types of AI use (e.g., generative AI for content creation vs. search-based AI for fact-checking). Different AI uses likely have different effects on engagement dimensions.

●Response: We sincerely thank the reviewer for pointing out this important nuance. We agree that treating AI usage primarily as a frequency variable without differentiating the specific types of AI tools (e.g., generative vs. search-based) is a limitation of our measurement instrument, as different AI functionalities may distinctively impact learning engagement. To address this, we have added this point to the research limitations (Section 6.3) as a critical direction for future studies.

9. Social desirability bias is unaddressed. Students may overreport "appropriate" AI usage or engagement levels.

●Response: Thank you for this helpful comment. We agree that social desirability bias is a potential concern in questionnaire-based research of this kind, and that students may have a tendency to overreport their AI usage and engagement when responding to related questions. In the present study, the survey was conducted anonymously, participation was voluntary, and no personally identifiable information was collected, which may help reduce respondents’ pressure to provide socially desirable answers. To address this concern more explicitly, we have added a brief statement in the limitations section (Section 6.3).

10. The paper invokes Self-Perception Theory (Bem, 1972), Cognitive Consistency Theory, and Affective Events Theory but does not integrate them into a coherent framework. How do these theories work together? Which predictions do they make about differential mediation across engagement dimensions?

●Response: Thank you for this insightful comment. We agree that the original manuscript did not sufficiently integrate the three theoretical perspectives. We have now added an integrated framework ("three-layer progressive consistency framework") of the three perspectives in the research model section (Section 3.4) to clarify how these theories work together and predict the mediating effects of different engagement dimensions based on these theories.

11. The authors claim Bayesian analysis is "more suitable for relatively small sample sizes" but provide no comparison with frequentist approaches or sensitivity analyses. The choice of priors is not specified, which is a fundamental omission in Bayesian work.

●Response: We thank the reviewer for these important points. We apologize for this omission. In the revised manuscript (Section 4.3), we have added a detailed Prior Specification subsection. Bayesian estimation employed Mplus default diffuse priors: factor loadings, path coefficients, and intercepts followed a normal distribution with a mean of 0 and a variance of 10¹⁰, while residual variances and covariances were assigned non-informative inverse-gamma priors. To examine the sensitivity of posterior inference to prior specifications, we re-estimated the model by replacing the priors for all structural path coefficients with weakly informative normal priors N(0,1), keeping all other specifications unchanged. The results showed that the Bayesian inference in this study is robust and not sensitive to the choice of priors. Following the reviewer's suggestion, we also estimated the same model using maximum likelihood (ML) estimation as a comparative approach (added to Section 5.4). Critically, the direction and significance of the core paths were entirely consistent with the Bayesian results. This demonstrates that the conclusions of this study are robust and consistent across two different estimation approaches (Bayesian and frequentist ML), providing strong evidence for the cross-methodological robustness of our findings.

12. The path from BE → AT (β = -0.074, CI [-0.205, 0.056]) is non-significant, yet the authors interpret this as "behavioral engagement operates through a mech

Attachments

Attachment

Submitted filename: Response to Reviewers.docx

Response to Reviewers

Dear Editors and Reviewers,

We would like to express our sincere gratitude to you for your rigorous evaluation of our manuscript (PONE-D-26-08042) and your highly constructive feedback. We have carefully considered all the comments and have revised the manuscript accordingly. We believe that these revisions have substantially improved the theoretical foundation, methodological rigor, and clarity of our paper.

Additionally, before proceeding to the point-by-point responses, we would like to briefly note that a “Declaration on Data Reuse" has been included at the very end of this document to clarify the relationship between this study and a recently published related work.

Below, we provide the point-by-point response to each comment. The original comments are followed by our responses and the corresponding modifications, which are highlighted in the revised manuscript.

Reviewer #1:

Comment 1. The authors have fully met the requirements of the reviewer comments. The manuscript is ready for acceptance from a review-response perspective.

●Response:

We sincerely thank the reviewer for the positive and encouraging assessment of our revisions. We are very glad to hear that our point-by-point responses have fully addressed all the concerns raised. We truly appreciate the reviewer’s time and constructive suggestions throughout this process, which have significantly improved the quality of our manuscript.

Reviewer #3:

Comment 1. The mediators measure perceived benefit not engagement. This is my key point. The instruments do not measure engagement. They measure agreement that AI enhances engagement. The authors say CE is "perception of whether in-class AI usage enhances cognitive engagement." Same for BE and EE. So these are evaluative. AI-referential. They are close to perceived usefulness. The authors invoke TAM and perceived usefulness themselves on the mediator→attitude paths. All same-method and highly overlapping constructs "I think AI boosts my thinking" predicting "I like AI" ishould be described more carefully.

Suggestion: relabel the constructs accurately throughout. They are perceived engagement benefits. Not engagement. and discuss how much indirect effect might be construct overlap rather than mechanism.

●Response:

We sincerely thank the reviewer for this crucial insight. We completely agree that our instruments measure the students' evaluative agreement that AI enhances their engagement, rather than objective engagement metrics. To accurately reflect this, we have systematically relabeled the constructs throughout the entire manuscript—including the title, abstract, hypotheses, instruments, and discussion—to "perceived cognitive engagement benefits (PCEB)", "perceived behavioral engagement benefits (PBEB)", and "perceived emotional engagement benefits (PEEB)".

Furthermore, we have followed your suggestion to explicitly discuss the issue of construct overlap. We added a new paragraph to the limitations section to address how the indirect effects might be partially attributable to this overlap or a generalized positive appraisal rather than a purely sequential mechanism.

●Modification:

(1) To ensure conceptual consistency from the very beginning, we have modified the title of the manuscript to replace the general term of learning engagement with our specific measured variable.

Modified title: In-class AI usage and university students' attitudes: The mediating role of perceived learning engagement benefits

(2) In Section 6.3 (Research limitations and future directions), we have added a new fifth paragraph between the fourth limitation and the subsequent discussion to explicitly acknowledge the measurement boundaries and address the potential issue of construct overlap.

Section 6.3 (Research limitations and future directions):

Fifth, the mediating constructs in this study measure students' perceived benefits of AI use regarding their engagement (PCEB, PBEB, PEEB), rather than engagement metrics. Since both the mediators and the outcome variable (attitudes) are self-report measures that reference AI and involve evaluative judgments, the observed indirect effects may be partly driven by construct overlap. Although our discriminant validity analysis confirmed the statistical distinctness of these constructs, future research would benefit from incorporating objective measures of engagement. (Section 6.3)

Comment 2. The revision adds a "three-layer progressive consistency framework." suggesting that it is developmental when in fact t is parallel. It is cross-sectional. Three mediators side by side. It cannot test temporal order. So the framework's core claims are should be more tentatively claimed. Either scale the theory back to what the model evaluates. Or state clearly that the framework is an interpretive heuristic. Not a tested structure.

●Response:

We greatly appreciate your rigorous evaluation of our theoretical framework and empirical model. You are correct that because our survey data is cross-sectional and the three mediators are evaluated in a parallel configuration, we cannot empirically verify a temporal order. We agree that our previous claims regarding a "developmental" or "progressive" trajectory were overstated. To address this, we have tempered our theoretical claims in Section 3.4. We revised the language to present a "three-dimensional theoretical framework" with "complementary interpretive angles" rather than a sequential or developmental process. Most importantly, we have added a clear statement explicitly defining the framework as an interpretive heuristic rather than a tested structure.

●Modification:

Section 3.4 (Research model):

It is important to note that, while this framework proposes multiple layers of psychological processing, the cross-sectional survey design does not allow for empirical tests of temporal sequence or causal direction among the dimensions. Accordingly, this theoretical framework should therefore be interpreted as a conceptual heuristic to map concurrent psychological channels rather than a temporally verified path structure. (Section 3.4)

Comment 3. The "proportion mediated" figures (53.6% / 54.9%) are unstable with small effects. De-emphasize them.

●Response:

We sincerely thank the reviewer for this valuable methodological advice. We fully agree that the "proportion mediated" statistic tends to be unstable when the indirect effect is relatively small.

Accordingly, we have removed the unstable figures. We have completely deleted all reports of the "proportion mediated" values (53.6% and 54.9%) from the revised manuscript.

•Modification:

The following sentence was deleted from Section 5.4 (Hypothesis testing):

"Of the significant indirect effects, CE and EE showed comparable mediating effects, accounting for 53.6% (95% CI [32.8%, 75.6%]) and 54.9% (95% CI [30.3%, 81.8%]) of the total indirect effects, respectively." (Deleted)

Comment 4. One recommendation contradicts a finding. The paper says use AI to "stimulate behavioral engagement." But BE does not relate to attitudes. Reconcile this.

●Response:

We sincerely appreciate the reviewer’s astute observation regarding this logical contradiction. You are correct that advising educators to "stimulate behavioral engagement" directly conflicted with our empirical finding that perceived behavioral engagement benefits do not significantly relate to or mediate student attitudes.

To resolve this contradiction and ensure strict alignment between our empirical evidence and pedagogical recommendations, we have systematically removed and rephrased the active advocacy for stimulating behavioral engagement in both the Practical Implications and Conclusions sections.

•Modification:

(1) In Section 6.2 (Practical implications), the third paragraph has been revised as follows:

A critical examination should be conducted on behavioral engagement in AI-supported classroom settings. The current study find that perceived behavioral engagement benefits cannot mediate the relationship between AI usage and attitudes toward AI, implying that superficial behavioral performance is insufficient to realize the educational value of AI. In teaching practice, AI should neither be used merely for the sake of doing so, nor should its usage be oversimplified, including such superficial practices as content generation and automated question answering [16]. Instead, emphasis should be placed on promoting deep learning and fostering emotional experiences. (Section 6.2)

(2) In Section 7 (Conclusions), the relevant pedagogical recommendation paragraph has been amended as follows:

Educational practice should move beyond the debate over whether AI use should be permitted, and instead the focus should be shifted to designing human-AI collaborative learning experiences that use AI as a catalyst for fostering deep learning and student development. This study recommends that educators integrate AI tools into their teaching design, optimize their guidance on in-class AI usage, and look beyond superficial behavioral participation to focus primarily on facilitating deep cognitive integration and fostering positive emotional experiences for their students. (Section 7)

Comment 5. Control variables (gender, grade, major) are not fully specified in the structural model. This matters. The disciplinary differences for CE and BE were significant.

● Response:

We sincerely thank the reviewer for identifying this critical issue. In response, we have made the following revisions:

We re-ran the structural equation model analysis. In the final adopted model, gender, grade, and major were specified as covariates predicting all endogenous variables (i.e., the independent variable, mediator, and dependent variable) in the structural model.

We compared the hypothesis testing results between the two models and confirmed that all hypothesized paths remain significant with unchanged directions before and after covariate adjustment (see the comparison below). This demonstrates that our findings are robust and not contingent upon whether covariates are controlled. Importantly, after controlling for major as a covariate, all our hypothesized relationships continue to hold, indicating that the observed effects are generalizable across disciplines.

● Modification:

In Section 5.4 (Hypothesis testing), we have updated the model description to explicitly clarify the inclusion of covariates, and the relevant paragraph has been revised as follows:

In this study, the hypotheses were tested using SEM conducted in Mplus 8.3. Bayesian analysis was employed with the Markov Chain Monte Carlo (MCMC) method for parameter estimation. The model employed two parallel chains over 40,000 iterations, with the first 10,000 iterations used as burn-in. Posterior samples were saved at intervals of every 10 iterations. Convergence diagnostics indicated that the final value of the potential scale reduction factor (PSRF) was 1.009, and the PSR values for all parameters were close to 1.000. This is well below the strict threshold of 1.05, thus verifying the reliability of the subsequent parameter estimation results [68, 69]. In addition, examination of the trace plots and autocorrelation plots for the parameters’ posterior distributions revealed no abnormal trends or high autocorrelation, further verifying the good model convergence. All analyses controlled for gender, grade, and major as covariates affecting the independent, mediating, and dependent variables. (Section 5.4)

Furthermore, the data in the corresponding tables and the path diagram have been updated to reflect the newly estimated coefficients after covariate adjustment:

1. Comparison of Table 6 (Before and After Modification)

Original Table 6: Standardized path coefficients of the SEM

Paths Coefficient estimates S.E. Lower 95%CI Upper 95%CI

CU→AT 0.197 0.058 0.083 0.308

CU→CE 0.338 0.061 0.212 0.453

CU→BE 0.272 0.066 0.139 0.397

CU→EE 0.374 0.061 0.249 0.489

CE→AT 0.380 0.068 0.242 0.509

BE→AT -0.074 0.067 -0.205 0.056

EE→AT 0.351 0.080 0.191 0.508

Revised Table 6: Standardized path coefficients of the SEM

Paths Coefficient estimates S.E. Lower 95%CI Upper 95%CI

CU→AT 0.185 0.058 0.071 0.299

CU→PCEB 0.338 0.062 0.212 0.454

CU→PBEB 0.284 0.066 0.151 0.410

CU→PEEB 0.383 0.062 0.254 0.499

PCEB→AT 0.384 0.069 0.244 0.516

PBEB→AT -0.072 0.068 -0.206 0.060

PEEB→AT 0.346 0.080 0.187 0.502

2. Comparison of Table 7 (Before and After Modification)

Original Table 7: Results of multiple mediating effect test (standardized)

Standardized effects Effect size proportion SE Estimates 95% CI

Lower Limit Higher Limit

Standardized total effects

CU→AT 100% 0.058 0.347 0.241 0.469

Standardized direct effects

CU→AT 56.8% 0.058 0.197 0.083 0.308

Standardized indirect effects

CU→CE→AT 28.8% 0.028 0.100 0.053 0.164

CU→BE→AT -4.3% 0.016 -0.015 -0.052 0.012

CU→EE→AT 29.7% 0.032 0.103 0.050 0.174

Total indirect effects 54.5% 0.039 0.189 0.120 0.273

Revised Table 7: Results of multiple mediating effect test (standardized)

Standardized effects Effect size proportion SE Estimates 95% CI

Lower Limit Higher Limit

Standardized total effects

CU→AT 100% 0.057 0.335 0.231 0.454

Standardized direct effects

CU→AT 55.2% 0.058 0.185 0.071 0.299

Standardized indirect effects

CU→PCEB→AT 53.8% 0.029 0.100 0.053 0.164

CU→PBEB→AT - 0.017 -0.015 -0.052 0.013

CU→PEEB→AT 54.8% 0.032 0.102 0.050 0.173

Total indirect effects 56.1% 0.039 0.188 0.119 0.273

3. Comparison of Path Diagram (Before and After Modification)

Original Fig 2: Mediating path diagram. (Before controlling for covariates)

Revised Fig 2: Mediating path diagram. (After controlling for covariates)

Journal Requirements:

1. If the reviewer comments include a recommendation to cite specific previously published works, please review and evaluate these publications to determine whether they are relevant and should be cited. There is no requirement to cite these works unless the editor has indicated otherwise.

●Response:

Thank you for the reminder. We confirm that no citations were recommended by the reviewers.

2. Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice.

●Response:

We thank the editor for this important reminder. Our responses are as follows:

(1) Retracted publications: We have carefully verified every entry in our reference list and confirm that none of the cited papers have been retracted. Therefore, no rationale for citing retracted articles is needed.

(2) Changes to the reference list: One replacement has been made. The original reference [61] has been substituted with a new publication. Details are provided below:

Removed (original [61]): Liu E, Dewaele JM, Wang J. Developing a short language classroom engagement scale (LCES) and linking it with needs satisfaction and achievement. System. 2024;120(1):103189. doi: 10.1016/j.system.2023.103189.

Added (new [61]): Fan W, Cheng L, Wang Y, Zhao Q, Li Y. In-class AI use and attitudes among university students: The different mediating roles of cognitive relief and cognitive offloading. Behav Sci. 2026;16(6):1014. doi: 10.3390/bs16061014.

The corresponding in-text citation has been updated accordingly, and all changes have been clearly marked in the revised manuscript for the editor's and reviewers' convenience.

Declaration on Data Reuse

In addition to addressing the reviewers' comments, we would also like to make a declaration regarding data reuse:

Data for the present study and Fan et al. (2026) (Available from: https://www.mdpi.com/2076-328X/16/6/1014� were sourced from the same large-scale survey project. The two investigations overlap on two constructs: in-class AI usage and user attitudes. Whereas Fan et a

Attachments

Attachment

Submitted filename: Response_to_Reviewers_auresp_2.docx