AI Labels, Perceived Authenticity, and Consumer Trust in User-Generated Reviews

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This paper studies a timely and useful question: how different levels of AI involvement in user-generated reviews affect consumer trust and perceived authenticity in digital marketplaces. The topic is relevant, and the paper has practical value for platforms and online sellers. I also think it is a good idea to separate AI-assisted content from fully AI-generated content. The overall result pattern is clear and interesting. However, the current version still has several major problems in research design, reporting, and presentation. The biggest problem is that the review text stayed the same across the three conditions, while the image and the AI label changed. Because of this, the paper cannot make strong claims about AI-generated reviews in general. There are also important problems with the testing of the hypotheses, the reporting of the sample, the references, and the English. For these reasons, I recommend major revision.

1. The main design problem is that the paper claims to compare human-created, AI-assisted, and AI-generated reviews, but the written review text did not change across conditions. Only the image and the AI label changed. Because of this, the study seems to test differences in review presentation and AI labeling more than differences in review authorship or review generation itself. The claims in the title, abstract, and conclusion should be narrowed, unless the authors add a design that directly changes the review text.

2. H3 is not tested in a fully convincing way. The paper says that AI disclosure reduces trust by lowering perceived authenticity, but the study does not clearly isolate disclosure as its own independent factor. The current analysis mainly tests scenario differences, not the direct effect of disclosure itself. The wording of H3 and the claim that it is fully supported should be reconsidered.

3. The reporting of the sample is inconsistent. The abstract mentions 370 participants, Table 1 shows 388, some tables use 369, and the repeated-measures ANOVA seems to be based on complete cases only. The paper needs a clear explanation of the full sample, exclusions, valid responses, missing data, and final analysis sample.

4. The methods section needs more detail. Please explain more clearly how the scenario variable was coded, how the mediation analysis was run, how the within-subject nature of the data was handled, and whether any assumption checks were performed for the repeated-measures ANOVA.

5. The measurement section also needs revision. The paper says “reliability and validity,” but it mainly reports internal consistency. Also, the item wording is not fully consistent between the main text and the appendix. In addition, the appendix mentions a reverse-coded item, but this is not clearly shown in the main paper. These points need to be corrected.

6. The reliability results for perceived authenticity in the AI-assisted condition are relatively weak. The paper should discuss this more carefully and explain what this means for the later hypothesis tests, especially for the mediation analysis.

7. The reference system needs major cleanup. There are many places with “Error! Reference source not found.” This should not appear in a review-ready manuscript. I also noticed some citation mismatches, where the source named in the text does not match the number in the reference list. Please check all in-text citations and references carefully.

8. The presentation can be improved. Some table references are wrong, some wording is repetitive, and a few template placeholders are still visible on the first page. These problems make the paper look unfinished.

9. The English needs substantial editing. The meaning is often understandable, but there are many awkward sentences, grammar problems, and unclear phrases. A careful language edit is needed before the paper can move forward.

10. The paper has a relevant topic and some promising results, but the current version overstates what the design can really support. If the authors narrow the claims, improve the methods and reporting, and fix the references and language, the paper could become much stronger.

Author Response

Dear Reviewer,

Thank you very much for taking the time to review our manuscript and for your valuable comments and suggestions. We greatly appreciate your careful reading and constructive feedback.

We have addressed all of your comments in detail. Please find our point-by-point responses in the attached document, where we explain how each of your remarks has been considered and incorporated into the revised manuscript.

We believe that these revisions have significantly improved the quality and clarity of the paper.

Thank you again for your contribution to the review process.

Kind regards,
The Authors

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The article's subject is highly pertinent, as it aligns with the ongoing debate over the ethics and transparency of using generative artificial intelligence in e-commerce. In light of the mounting ethical dilemmas posed by the manipulation of consumer expectations through often unrealistic images generated by GenAI, investigating the impact of disclosure on perceived authenticity and trust addresses the key challenges currently facing both digital platforms and regulatory bodies. Given the growing expectations for institutional regulation in this area, it is essential to gain a deeper understanding of how consumers perceive and respond to AI-driven interventions.

The article possesses several distinct advantages, most notably a robust theoretical foundation based on the Stimulus-Organism-Response (SOR) model and Signalling Theory, which provides an appropriate and rigorous framework for explaining how AI involvement in labelling influences authenticity and trust.

A significant merit of this research is also its move beyond a simple Human-AI divide through a nuanced categorisation of AI involvement that captures the practical realities of modern content creation and provides more precise insights than previous studies.

The research conducted by the authors demonstrates high methodological rigour through the technically sound application of repeated-measures ANOVA and mediation analysis, which provides a statistically validated description of the role of perceived authenticity in mediating the relationship between AI involvement and trust.

The research carries significant practical implications, particularly vital as GenAI fundamentally reshapes the reality for both consumers and firms. By offering actionable advice on content labelling, the study provides a timely roadmap for e-commerce platforms and regulators, suggesting that a hybrid approach (AI-assisted) is the most viable strategy for balancing technological innovation with user engagement and trust.

Despite these merits, several areas remain that may raise concerns for the reviewer, particularly regarding measurement reliability issues; specifically, the Perceived Authenticity scale demonstrated low internal consistency in the AI-assisted scenario (α=0.564), which, despite the authors' acknowledgement, represents a critical weakness that potentially undermines the validity of the mediation analysis for that specific condition.

Furthermore, the research is constrained by significant methodological limitations regarding external validity; specifically, the reliance on convenience sampling among Latvian digital platform users limits the generalisability of the findings to other cultural or market contexts. Moreover, the study focuses on self-reported attitudes rather than actual consumer behaviour; the inclusion of behavioural data (e.g., actual click-through rates or purchase intent in a simulated shop) would have strengthened the results.

The manuscript contains several technical oversights, such as "Error! Reference source not found.", and inconsistent formatting of citations. These issues must be resolved to meet the standards of a leading international journal.

In summary, I recommend the manuscript for publication provided the authors rectify the technical errors, address the aforementioned statistical concerns in detail, and clarify why broad generalisations have been drawn despite these evident methodological limitations.

Author Response

Dear Reviewer,

Thank you very much for taking the time to review our manuscript and for your valuable comments and suggestions. We greatly appreciate your careful reading and constructive feedback.

We have addressed all of your comments in detail. Please find our point-by-point responses in the attached document, where we explain how each of your remarks has been considered and incorporated into the revised manuscript.

We believe that these revisions have significantly improved the quality and clarity of the paper.

Thank you again for your contribution to the review process.

Kind regards,
The Authors

Author Response File: Author Response.docx

Reviewer 3 Report

Comments and Suggestions for Authors

The manuscript addresses a topic of clear contemporary relevance, namely the impact of artificial intelligence on user-generated content and its implications for consumer trust in digital marketplaces. The distinction between human-created, AI-assisted, and fully AI-generated reviews is conceptually appropriate and aligns with current developments in digital marketing and platform governance. The experimental approach adopted by the authors is methodologically suitable, and the use of repeated-measures ANOVA and mediation analysis is, in principle, adequate for addressing the research question.

However, despite these positive aspects, the manuscript in its current form presents several substantive weaknesses that significantly limit its scientific contribution and readiness for publication.

First, the manuscript contains numerous unresolved referencing errors (e.g., “Error! Reference source not found.”), which constitute a serious technical flaw. These errors undermine the credibility of the work and prevent a proper assessment of the theoretical foundation. This issue alone indicates insufficient editorial rigor and must be fully corrected before the manuscript can be meaningfully evaluated.

More fundamentally, the theoretical contribution of the study remains limited and insufficiently articulated. While the manuscript adopts signaling theory as its primary framework, it does not convincingly demonstrate how it advances this theoretical perspective. The study largely applies existing concepts (e.g., authenticity, transparency, trust) predictably, thereby confirming relationships already well established in the literature. The positioning of the research gap is relatively generic, and the manuscript does not clearly specify what new theoretical insight is generated. As it stands, the contribution appears incremental rather than substantive.

The literature review, although structured, is predominantly descriptive and lacks critical depth. The manuscript tends to summarize prior studies without engaging in meaningful synthesis or theoretical integration. There is limited effort to reconcile conflicting findings or to develop a more nuanced conceptual argument. This weakens the foundation for the hypotheses, which, in their current form, are largely intuitive and insufficiently challenging from a theoretical standpoint.

From a methodological perspective, although the experimental design is appropriate, there are conceptual limitations that deserve closer attention. Notably, the decision to keep the review text identical across all experimental conditions, varying only the visual elements and disclosure labels, raises concerns about construct validity. In practice, this design primarily captures the effect of labeling and perceived AI involvement rather than differences in content authenticity per se. This limitation has important implications for the interpretation of the findings, yet it is not sufficiently problematized in the manuscript.

The sampling strategy further constrains the study’s contribution. The use of a convenience sample drawn from a single national context significantly limits external validity. While this limitation is acknowledged, its implications are not critically examined. Given the cultural and contextual sensitivity of trust and perceptions of AI, this represents a non-trivial constraint on the generalizability of the findings.

Regarding measurement, the reported reliability levels are uneven, with some Cronbach’s alpha values falling below commonly accepted thresholds. Although the authors attempt to justify this, the explanation remains somewhat superficial. Additional evidence supporting the robustness of the measurement model would strengthen the methodological rigor.

The results are clearly presented and statistically consistent; however, they largely confirm expected patterns. The finding that fully AI-generated content is perceived as less trustworthy and authentic than human-created content is not particularly novel. Similarly, the intermediate positioning of AI-assisted content aligns closely with prior research. As such, the empirical contribution is somewhat limited by the predictability of the results.

The discussion section does not sufficiently compensate for this limitation. Rather than offering deeper theoretical interpretation or exploring alternative explanations, it tends to reiterate the results and align them with existing literature. A more critical and reflective discussion would be necessary to elevate the contribution of the study.

The practical implications are relevant and potentially valuable for digital platforms and marketers. However, they are presented at considerable length and with some degree of repetition. A more concise and focused articulation would improve their impact.

Comments on the Quality of English Language

In addition, the overall writing quality requires significant improvement. The manuscript contains grammatical inaccuracies, typographical errors, and instances of unclear or awkward phrasing. There are also several redundancies, including repeated statements of findings, particularly in the abstract and discussion sections. These issues affect readability and suggest that the manuscript has not undergone adequate linguistic revision.

 

Author Response

Dear Reviewer,

Thank you very much for taking the time to review our manuscript and for your valuable comments and suggestions. We greatly appreciate your careful reading and constructive feedback.

We have addressed all of your comments in detail. Please find our point-by-point responses in the attached document, where we explain how each of your remarks has been considered and incorporated into the revised manuscript.

We believe that these revisions have significantly improved the quality and clarity of the paper.

Thank you again for your contribution to the review process.

Kind regards,
The Authors

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

I appreciate the authors’ efforts in revising the manuscript and in preparing a detailed response letter. The revised version is clearly improved in several respects. In particular, the manuscript now explains more explicitly that the review text remained constant across conditions, the reporting of the sample is more transparent, the methods section is more detailed, and the discussion takes a more cautious view of the relatively weak internal consistency of the perceived authenticity scale in the AI-assisted condition. These revisions strengthen the manuscript.

 

At the same time, after reviewing the revised manuscript itself, I do not believe that all of the major concerns have been fully resolved. Several important issues remain, including unresolved citation errors, inconsistencies in the use of key terms, and some remaining misalignment between the title, the hypotheses, and what the study design actually tests. In my view, the manuscript would therefore benefit from a further round of revision before it is ready for consideration.

 

1. The overall framing has improved, but the central claim is still not fully aligned with the design. The revised manuscript now states more clearly that the review text remained unchanged and that only the labels and images were varied. This clarification is helpful. However, the title still emphasizes “The Impact of AI Disclosure,” whereas the actual manipulation combines AI-related labels and visual cues. Because disclosure was not isolated as an independent factor, the title and some interpretive statements still appear somewhat stronger than the design can fully support.

 

2. A related issue remains in the wording of the hypotheses and the summary of results. For instance, Table 10 refers to “reviews labeled as human-created,” but the baseline condition is more accurately described as a review presented without AI-related information, rather than a direct human-created label. This distinction is important and should be expressed more consistently throughout the manuscript.

 

3. The reference system has not yet been fully corrected. The revised manuscript still contains visible citation errors, including examples such as “[1010],” “[19,20Error! Reference source not found.],” and “[10Error! Reference source not found.].” This remains an important presentation and quality-control issue and should be carefully resolved throughout the manuscript.

 

4. The methods section is improved, but the reporting of assumption checks is still somewhat limited. The response letter mentions procedures such as Mauchly’s test of sphericity and consideration of the Greenhouse-Geisser correction, but the revised manuscript does not clearly report the corresponding results. If these checks were conducted, their outcomes should be reported explicitly enough for readers to evaluate the analysis.

 

5. The discussion of measurement quality is now more careful, which I appreciate. However, some inconsistencies remain. The manuscript uses “AT1” in the main text, while Appendix B still reports “ATP1.” In addition, the citation used for the AI transparency item still appears problematic, as the cited source is Podsakoff et al. (2003), which is a common method bias reference rather than a source for an AI disclosure or transparency measure. This point should be checked and corrected.

 

Overall, the manuscript has improved and the authors have made a serious effort to address the previous comments. Nevertheless, the current revision does not yet fully resolve the main concerns. I would therefore recommend a further round of revision, with particular attention to fully correcting the references and remaining inconsistencies, tightening the alignment between the title and the actual design, and making the methodological reporting more complete.

Author Response

Dear Reviewer,

Thank you very much for taking the time to review our manuscript and for your valuable comments and suggestions. We greatly appreciate your careful reading and constructive feedback.

We have addressed all of your comments in detail. Please find our point-by-point responses in the attached document, where we explain how each of your remarks has been considered and incorporated into the revised manuscript.

We believe that these revisions have significantly improved the quality and clarity of the paper.

Thank you again for your contribution to the review process.

Kind regards,
The Authors

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The revised version of the manuscript shows a clear and substantial improvement compared to the original submission. The authors have addressed several of the concerns raised in the previous review seriously and constructively, particularly regarding methodological transparency, clarification of the study’s limitations, and the articulation of the experimental design.

One of the most important improvements concerns the clarification of the experimental manipulation. The revised manuscript now more explicitly acknowledges that the study primarily captures the effects of AI-related labels and presentation cues rather than substantive differences in review content itself. This clarification significantly improves the internal coherence of the study and provides a more accurate interpretation of the findings. The limitations section has also been considerably strengthened, especially regarding construct validity, external validity, and the implications of using identical review texts across conditions.

The discussion of methodological limitations is now more balanced and reflective. The authors appropriately acknowledge the constraints associated with the Latvian convenience sample, the absence of behavioral measures, and the limitations related to the authenticity scale. The additional reporting of reliability indicators and measurement-related analyses also improves the methodological transparency of the study.

The manuscript has further benefited from revisions to the discussion and theoretical framing. The effort to position perceived authenticity as an interpretive mechanism within signaling theory has strengthened the conceptual coherence of the study and improved the articulation of the proposed contribution.

Nevertheless, despite these improvements, some concerns remain.

First, although many technical issues were corrected, the manuscript still contains unresolved referencing and formatting errors (e.g., “Error! Reference source not found.”). The persistence of these issues at this stage of the review process is problematic and should be fully resolved before publication. A careful editorial revision of the entire manuscript is still required.

Second, while the theoretical framing is clearer than in the original version, the overall theoretical contribution remains somewhat limited and incremental. The study continues to confirm relationships that are largely expected and already well supported in prior literature, namely that fully AI-generated content is perceived as less authentic and trustworthy than human-created content, while AI-assisted content occupies an intermediate position. The revised manuscript improves the articulation of this contribution, but the extent to which it substantially advances signaling theory or the broader literature on AI-mediated communication remains somewhat modest.

In addition, although the literature review and discussion sections are now more integrated, parts of the manuscript still remain descriptive rather than critically analytical. Greater theoretical synthesis and deeper engagement with alternative interpretations would further strengthen the paper.

The issue of measurement reliability also remains relevant. Although the authors have provided additional justification and supplementary indicators, the relatively low Cronbach’s alpha reported for the AI-assisted authenticity construct continues to represent a limitation that should be acknowledged cautiously when interpreting the mediation results.

Finally, the manuscript would still benefit from additional language polishing and stylistic refinement. While readability has improved, some sections — particularly the practical implications — remain somewhat repetitive and overly extended.

Overall, the revised manuscript represents a meaningful improvement over the original submission. The authors have addressed most of the major concerns raised during the first review round, and the study is now methodologically more transparent and conceptually clearer. However, some editorial issues and limitations regarding theoretical originality and analytical depth still require attention before the manuscript can be considered fully ready for publication.

Comments on the Quality of English Language

In addition, the overall writing quality requires significant improvement. The manuscript contains grammatical inaccuracies, typographical errors, and instances of unclear or awkward phrasing. There are also several redundancies, including repeated statements of findings, particularly in the abstract and discussion sections. These issues affect readability and suggest that the manuscript has not undergone adequate linguistic revision.

 

Author Response

Dear Reviewer,

Thank you very much for taking the time to review our manuscript and for your valuable comments and suggestions. We greatly appreciate your careful reading and constructive feedback.

We have addressed all of your comments in detail. Please find our point-by-point responses in the attached document, where we explain how each of your remarks has been considered and incorporated into the revised manuscript.

We believe that these revisions have significantly improved the quality and clarity of the paper.

Thank you again for your contribution to the review process.

Kind regards,
The Authors

Author Response File: Author Response.docx