03 · What You Need to Know
How to Handle Editorial Questions About Undisclosed AI Use
Why Is Discovery During Peer Review a Distinct Problem?
Peer review is an evaluation of the manuscript's scientific quality, methodological appropriateness, originality, and contribution.
When undisclosed AI assistance is identified during this process, editors may need to determine whether the omission affects the accuracy of the submission declarations or the evidence available to reviewers.
The issue can also influence confidence in the authors' descriptions of how the work was prepared.
However, discovery during peer review does not automatically make the underlying research unreliable.
A manuscript may contain accurately verified AI-assisted work that was not disclosed as required. Conversely, a manuscript may contain serious errors that need correction regardless of whether AI was used.
The editorial inquiry should distinguish transparency from scientific validity and possible misconduct.
First, Determine Whether AI Use Is Established or Merely Suspected
Not every allegation of AI-generated writing is supported by reliable evidence.
A reviewer may suspect AI use because of repetitive phrasing, unusually polished language, fabricated references, or the output of an automated AI-text detector.
These indicators differ in evidentiary value.
A verified record of AI-assisted drafting establishes a particular use more directly than a stylistic impression.
A fabricated citation establishes a bibliographic problem but does not, by itself, prove which tool or person generated it.
Likewise, an AI detector's classification should not automatically be treated as proof of undisclosed generative AI use.
Research by Liang and colleagues examined how GPT detectors can misclassify writing by non-native English speakers, illustrating concerns about using detector output as decisive evidence.
The authors and editor should therefore distinguish suspicion from independently established facts.
What if the Reviewer Relies on an AI-Text Detector?
Automated AI-text detectors attempt to infer whether text was generated by AI from characteristics of the writing.
Their outputs are probabilistic assessments or classifications, not direct records of the manuscript's production history.
Performance may vary with the detector, model, text length, genre, language background, and editing process.
A positive detection result does not establish that an author violated a journal policy.
If an editorial inquiry relies on detector output, the authors can request clarification about the concern and provide relevant evidence of their writing process.
Such evidence might include dated drafts, version histories, analytical records, and documentation of any actual AI assistance.
Researchers should not fabricate drafts or alter records to create an appearance of human-only authorship.
The appropriate response is to establish what happened, not to attempt to defeat a detector.
Respond to the Editor Rather Than Arguing Directly With the Reviewer
Editorial inquiries should ordinarily be handled through the journal's established communication channels.
The corresponding author should acknowledge the inquiry, coordinate with coauthors, and provide an accurate response.
If additional time is needed to reconstruct AI use or verify affected analyses, the authors can explain what remains under investigation and request a reasonable opportunity to complete the review.
Do not contact anonymous reviewers directly unless the journal explicitly authorizes it.
A professional response should focus on evidence and the applicable requirements rather than speculation about the reviewer's motives.
Even when an allegation is mistaken, factual clarification is more useful than treating the inquiry itself as misconduct by the reviewer.
Review the Journal's Actual AI Policy
Before responding to an alleged disclosure violation, establish what the journal required.
Relevant documents may include the author instructions, submission declarations, editorial policies, and requirements applicable when the manuscript was submitted.
Some journals require disclosure of generative AI used in drafting or substantive content development. Others distinguish routine language assistance from activities affecting research methods.
The International Committee of Medical Journal Editors recommends transparency about AI assistance and emphasizes human responsibility for scholarly content.
However, its recommendations do not automatically establish the policy of every journal.
If the relevant policy changed after submission, the editor may need to clarify how the newer requirements apply to the manuscript's current status.
The authors should avoid assuming either that every AI interaction was prohibited or that no disclosure was necessary.
Determine What the AI Actually Did
The seriousness of an editorial concern depends on the contribution that was omitted.
| AI involvement |
Potential editorial concern |
Evidence or review needed |
| Language polishing |
Missing required disclosure or altered scientific meaning |
Original and revised manuscript passages |
| Substantive drafting |
Unsupported claims, inaccurate citations, or unclear attribution |
Source verification and drafting records |
| Literature summarization |
Misrepresented findings or nonexistent references |
Original publications and claim-source checks |
| Statistical programming |
Unverified computations or incomplete methodological reporting |
Code, data dictionary, analysis plan, and independent checks |
| Qualitative coding |
Inadequate analytical transparency or distorted participant meaning |
Coding framework and original authorized data |
| Figure generation |
Inaccurate representation or insufficient explanation of how the figure was produced |
Underlying data and figure-production records |
| Participant-data processing |
Potential privacy, consent, or ethics noncompliance |
Approved procedures, data-sharing conditions, and relevant institutional review |
This comparison illustrates possible concerns rather than prescribing a universal editorial sanction.
The same activity may be acceptable in one research context and restricted in another.
What if AI Was Used Only for Grammar Correction?
Language assistance may have limited scientific consequences when it preserves the original meaning and introduces no new claims.
Nevertheless, whether disclosure was required depends on the journal's policy.
Researchers should identify the nature of the assistance and verify that the editing did not alter methods, findings, certainty, or interpretation.
If the journal required disclosure of that activity, the authors should correct the omission.
If the policy explicitly excluded such assistance, the authors can explain that distinction to the editor.
Neither conclusion should be assumed without checking the applicable instructions.
What if AI Generated Part of the Literature Review?
AI-assisted literature review preparation may raise concerns about both transparency and evidentiary accuracy.
Generative AI can produce plausible but nonexistent references or misrepresent genuine publications.
The authors should identify the affected passages and verify consequential claims against original scholarly sources.
For each cited study, establish that the publication exists and that it supports the relevant statement.
If fabricated references or unsupported conclusions are discovered, the authors should correct the manuscript and explain material changes to the editor.
A disclosure statement does not make unreliable references acceptable.
What if AI Generated Statistical Code or Research Results?
Undisclosed AI-assisted analysis can have greater methodological significance because it may affect how the study's evidence was produced.
Researchers should establish which procedures were generated or modified with AI assistance and whether the resulting code was independently validated.
For example, the team may need to inspect data transformations, variable selection, model specifications, and reported estimates.
Where consequential results depend on generated code, independent reproduction or qualified methodological review may be appropriate.
If the code is accurate, the journal may still require a corrected disclosure or methodological description.
If the code is wrong, the authors must address the scientific error separately.
The process for correcting incorrect results produced by AI-generated code should determine which analyses and conclusions are affected.
What if the AI Use Was Undisclosed Because a Coauthor Concealed It?
The corresponding author may discover during peer review that another contributor used AI without informing the research team.
The team should establish what occurred and whether the coauthor understood the applicable expectations.
An honest misunderstanding differs from deliberately making a false declaration.
However, the coauthor's intent does not determine whether the submitted research requires correction.
Relevant outputs should be verified, and any required disclosure should be supplied through the editorial process.
Where serious concerns about deliberate misrepresentation arise, the team should follow appropriate institutional procedures rather than making unsupported accusations.
Can a Reviewer Demand Access to Every AI Prompt?
There is no universal requirement that authors provide complete AI interaction histories to reviewers.
Journals may request documentation necessary to evaluate methodological transparency, research integrity, or compliance with their policies.
Researchers should determine which records are relevant and which can appropriately be shared.
AI conversations may contain confidential research information, unpublished ideas, or participant data that cannot be disclosed without authorization.
The team should not publicly release protected information merely to demonstrate transparency.
Where complete records cannot be shared, the authors can explain the limitation and provide appropriate alternative documentation through authorized channels.
The goal is meaningful verification, not unrestricted disclosure of every interaction.
What if the Authors Cannot Reconstruct the AI Contribution?
Incomplete documentation can complicate an editorial inquiry.
A coauthor may no longer have the original AI conversation or may not know which model version was used.
The team should recover information from available records, including manuscript versions, code repositories, notes, and saved outputs.
Do not invent prompts, dates, or model identifiers.
Explain what can be established and what remains uncertain.
If the missing information prevents adequate verification of consequential research procedures, the team may need to repeat the analysis or seek qualified review.
The inability to reconstruct an interaction does not automatically prove misconduct, but it may limit the defensibility of affected work.
What if the Editor Requests a Corrected AI Disclosure?
Provide a statement consistent with the actual contribution and the journal's instructions.
Where required, identify the AI service, available model information, tasks performed, and human verification undertaken.
Avoid vague wording that obscures substantive AI involvement.
For example, describing AI-generated statistical programming only as language assistance would misrepresent the activity.
Equally, do not exaggerate the role of AI when it performed a limited editorial task.
The revised disclosure should be reviewed by the responsible authors before submission.
If the journal also requests changes to the Methods section, ensure that those changes accurately describe the analytical workflow.
Can the Journal Request Additional Peer Review?
Yes, depending on the significance of the new information and the journal's procedures.
If undisclosed AI assistance materially affected methodology or results, the editor may need additional expertise to assess the revised work.
For example, an AI-assisted classification procedure may require review of its validation methods.
Additional review is not automatically necessary for every corrected disclosure.
The editor may instead permit an administrative update or request a limited revision.
The appropriate response should be proportionate to the scientific consequences of the AI contribution.
Could the Manuscript Be Rejected?
Rejection is a possible editorial outcome, but it is not an automatic consequence of undisclosed AI use.
The journal may consider whether its policies were violated, whether the omission was material, whether the underlying work is reliable, and whether adequate correction is possible.
Some problems may be resolved through disclosure and verification.
Others may involve prohibited activity or substantial uncertainty about the findings.
Authors should not assume that acknowledging the omission guarantees acceptance or rejection.
They should provide accurate information and allow the editorial process to determine the appropriate outcome.
What if the Reviewer Accuses the Authors of Misconduct?
A reviewer may raise a research-integrity concern, but a formal misconduct finding requires the applicable institutional or regulatory process.
Under the United States Public Health Service framework, research misconduct concerns fabrication, falsification, or plagiarism within its specified scope and requirements.
Undisclosed AI use does not automatically establish those elements.
Editors may refer serious concerns to an institution or other responsible authority when appropriate.
The authors should cooperate with legitimate inquiries, preserve relevant records, and distinguish established facts from allegations.
They should not treat every critical reviewer comment as a misconduct determination.
How Should Authors Respond if the Allegation Is Incorrect?
If the manuscript did not involve the alleged AI use, the authors should explain the relevant facts and provide available supporting documentation.
Version histories, research notes, original analytical files, and source records may help establish how the work was prepared.
Authors should avoid claiming that stylistic characteristics alone prove human authorship, just as those characteristics do not reliably prove AI authorship.
The response should address the editor's specific concern.
For example, if the allegation arose from a suspicious citation, verify the reference regardless of whether AI was involved.
A mistaken inference about AI use does not eliminate the obligation to correct an independently established manuscript error.
What if the AI Use Involved Confidential Research Data?
Undisclosed processing of participant information may raise concerns beyond publication reporting.
The research team should determine what data were supplied, whether the system was approved, and what protections applied.
Consent conditions, ethics approvals, contractual restrictions, and applicable privacy requirements may govern the activity.
Where a possible breach occurred, institutional privacy, security, or ethics procedures may need to be initiated.
The editor should receive information relevant to the manuscript and applicable reporting obligations, but confidential participant records should not be shared indiscriminately.
A corrected AI disclosure cannot retrospectively authorize prohibited data processing.
How Should the Research Team Document the Editorial Inquiry?
Preserve the editor's questions, the team's factual investigation, relevant AI records, verification procedures, and the final response.
Maintain appropriate versions of the manuscript and supporting analyses.
If the inquiry identifies an error, document the correction and its effects on the scientific conclusions.
Where authors disagree about the response, record the basis for consequential decisions and follow applicable institutional procedures.
Documentation should remain accurate and should protect confidential research material.
Do not reconstruct false records to satisfy an editorial request.
How Is This Different From Discovering the Omission Yourself?
When researchers independently identify undisclosed AI use after submission, they should initiate the appropriate correction without waiting for editorial inquiry.
When a reviewer or editor raises the issue, the team must also respond to specific questions and evidence presented during peer review.
The scientific and disclosure responsibilities overlap, but the communication context differs.
The process for addressing undisclosed AI use discovered after manuscript submission focuses on proactive correction.
An active peer-review inquiry additionally requires careful assessment of allegations, factual clarification, and cooperation with the editor's evaluation.
Prevent Future Editorial Disputes Through Pre-Submission Review
Research teams should establish clear expectations about AI assistance before manuscript submission.
Contributors should identify relevant AI-assisted activities, verify consequential outputs, and review the final disclosure statement.
Where AI contributes to methodology or analysis, sufficient information should be retained to explain the procedure and reproduce the results.
The corresponding author should compare the team's records with the journal's requirements.
These measures cannot prevent every editorial question, but they can reduce uncertainty about what AI did and whether the research was adequately verified.
Watch Out
Do not treat an AI detector result, unusual writing style, or reviewer suspicion as conclusive evidence of undisclosed AI use. Equally, do not conceal confirmed AI assistance when disclosure was required. Respond through the editor, establish the facts, and verify consequential research content independently.
