What Should You Do if an AI Error Is Discovered After Manuscript Submission?

What Should You Do When You Discover an AI Error After Submitting Your Research Paper?

You have submitted a manuscript to a scholarly journal. The submission has passed the initial technical checks and may already be undergoing peer review.

While preparing another research project, you discover that a generative AI system introduced an error into the submitted manuscript.

Perhaps a citation does not exist. An AI-generated statistical script used the wrong variable. A paragraph incorrectly describes the study's findings, or a participant quotation was altered during AI-assisted editing.

The manuscript is no longer simply a private working document. Editors and reviewers may already be evaluating the research based on the inaccurate information.

Should you wait until the journal requests revisions? Should you contact the editor immediately? Could notifying the journal jeopardize publication?

The appropriate response depends on what is wrong, whether the error affects the reliability of the work, and how far the manuscript has progressed through the editorial process.

Verify the Error and Notify the Journal Promptly When It Is Material

How to Handle AI-Generated Errors During Journal Review

Why Does Discovery After Submission Change the Researcher's Responsibilities?

Before submission, researchers can ordinarily correct drafting errors through their internal quality-assurance procedures.

After submission, the manuscript has entered an editorial process in which others may rely on the accuracy of the submitted material.

Editors assess whether the work meets the journal's requirements. Reviewers evaluate methodological quality, evidentiary support, and scientific contribution. Editorial decisions may be influenced by information that the authors subsequently discover to be inaccurate.

This creates a responsibility to consider whether the journal needs updated information to evaluate the manuscript fairly.

Submission does not transfer responsibility for accuracy from authors to editors or reviewers.

The International Committee of Medical Journal Editors emphasizes that human authors remain accountable for the accuracy and integrity of AI-assisted scholarly content under its recommendations.

Specific journal procedures differ, but authors should not assume that an error becomes the editor's responsibility merely because the manuscript has been submitted.

First, Establish Whether the Suspected Error Is Real

Before contacting the journal with a substantive correction, identify the precise problem and verify it using appropriate original evidence.

If AI generated a suspicious citation, examine publisher records, bibliographic databases, and the actual publication where available.

If AI-generated code produced an unexpected result, inspect the analytical procedure and reproduce the relevant calculations.

If the problem concerns an inaccurate description of the methodology, compare the manuscript with the approved protocol, research records, and actual procedures.

For participant quotations, return to the original authorized transcript or recording.

Verification should be timely but sufficiently rigorous to distinguish an actual error from a misunderstanding or legitimate methodological disagreement.

Asking the same AI system to confirm its earlier mistake does not constitute independent verification.

Assess the Error's Scientific Materiality

Not every error discovered after submission requires the same response.

Materiality concerns whether the problem meaningfully affects the manuscript's accuracy, interpretation, scientific reliability, or compliance with applicable requirements.

This table is a practical guide, not a universal editorial policy. Journals may impose specific reporting requirements regardless of whether authors regard an error as minor.

Do Not Assume That an AI Error Is Automatically Minor

The apparent size of a mistake does not establish its scientific importance.

A single incorrect symbol in a statistical formula can change a computation. One misclassified primary outcome can affect an analysis, depending on the study design and dataset.

Conversely, a conspicuous wording error may have limited consequences if the underlying results and interpretation remain accurate.

Researchers should examine what the error affects rather than judging seriousness by the number of incorrect words or lines.

The relevant questions include whether the error changes reported findings, misrepresents evidence, affects participant protections, or alters the conclusions editors are being asked to evaluate.

Inform the Coauthors Before or Alongside Editorial Communication

Consequential corrections should ordinarily involve the responsible research team.

The corresponding author should coordinate communication with the journal, while relevant coauthors should participate in verifying and correcting material within their areas of responsibility.

For example, a statistical error may require review by the analyst and principal investigator. A misrepresented participant quotation may require consultation with researchers responsible for qualitative analysis.

Authors should establish the facts and avoid attributing misconduct or negligence before the circumstances have been examined.

If urgent risks or editorial deadlines make full consultation impossible, the corresponding author may need to notify the journal that an investigation is underway while continuing to coordinate with coauthors.

Authorship agreements and institutional procedures may determine how disagreements are handled.

When Should the Journal Be Notified Immediately?

Prompt notification is generally warranted when a verified error could materially influence editorial assessment or the reliability of the submitted manuscript.

Examples include incorrect primary analyses, fabricated evidence supporting central claims, inaccurate ethics declarations, or errors that substantially change the study's conclusions.

Researchers should not wait for peer reviewers to discover such problems.

Nor should they conceal a material error because disclosure might delay publication.

Editorial communication should describe what has been verified, which sections are affected, whether corrected analyses are available, and what further work remains.

If the investigation is incomplete but the potential consequences are serious, the authors can explain the uncertainty and request guidance on whether editorial consideration should be paused.

How Should Authors Communicate the Error to the Editor?

The communication should be factual, concise, and sufficiently detailed to support an appropriate editorial decision.

Identify the manuscript using its official submission identifier, title, and corresponding author information.

Describe the verified error, affected sections, potential consequences, and corrective action undertaken or proposed.

Where relevant, explain that AI assistance contributed to the problem and identify the nature of that contribution.

Do not claim that the corrected findings are reliable until appropriate verification has been completed.

The editor may request revised files, supporting information, additional explanation, or a different procedural response.

Authors should follow the journal's submission and communication requirements rather than assuming that uploading a replacement file without explanation is sufficient.

Should Authors Mention That AI Caused the Error?

The origin of the error may be relevant to understanding the workflow and determining whether disclosure requirements were satisfied.

For example, if AI-generated code caused incorrect results, identifying that contribution may help explain which analytical procedures require additional verification.

If AI-generated writing introduced a false methodological statement, the authors should establish whether the tool's use was already disclosed and whether the journal requires additional information.

However, attributing the error to AI does not reduce the need to correct the manuscript.

The ICMJE's AI guidance emphasizes human responsibility for scholarly content, while journal-specific disclosure requirements may differ.

Researchers should accurately describe AI's contribution without using it as a substitute for accountability.

What if the Error Is Discovered While Peer Review Is Underway?

When reviewers are already evaluating the manuscript, a material correction may affect the validity of their assessments.

The editor may need to provide corrected information to reviewers, request additional review, or pause the process while the authors complete verification.

Authors should not contact anonymous reviewers directly unless the journal explicitly authorizes that communication.

For a minor issue, the editor may permit correction during the next revision.

For an error affecting primary results, waiting until the next revision could allow the review to proceed on materially inaccurate information.

The appropriate timing should therefore reflect the seriousness of the error and the journal's instructions.

What if the Manuscript Has Been Accepted but Not Published?

Acceptance does not eliminate the authors' responsibility to correct verified errors.

If a material problem is discovered after acceptance, the authors should promptly contact the journal before publication proceeds.

The editorial team may need to reconsider acceptance, request corrected analyses, delay production, or determine whether the manuscript remains publishable.

Authors should not assume that errors can safely be corrected during proof review.

Proof corrections are generally intended for limited production-related changes, and journals may restrict substantive revisions at that stage.

A significant analytical or interpretive correction requires explicit editorial coordination.

When Is Withdrawal Appropriate?

Withdrawal may be appropriate when the submitted manuscript cannot be made scientifically reliable within the current editorial process.

For example, a coding error may invalidate the primary analysis, and the necessary reanalysis may require substantial reconstruction of the study.

Alternatively, fabricated references may have materially distorted the literature review and research rationale.

Withdrawal is not automatically necessary whenever AI contributes to an error.

If the problem can be corrected transparently and the manuscript remains defensible, the journal may permit revision.

The decision should be discussed with the editor and responsible coauthors, considering the journal's policies and the consequences for the research record.

Authors should distinguish a formal withdrawal request from simply abandoning communication with the journal.

What if Corrected Results Change the Main Conclusion?

Changes to the primary conclusion deserve particular attention.

Suppose corrected AI-generated code produces a substantially smaller effect estimate than originally reported.

The researchers should evaluate the revised estimate, uncertainty, and substantive interpretation rather than focusing only on whether statistical significance changed.

If the original conclusion is no longer supported, the manuscript must be revised accordingly.

Editors should receive enough information to understand the consequences for the study's scientific contribution.

The goal is not to preserve the manuscript's original narrative, but to ensure that the work under consideration accurately represents the corrected evidence.

What if the Error Concerns a Fabricated Citation?

A nonexistent AI-generated reference may be an isolated bibliographic problem or evidence of a broader weakness in the literature review.

Researchers should verify the citation, examine the claim it was supposed to support, and audit other references produced through the same unverified workflow.

If the reference supports a central argument, removing it without revising the underlying claim may leave the manuscript scientifically misleading.

The appropriate process for correcting an AI-fabricated citation includes both bibliographic verification and examination of claim-source alignment.

If the manuscript has already been submitted, the significance of the correction determines whether prompt editorial notification is necessary.

What if AI-Generated Code Produced the Incorrect Results?

Correcting a computational error requires more than replacing reported numbers.

Researchers should preserve the original data and code, identify the faulty operation, validate the correction, and regenerate affected outputs.

Where feasible, consequential results should be independently reproduced.

The revised manuscript should accurately report the corrected analyses and explain any material changes.

The process for correcting AI-generated code that produced incorrect research results may reveal that multiple tables, figures, or interpretations require revision.

Editors should be informed when those changes materially affect the submitted work.

Preserve an Audit Trail of the Correction

Researchers should retain appropriate records of the original error, verification process, corrected material, and communications with the journal.

Relevant documentation may include original and revised manuscript versions, analytical scripts, verified outputs, and coauthor decisions.

Where AI materially contributed to the error, preserve relevant tool information and interactions as required and permitted.

Records containing confidential participant data or unpublished material should be protected according to applicable institutional and contractual requirements.

A useful audit trail makes it possible to explain what changed, why it changed, and whether the corrected manuscript remains scientifically defensible.

Does Discovering an AI Error Mean Research Misconduct Occurred?

No. A verified error does not automatically establish research misconduct.

Formal definitions and procedures 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. Honest errors are not automatically misconduct.

However, deliberate concealment, knowingly false reporting, or other serious conduct may raise distinct integrity concerns.

The appropriate response is to establish the facts, correct the research, and follow relevant institutional procedures when misconduct concerns arise.

The presence of AI does not independently determine intent, culpability, or the appropriate editorial outcome.

How Can Researchers Reduce the Risk of Post-Submission AI Errors?

Pre-submission quality assurance should examine consequential AI contributions before the manuscript enters peer review.

Useful procedures include verifying citations against original publications, testing AI-generated analytical code, comparing rewritten passages with source material, and checking claims about methods and ethics compliance.

Coauthors should review material relevant to their responsibilities.

A final consistency check can compare the abstract, results, tables, figures, discussion, and conclusions against verified outputs.

Such procedures cannot guarantee an error-free manuscript, but they may reduce the likelihood that AI-generated inaccuracies remain undetected until editorial review.

Correcting an AI-Generated Statistical Error During Peer Review

The error is material because it changes the analysis of the primary outcome and affects the manuscript's central conclusion.

Waiting until reviewers request revisions would allow them to continue evaluating findings the authors already know are inaccurate.

The appropriate response is to verify the error, correct the evidence, and communicate promptly through the journal's established process.

The fact that AI originally generated the code explains part of the workflow but does not remove the authors' responsibility for the submitted analysis.

Common Misconceptions About AI Errors Discovered After Submission

A Practical Decision Framework for Post-Submission AI Errors

The appropriate response begins with independent verification and an assessment of materiality.

Then determine whether the journal needs updated information immediately or whether the correction can be handled through an authorized routine revision.

The key question is whether the editor and reviewers are evaluating an accurate representation of the research.

Authors should act promptly when the answer is no, while ensuring that proposed corrections are independently verified and appropriately documented.

What to Check After Finding an AI Error in a Submitted Manuscript

Frequently Asked Questions About AI Errors After Manuscript Submission

Correct Material Errors Before They Become Part of the Published Record

How to Cite This Guide

This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.