03 · What You Need to Know
How to Assess and Correct Missing AI Disclosures
First, Determine Whether Disclosure Was Actually Required
Before describing an omission as a policy violation, researchers should establish which requirements governed the work.
Relevant sources may include the target journal's author instructions, institutional research policies, funder conditions, conference submission requirements, and rules applicable to the research activity.
These requirements are not necessarily identical.
A publisher may require disclosure of generative AI assistance in manuscript preparation, while an institution may separately regulate the use of AI to process confidential participant information.
Some policies distinguish ordinary spelling or grammar assistance from substantive AI-generated content. Others require broader reporting of AI involvement.
The International Committee of Medical Journal Editors provides recommendations concerning AI-assisted scholarly work, including disclosure and human author responsibility. These recommendations are influential but should not be treated as automatically binding on every journal or discipline.
Researchers should identify the actual policy that applied to their work, including the relevant submission date and any instructions incorporated into the submission process.
Why Does the Timing of the Policy Matter?
AI disclosure requirements have evolved as generative AI tools have become more widely used in research.
A journal may revise its author instructions between the time a manuscript is submitted and the time it is accepted or published.
Researchers should therefore distinguish requirements that applied when the relevant activity occurred from those introduced later.
A newly adopted policy does not automatically establish that an earlier omission violated the previous policy.
However, the journal may still require an updated disclosure during revision, acceptance, or production.
Where the historical policy cannot be established, researchers should avoid claiming certainty about a past violation. They can explain the circumstances and seek clarification from the responsible editorial authority.
The objective is accurate compliance assessment, not retrospective application of rules without examining their scope.
Identify What the AI Actually Did
A meaningful disclosure should describe the contribution rather than merely state that an AI tool was used.
Researchers should reconstruct the relevant AI-assisted activities using available prompts, outputs, code, document versions, and research records.
For example, AI may have corrected grammar, suggested literature search terms, generated analytical code, classified survey responses, or drafted substantive passages.
These activities differ in their potential effects on scientific content and in the verification they require.
| AI contribution |
Information to establish |
Additional concern |
| Language editing |
Which passages were edited and whether scientific meaning changed |
Compliance with the journal's writing-assistance policy |
| Substantive drafting |
Which sections or claims were generated or substantially rewritten |
Factual accuracy, attribution, and author review |
| Literature assistance |
Whether AI suggested searches, summarized sources, or generated references |
Citation authenticity and claim-source alignment |
| Statistical programming |
Which code or analytical procedures were AI-assisted |
Independent validation and reproducibility |
| Data classification |
Which observations were classified and how outputs were validated |
Classification error, bias, and methodological reporting |
| Participant-data processing |
What information was supplied and how it was handled |
Privacy, consent, security, and ethics requirements |
The table identifies questions worth examining. It does not imply that every listed activity requires the same disclosure under every policy.
Distinguish Missing Disclosure From Inaccurate Research Content
A manuscript may contain accurate AI-assisted material that was not disclosed as required.
Alternatively, undisclosed AI assistance may have introduced fabricated references, incorrect analyses, or misleading statements.
These situations require different corrective actions.
Disclosure omission
Required information about AI involvement was not reported, regardless of whether the resulting content is accurate.
Scientific inaccuracy
The research contains incorrect or unsupported information that requires substantive verification and correction.
Possible policy or integrity breach
The circumstances may involve prohibited AI use, deliberate concealment, mishandling of protected information, or another issue governed by applicable rules.
Adding an AI disclosure does not correct inaccurate statistical results or fabricated citations.
Similarly, verifying that AI-generated content is accurate does not automatically satisfy a disclosure requirement.
Researchers should examine both the transparency of the AI contribution and the reliability of the resulting work.
What Should an Accurate AI Disclosure Contain?
The appropriate wording depends on the journal's requirements and the actual research activity.
A useful disclosure may identify the AI tool or service, available model designation, the tasks performed, the extent of the contribution, and the verification or human oversight undertaken.
Some publishers prescribe a particular statement or manuscript location. Others ask for information in the Methods section, acknowledgments, or submission system.
Researchers should not invent model versions or claim verification procedures that were not performed.
For example, a disclosure should not state that all AI-generated references were independently checked unless that verification actually occurred.
When model information is unavailable, document what can reliably be established and follow the journal's instructions.
A concise, accurate description is preferable to a broad statement that obscures how AI affected the research.
What if the Manuscript Has Not Yet Been Submitted?
When the manuscript remains a working draft, the omission can generally be addressed before submission.
Researchers should identify the relevant AI activities, verify the affected work, and prepare the disclosure required by the intended journal.
Coauthors should have an opportunity to review consequential AI contributions and confirm the accuracy of the final manuscript.
If the AI use involved participant data or institutional restrictions, any substantive compliance concerns should be addressed separately.
The research team should not assume that a disclosure statement retrospectively authorizes an activity that required prior approval.
At this stage, the objective is to ensure that the submission accurately reports both the research and relevant AI assistance.
What if the Manuscript Has Already Been Submitted?
Once the manuscript enters editorial consideration, authors should determine how to correct a missing required disclosure through the journal's procedures.
If the omission concerns material AI assistance that the journal explicitly required authors to report, contacting the editorial office or handling editor is generally appropriate.
The corresponding author should explain what AI did, identify the omitted information, and request guidance on submitting an updated statement.
The journal may permit an administrative update, request revised files, or consider whether additional verification is necessary.
Authors should not silently modify a submitted manuscript outside the journal's authorized process.
Where the omission concerns a minor activity and policy interpretation is uncertain, clarification from the editorial office may be more appropriate than assuming a serious violation occurred.
What if the Manuscript Is Under Peer Review?
Undisclosed AI assistance may be relevant to reviewers when it affects methodological procedures, analytical outputs, or substantive manuscript content.
For example, AI-generated qualitative classifications may require reporting in the Methods section because they form part of how the research evidence was produced.
If such information was omitted, the editor may need to determine whether reviewers should receive an updated methodological description.
Authors should communicate through the journal rather than contacting anonymous reviewers directly.
The appropriate response depends on the AI contribution and whether the omission could materially affect editorial assessment.
Correcting a disclosure should not be treated as an opportunity to rewrite the history of the research process.
What if the Article Has Already Been Published?
A missing AI disclosure discovered after publication may require consultation with the journal about a formal correction or another documented update.
The appropriate mechanism depends on the publisher's policies, the nature of the omitted information, and whether the article's scientific reliability is affected.
For example, an omitted statement about language assistance may be addressed differently from an undisclosed AI classification procedure that materially influenced the study's findings.
The Committee on Publication Ethics provides guidance concerning publication concerns and corrections to the scholarly record.
Researchers should not assume that every missing AI disclosure requires retraction.
However, they should also not assume that adding a statement to a personal website adequately corrects the official publication.
The detailed response to correcting a missing AI disclosure after publication depends on the journal's editorial procedures.
Does Every Omitted AI Disclosure Require Notifying the Journal?
Not necessarily.
The answer depends on whether the relevant activity was covered by the journal's disclosure requirements and whether the manuscript has already entered editorial or published records.
A researcher who used a conventional spelling checker may not have the same obligations as someone who used a generative model to produce substantive analytical content.
Yet the distinction cannot be resolved solely by calling an activity minor. The applicable policy may define disclosure more broadly.
Researchers should check the journal's actual instructions rather than relying on assumptions about what editors would consider important.
Where a required disclosure was omitted from a submitted or published manuscript, the authors should follow the journal's correction or clarification procedures.
How Serious Is the Omission?
The seriousness of missing AI disclosure depends partly on what the AI contributed.
Undisclosed grammar assistance may raise a reporting issue under a policy requiring such disclosure, but it does not necessarily affect the study's scientific findings.
Undisclosed AI-generated code used for the primary analysis may have greater methodological significance, particularly if its implementation and validation were not adequately documented.
Undisclosed processing of identifiable participant information may raise privacy or ethics concerns beyond publication transparency.
The degree of concern should reflect the actual activity, applicable requirements, and consequences.
This is why the seriousness of undisclosed AI use depends on what the AI actually did.
Researchers should avoid treating all omissions as equivalent.
Does Forgetting a Disclosure Automatically Mean Research Misconduct?
No. A missing disclosure does not automatically establish research misconduct.
Formal misconduct definitions vary across jurisdictions and institutions.
Under the United States Public Health Service framework, research misconduct concerns fabrication, falsification, or plagiarism within the scope of the applicable rules and required standards.
A reporting omission does not automatically meet those definitions.
Nevertheless, deliberate concealment, false declarations, or other conduct may raise separate institutional or publication-integrity concerns.
The relevant facts include what the policy required, what the researcher knew, what was declared, and how the omission was handled after discovery.
An honest misunderstanding of a changing policy should not be equated automatically with intentional deception.
Why Does Intent Matter Without Eliminating Corrective Responsibility?
The circumstances of an omission may matter when evaluating possible culpability.
A researcher who misunderstood an ambiguous policy may have acted differently from someone who knowingly denied using AI after being explicitly asked about it.
However, both situations may require correction if the published disclosure is inaccurate.
The distinction between intent and corrective responsibility is important.
Researchers may need to correct an honest mistake even when there is no basis for alleging misconduct.
Conversely, correcting the omission does not automatically resolve a separate concern about deliberate misrepresentation.
Appropriate institutional processes should address such questions when necessary.
What if the Researcher Is Unsure Which AI Model Was Used?
Some AI services do not consistently expose complete model or version information to users.
Researchers should document the information that can be reliably recovered, including the service, approximate or recorded dates, available model designation, and tasks performed.
Relevant chat histories, project records, and document revisions may assist with reconstruction.
Do not invent a version identifier merely to make the disclosure appear complete.
If important details cannot be established, describe the limitation accurately and follow the journal's requirements.
Transparency is better served by acknowledging unavailable information than by supplying false precision.
Should Researchers Recheck the AI-Assisted Work?
Yes, particularly when the undisclosed activity could affect research validity or scientific conclusions.
For example, if AI generated analytical code, verify the code's operations and regenerate consequential results where appropriate.
If AI suggested citations, confirm the publications exist and support the relevant claims.
If AI summarized participant responses, examine whether the summaries preserve the original meaning.
The discovery of a disclosure omission does not establish that the outputs are wrong.
However, it provides an opportunity to confirm that appropriate verification occurred and that the research remains defensible.
Adding a disclosure should not become a substitute for checking the work.
What if the AI Use Was Prohibited Rather Than Merely Undisclosed?
Prohibited AI use is a different issue from failing to report otherwise permitted assistance.
For example, a research team may have entered confidential participant information into an AI service that was not approved for such data.
Adding an AI-use statement would not retrospectively resolve the possible privacy or ethics breach.
The researchers may need to consult an institutional research ethics committee, information-security office, data-protection officer, or other responsible authority.
Applicable reporting obligations depend on the data, agreements, institution, and jurisdiction.
Researchers should not assume that editorial disclosure alone remedies a substantive compliance problem.
Watch Out
Do not treat a missing AI disclosure as merely a wording problem until you have established what AI actually did. If the activity involved unverified analyses, confidential data, prohibited procedures, or inaccurate declarations, those issues require separate assessment and may need additional corrective action.
How Should the Research Team Document the Correction?
A useful record identifies the omitted AI contribution, applicable disclosure requirement, original reporting status, verification performed, and corrective communication.
Where available, retain relevant model information, prompts, outputs, manuscript versions, and editorial correspondence.
Documentation should be proportionate and should protect confidential research materials.
If the omission affects a major research decision, record how the AI contribution influenced the workflow and how the resulting decision was independently justified.
These records can help distinguish a reporting oversight from a more consequential problem.
They also support consistency between internal research documentation and public statements about AI use.
How Can Researchers Prevent Missing Disclosures?
AI disclosure is easier to manage when the research team records relevant AI assistance as it occurs.
Researchers can maintain a task-level log identifying the tool, activity, date, responsible contributor, and verification procedure.
Before submission, the corresponding author can compare this record with the journal's current instructions.
Coauthors should be asked to identify relevant AI assistance they used, including contributions to writing, coding, data processing, or analysis.
A final disclosure review should be completed alongside checks of authorship, funding, conflicts of interest, ethics statements, and data availability.
This approach may reduce omissions without requiring publication of every AI conversation.
