03 · What You Need to Know
Why AI-Generated Revision Promises Can Misrepresent the Research Record
Reviewer Requests Are Not Evidence of Completed Revisions
A reviewer comment describes a concern or recommends an action. It does not establish that the authors have carried out that action.
For example, a request to conduct subgroup analysis does not mean the analysis is appropriate, feasible, or complete.
Similarly, a request to clarify participant recruitment does not establish that the Methods section has already been revised.
When generative AI receives reviewer comments without a verified record of author actions, it may produce conventional response language that assumes the requested work has been completed.
This creates a mismatch between the response letter and the research record.
Why AI May Convert Requests Into Promises
Response-to-reviewers letters follow recognizable academic writing conventions. Authors frequently use expressions such as "We have revised," "We have added," and "We have conducted."
Generative AI can reproduce these patterns even when it lacks information about the actual manuscript changes.
The problem is especially likely when a prompt asks the system to produce a complete response letter from reviewer comments alone.
Because the system is expected to generate a coherent answer, it may fill informational gaps with plausible revision statements rather than clearly identifying what remains unknown.
Such statements can appear credible because they resemble ordinary academic correspondence.
What Kinds of False Revision Promises Might AI Generate?
The risk extends beyond additional statistical analyses.
| Reviewer Request |
Possible AI-Generated Promise |
What Must Be Verified |
| Conduct sensitivity analysis |
"We have completed the requested robustness checks." |
Whether the analysis was performed and verified |
| Clarify sampling procedures |
"We have expanded the sampling subsection." |
Whether the revised text actually exists |
| Report additional validity evidence |
"We have conducted further instrument validation." |
Whether additional validation was performed |
| Discuss study limitations |
"We have comprehensively addressed these limitations." |
Which limitations were actually added or revised |
| Collect additional participant data |
"We have expanded the sample." |
Whether new data were collected with appropriate authorization |
| Improve theoretical justification |
"We have incorporated the recommended theoretical framework." |
Whether the framework was adopted and integrated appropriately |
| Revise a results table |
"The corrected results are now presented in Table 4." |
Whether the table exists and contains the described results |
These are hypothetical examples of misleading generated statements. They illustrate why authors should never treat a fluent response as evidence that the corresponding work occurred.
A Promise Is Different From a Claim of Completed Work
There is an important distinction between stating that a revision will be made and claiming that it has already been completed.
"We will conduct an additional analysis" expresses a future commitment. "We have conducted the analysis" asserts a completed action.
Both statements require verification, but in different ways.
A future commitment should reflect an action that the authors have explicitly approved and can reasonably undertake. A completed-action statement should be supported by an existing research record.
In a final resubmission response, describing unfinished work as a future promise may also be inadequate if the editor expects the revised manuscript to contain the completed changes.
Proposed revision
An action suggested by a reviewer, editor, or author but not yet approved.
Planned revision
An action the authors have decided to undertake but have not completed.
Completed revision
An action that has been performed and verified against the manuscript or research record.
Declined revision
A recommendation not adopted, with an appropriate explanation of the decision.
AI-assisted drafting should preserve these distinctions rather than converting every request into a completed action.
Why False Revision Promises Matter for Research Integrity
A response letter is part of the scholarly record associated with editorial evaluation. Editors and reviewers may rely on its statements when determining whether concerns have been addressed.
If authors claim to have completed work that does not exist, the letter misrepresents the revision process.
Not every inaccurate statement constitutes deliberate misconduct. An accidental drafting error should be distinguished from knowingly making a false claim.
Nevertheless, the absence of intent does not make the statement accurate or eliminate the need for correction.
Repeated or consequential inaccuracies may undermine editorial confidence and prompt further questions about the reliability of the submission.
AI May Promise Analyses That Should Not Be Conducted
Some reviewer requests are scientifically inappropriate or cannot be supported by the available data.
For example, a reviewer may request extensive subgroup analyses even when the sample contains too few observations for meaningful subgroup inference.
AI might nevertheless draft a response promising to conduct all requested comparisons.
Such a promise could pressure the authors into performing unnecessary or misleading analyses simply to satisfy the response letter.
The scientific appropriateness of the requested work must be evaluated before any commitment is made.
Authors may instead provide a reasoned explanation for declining an analysis, subject to the editor's instructions.
Conditional Language Can Also Create Misleading Commitments
AI-generated responses may use apparently cautious expressions that still imply an undertaking.
For example, "We will consider conducting additional experiments" may be too vague to address a substantive reviewer concern.
Similarly, "We intend to strengthen the validation procedure" may imply that a methodological revision is forthcoming without specifying what will actually be done.
Authors should distinguish a genuine commitment from a possibility under consideration.
Where the journal requires a completed revision, conditional language should not be used to conceal unfinished work.
Why Page and Line References Can Make False Promises More Convincing
Generative AI may accompany an invented revision with a plausible manuscript location.
For example, it might state that the new analysis appears on page 12, lines 280–305, even when no such passage exists.
This can make the response appear more credible while increasing the discrepancy between the letter and the manuscript.
Authors should insert page and line references only after checking the actual revised document.
Placeholders are preferable when locations have not yet been verified.
How Should Authors Handle Partially Completed Revisions?
Some reviewer concerns are addressed only in part.
For example, the authors may expand their explanation of missing-data handling but decide that the requested sensitivity analysis is not feasible.
The response should describe both decisions accurately.
It may explain what was clarified, why additional analysis was not performed, and how the manuscript's limitations were revised.
AI should not convert a partial response into a claim that the entire concern was resolved.
The related problem of AI overstating how fully reviewer concerns were addressed can arise even when some genuine revision work occurred.
Can AI Help Prevent False Promises?
Potentially, if the workflow requires the system to use verified revision records rather than infer completed work from reviewer comments.
Authors can maintain a revision tracker identifying each request, the decision taken, the completion status, and the corresponding manuscript evidence.
AI may then help draft responses using those verified records.
However, an instruction not to invent revisions does not guarantee that the model will comply perfectly.
Every final statement should still be checked by the authors.
Why the Final Manuscript Must Be the Reference Point
Drafting notes, revision plans, and AI conversations may describe work that was considered but never completed.
The final revised manuscript and supporting research records provide the evidence needed to verify claims about completed revisions.
For example, a response stating that a new regression model was estimated should correspond to documented code, outputs, and accurately reported results.
A statement that the theoretical framework was revised should correspond to identifiable changes in the manuscript.
Where the response letter and manuscript disagree, the discrepancy should be resolved before submission.
What If an Incorrect Promise Has Already Been Submitted?
If authors discover that a response letter contains an inaccurate commitment or claim of completed work, they should correct the record promptly.
The appropriate action depends on the journal's procedures and the stage of editorial processing.
Authors may need to contact the handling editor, explain the discrepancy, and provide a corrected response or manuscript.
They should not quietly create unsupported results or alter the research record to make the original statement appear true.
Transparent correction is preferable to allowing an inaccurate representation to remain unaddressed.
Confidentiality and Disclosure Still Apply
Reviewer comments, manuscript drafts, and revision records may contain unpublished information or confidential editorial material.
Authors should check the journal's AI-use policies and the service's data-handling arrangements before submitting such material to an external AI tool.
Where assistance is permitted, it should be used within authorized processing conditions.
Disclosure requirements vary across journals and publishers.
Authors remain responsible for the accuracy of the final response regardless of whether AI involvement was disclosed.
