03 · What You Need to Know
How to Resolve Undisclosed AI Use After Submission
Why Does Submission Change the Situation?
Before submission, researchers can ordinarily correct an incomplete AI-use statement as part of manuscript preparation.
After submission, the manuscript becomes part of an active editorial process. Editors and reviewers may rely on its methodological descriptions, declarations, and reported findings when assessing the work.
Consequently, an omitted AI contribution may be relevant not only to transparency but also to the accuracy of the information available to those evaluating the paper.
For example, undisclosed AI assistance in grammar correction may have limited methodological significance. Undisclosed AI classification of participant responses may directly affect how reviewers interpret the analytical procedure.
Researchers should therefore distinguish administrative disclosure requirements from information necessary to evaluate scientific validity.
The International Committee of Medical Journal Editors emphasizes human author accountability for AI-assisted scholarly content. Its recommendations also address disclosure of AI assistance, although individual journals may impose different requirements.
Establish Whether the Journal Required Disclosure
The first question is whether the relevant AI activity fell within the journal's actual disclosure policy.
Consult the author instructions, submission declarations, and other requirements governing the manuscript.
Pay attention to the version of the policy applicable at submission. AI policies may change, and a later revision should not automatically be treated as proof of an earlier violation.
Some journals distinguish ordinary language assistance from substantive generative AI contributions. Others require broader declarations.
The research team should also check institutional, funder, and project-specific requirements where relevant.
If the policy is ambiguous, seek clarification rather than assuming either that disclosure was unnecessary or that misconduct occurred.
Determine Exactly What AI Contributed
An accurate response requires reconstructing the actual AI-assisted activity.
Identify the tool or service, available model information, the task performed, the materials processed, and how the output entered the submitted manuscript.
| Undisclosed activity |
Primary question |
Appropriate initial action |
| Grammar or readability editing |
Did the journal require disclosure, and was meaning preserved? |
Compare revised text with the original and check the policy. |
| Substantive manuscript drafting |
Were new claims, references, or interpretations introduced? |
Verify affected passages and prepare an accurate contribution statement. |
| Literature review assistance |
Were citations and summaries independently checked? |
Retrieve original sources and verify claim-source alignment. |
| Statistical code generation |
Were the implemented analyses validated? |
Inspect code and independently reproduce consequential results. |
| Qualitative classification |
Did AI influence coding, themes, or interpretation? |
Examine the analytical framework and source data. |
| Participant-data processing |
Was the information permitted to be supplied to the system? |
Assess confidentiality, ethics, and institutional requirements. |
Do not infer the seriousness of the omission solely from the tool's name. The actual activity and its consequences matter more.
Distinguish Three Problems That May Occur Together
Undisclosed AI use can involve different issues requiring different responses.
Reporting omission
Required information about AI assistance was not provided to the journal.
Research reliability concern
AI-generated material may contain errors or lack adequate verification.
Governance concern
AI use may have breached an institutional restriction, confidentiality obligation, or approved research procedure.
These issues should be investigated separately.
Adding a disclosure does not correct inaccurate code. Reanalyzing data does not resolve an unauthorized disclosure of confidential participant information. Likewise, discovering an omitted AI-use statement does not establish that the research findings are incorrect.
A defensible response identifies which problems actually exist and addresses each according to its governing requirements.
Verify Consequential AI-Assisted Work Before Declaring It Reliable
Once the AI contribution is identified, assess whether its outputs were appropriately verified.
For analytical code, compare the implementation with the research design and reproduce consequential calculations. For AI-generated references, retrieve the original publications. For AI-assisted qualitative analysis, compare classifications and interpretations with the authorized source material.
Do not rely exclusively on the same AI system to certify its previous work.
Verification should be proportionate to the consequences of error.
Where the research team lacks sufficient expertise to evaluate a consequential AI-assisted procedure, qualified methodological review may be necessary.
Disclosure is a transparency requirement. It does not establish that the underlying research is scientifically valid.
Should the Corresponding Author Contact the Journal?
When a required disclosure was omitted from a submitted manuscript, the corresponding author should ordinarily communicate with the editorial office or handling editor to correct the submission.
The communication should explain the omitted contribution and request instructions for updating the relevant declaration or manuscript section.
For substantive AI involvement affecting methodology or results, timely communication is particularly important because reviewers may need the information to evaluate the work.
Researchers should not assume that a replacement manuscript can simply be uploaded without editorial authorization.
Nor should they contact anonymous reviewers directly unless the journal explicitly permits it.
If the omission concerns an activity whose disclosure status is uncertain, the authors may ask the editor for clarification.
What Information Should Be Provided to the Editor?
An editorial communication should be factual and sufficiently specific to explain the problem.
Identify the manuscript using its submission identifier and title. Describe the AI contribution, the applicable disclosure requirement, and the correction proposed.
Where relevant, identify the manuscript sections or analyses affected and explain what verification has been performed.
Distinguish established facts from work still under investigation.
For example, do not claim that all AI-generated references were checked unless the authors actually retrieved and examined them.
The journal may request an updated disclosure, revised Methods section, corrected manuscript, or additional supporting information.
Follow its instructions and preserve relevant correspondence.
Can the Disclosure Wait Until the Next Revision?
There is no universal answer independent of journal policy and the significance of the omission.
If the omission is minor and the journal permits administrative corrections during revision, an editor may authorize that approach.
However, researchers should not independently decide to withhold a known material omission when it could influence the ongoing editorial assessment.
For example, if AI generated classifications central to the study's results, reviewers may reasonably need an accurate account of that procedure before completing their evaluation.
The corresponding author should seek editorial guidance rather than assuming that an eventual revision will resolve the problem.
What if the Manuscript Is Still Awaiting Initial Screening?
When the submission remains at an early editorial stage, correcting the disclosure may be procedurally straightforward.
The journal may permit an updated submission declaration or replacement file before the manuscript is assigned to reviewers.
Nevertheless, the authors should follow the journal's established process rather than attempting an undocumented change.
Where the omitted activity raises a substantive research-integrity concern, the team should investigate that concern even if the manuscript has not yet entered peer review.
The timing may affect the administrative response, but it does not remove the need for accurate reporting.
What if Peer Review Has Already Started?
If reviewers are evaluating the manuscript, the editor must determine whether the corrected information should be supplied to them.
Undisclosed AI-generated analytical procedures may require additional methodological detail. An omitted language-editing disclosure may be addressed through a different process.
The journal may ask for revised files, provide supplementary information to reviewers, or defer the correction until an authorized revision.
The research team should not assume that every disclosure correction requires restarting peer review.
Equally, it should not assume that a material methodological omission can be handled without considering its effect on reviewer assessment.
The appropriate decision belongs to the editorial process.
What if the Manuscript Has Been Accepted?
Acceptance does not eliminate the responsibility to correct inaccurate declarations.
If the omission is discovered after acceptance but before publication, contact the journal promptly.
Production workflows may have begun, and the publisher may need to amend the disclosure before the article becomes publicly available.
If the AI contribution also raises concerns about the reliability of results, the journal may need to reconsider whether the manuscript is ready for publication.
Do not assume that substantive changes can be made during routine proof correction without editorial approval.
What if a Coauthor Used AI Without Informing the Team?
Undisclosed AI use by a coauthor introduces an additional collaborative responsibility.
The team should establish what the coauthor did, what outputs entered the manuscript, and whether relevant policies or agreements were breached.
Do not automatically assume deliberate concealment.
However, if consequential AI-assisted material was not communicated to responsible collaborators, the team may need to verify that work before confirming the accuracy of the submission.
The appropriate response to a coauthor's undisclosed AI use should reflect the activity, available evidence, and applicable authorship responsibilities.
What if the AI Contribution Was Scientifically Correct?
Accurate AI-assisted work may still require disclosure.
For example, a researcher may have used AI to generate statistical code that was subsequently reviewed and independently validated.
The analysis may be defensible, but a journal requiring disclosure of AI-assisted programming may still expect the contribution to be reported.
Conversely, disclosing an unverified analysis does not make it scientifically reliable.
The research team should address both accuracy and transparency rather than treating either as a substitute for the other.
What if AI Use Violated an Institutional or Ethics Requirement?
An undisclosed activity may involve more than a missing journal statement.
Suppose identifiable participant transcripts were uploaded to an external AI service that was not approved for the research project.
The team should determine what information was shared, which contractual and technical protections applied, and whether the processing was authorized.
Depending on the circumstances, consultation with an ethics committee, information-security office, or data-protection authority may be required.
Adding a disclosure does not retrospectively authorize prohibited processing.
Institutional incident-response procedures may need to operate alongside editorial communication.
Does an Omitted Disclosure Automatically Constitute Misconduct?
No. An omitted AI-use statement does not automatically establish research misconduct.
Formal definitions differ among jurisdictions and institutions.
Under the United States Public Health Service framework, research misconduct concerns fabrication, falsification, or plagiarism within its applicable scope and requirements. An ordinary reporting omission does not automatically meet those definitions.
However, knowingly false declarations, deliberate concealment, fabricated evidence, or other conduct may raise separate concerns under applicable policies.
Researchers should distinguish an honest misunderstanding from deliberate misrepresentation through a fair assessment of the evidence.
Could the Journal Reject or Withdraw the Manuscript?
Possible editorial outcomes depend on the journal's policies and the seriousness of the problem.
An omitted disclosure that can be corrected may not require withdrawal or rejection.
If the undisclosed activity involved prohibited conduct or introduced substantial uncertainty about the findings, the journal may require further investigation or take other editorial action.
Authors should not assume that voluntary correction guarantees a particular publication outcome.
Nevertheless, knowingly withholding material information can create additional problems beyond the original omission.
The defensible course is to provide accurate information and cooperate with the journal's procedures.
Document the Correction Without Rewriting the Research History
Maintain a record of the original submission, omitted AI contribution, applicable policy, verification performed, and subsequent communication.
Where available, preserve relevant prompts, generated code, document versions, and analytical outputs.
Do not invent model details or verification procedures to make the disclosure appear more complete.
Confidential information should remain protected according to applicable requirements.
Documentation should explain what actually happened, not reconstruct a more favorable version of the process.
Prevent Similar Omissions Before Future Submissions
A research team can reduce disclosure omissions by maintaining a record of relevant AI-assisted tasks throughout the project.
Before submission, the corresponding author should ask contributors to identify AI assistance involving writing, literature review, programming, data processing, and analysis.
The team should compare those contributions with the journal's current requirements and verify consequential outputs.
This process can be incorporated into the ordinary pre-submission review alongside authorship, funding, ethics, and conflict-of-interest declarations.
Clear expectations may be particularly important in collaborative projects where AI assistance occurs across several research activities.
Watch Out
Do not knowingly leave a required AI disclosure incomplete while peer review continues, particularly when the omitted activity materially affects the research methods or results. Correct the reporting issue through the journal and separately investigate any scientific or ethical concerns associated with the AI contribution.
