03 · What You Need to Know
What AI Disclosure Means in the Peer-Review Process
Why Is AI Disclosure Relevant to Peer Review?
Peer review is a confidential scholarly assessment conducted on behalf of a journal. Editors invite reviewers because their expertise is relevant to evaluating the submitted research.
When generative AI contributes to a review, the editor may reasonably need to understand how the technology was used and whether the process complied with the journal's requirements.
For example, using AI to improve sentence structure is different from asking it to identify methodological weaknesses or recommend rejection.
Both activities may involve generative AI, but their implications for independent judgment, confidentiality, and accountability differ.
Disclosure allows the journal to evaluate the nature of the assistance rather than assuming that every AI-assisted review involved the same procedure.
What Do the ICMJE Recommendations Say?
The ICMJE's recommendations on artificial intelligence in scholarly publishing address AI use by peer reviewers.
They advise reviewers to follow the journal's stated policy or request permission before using AI to facilitate the review. Reviewers should disclose their AI use to the journal and remain responsible for the accuracy and validity of the resulting content.
The recommendations also emphasize that submitted manuscripts are privileged communications and that AI use must not compromise their confidentiality.
These recommendations are influential, particularly in biomedical publishing, but they are not automatically binding on every journal. Individual publishers and journals may establish more specific or restrictive requirements.
Reviewers should consult the current ICMJE guidance on AI use by reviewers alongside the instructions governing their particular assignment.
Does Every Type of AI Assistance Require the Same Disclosure?
Not necessarily. The appropriate disclosure depends on the journal's policy and the nature of the assistance.
Some journals may distinguish routine language support from substantive AI involvement. Others may require disclosure of any generative AI use associated with the review.
| AI Activity |
Potential Disclosure Issue |
What to Verify |
| Grammar and language editing |
May be covered by general AI disclosure rules |
Whether the journal permits and requires reporting of language assistance |
| Organizing reviewer comments |
May influence the presentation and prioritization of concerns |
Whether the activity is authorized and how it should be described |
| Explaining general methodological concepts |
May support the reviewer's technical understanding |
Whether such assistance falls within the journal's policy |
| Suggesting manuscript weaknesses |
Contributes to the substantive evaluation |
Whether permission and specific disclosure are required |
| Generating a draft peer review |
May involve substantial delegation of the assessment |
Whether the activity is prohibited or explicitly authorized |
| Processing manuscript content |
Raises separate confidentiality and data-handling concerns |
Whether the specific tool and disclosure of content were authorized |
These distinctions are practical categories, not a universal disclosure standard. The journal's actual instructions determine what is required.
Disclosure and Permission Are Two Different Requirements
One of the most important distinctions is between obtaining permission to use AI and disclosing that AI was used.
Permission concerns whether an activity is authorized before it occurs. Disclosure concerns communicating the involvement of AI to the journal.
A reviewer cannot assume that acknowledging AI use afterward makes an otherwise prohibited activity permissible.
Permission
Establishes whether the reviewer may use the AI tool for the proposed purpose under the journal's rules.
Disclosure
Informs the journal about the nature and extent of AI assistance that occurred.
Accountability
Requires the reviewer to verify and take responsibility for the final scholarly assessment.
These obligations complement one another. Satisfying one does not automatically satisfy the others.
What Information Should an AI Disclosure Contain?
Where disclosure is required, a useful statement should accurately describe the role of AI in preparing the review.
Depending on the journal's instructions, relevant information may include the tool or model used, the purpose of the assistance, whether it affected substantive evaluation, and whether confidential manuscript material was processed in an authorized environment.
Reviewers should avoid vague statements such as "AI was used" when the journal requires more specific information.
They should also avoid overstating the independence of the review if AI materially influenced the criticisms or recommendations.
The disclosure should be proportionate to the activity and follow any journal-provided format.
Where Should the Disclosure Be Submitted?
Some editorial systems may provide a dedicated declaration field or questionnaire. Other journals may instruct reviewers to communicate AI assistance directly to the handling editor.
Reviewers should follow the specified process rather than assume that disclosure belongs in the comments intended for authors.
A confidential message to the editor may be appropriate when the journal requests disclosure but does not provide a designated field.
The distinction matters because reviewer reports often contain separate communications for authors and editors.
Disclosure should reach the appropriate editorial authority without unnecessarily revealing confidential information.
Does AI-Assisted Grammar Editing Need to Be Disclosed?
The answer depends on the journal's policy.
Some policies may permit limited language editing while still requiring acknowledgment of generative AI assistance. Others may exempt certain routine editorial tools or define different requirements for generative and conventional software.
Reviewers should not assume that grammar editing is automatically exempt because it does not involve scientific judgment.
Furthermore, even language editing can create confidentiality concerns if reviewer comments contain unpublished manuscript information.
The conditions governing AI-assisted improvement of peer-review comments should therefore be checked before the text is submitted to a tool.
What If AI Influenced the Scientific Criticisms?
Substantive AI involvement deserves particular attention because it may influence the assessment provided to the editor.
For example, a model might suggest checking for confounding, identify an apparent inconsistency in a statistical table, or recommend a different analytical approach.
If the reviewer subsequently investigates and incorporates such concerns, the AI has contributed to the development of the assessment, even if the final wording is entirely human-written.
Where the journal permits this assistance, disclosure should accurately reflect its substantive role.
Independent verification remains essential. A reviewer should not include AI-generated criticisms they cannot understand or defend.
Does Disclosure Resolve Manuscript Confidentiality Concerns?
No.
Suppose a reviewer uploads an unpublished manuscript to a general-purpose AI service without authorization and subsequently informs the editor that AI was used.
The disclosure does not reverse the transfer of confidential information or establish that the processing was permitted.
Depending on the journal's policy and the circumstances, the reviewer may need to report a potential confidentiality incident separately.
Reviewers should not assume that a provider's no-training policy, private interface, or deletion option resolves the issue.
The requirements discussed in AI use and unpublished manuscript confidentiality remain applicable regardless of whether the reviewer makes a disclosure.
Can Reviewers Use AI Without Disclosing the Manuscript?
Potentially, where permitted by the journal.
A reviewer might ask a general question about a statistical assumption or request a generic structure for constructive feedback without sharing manuscript-specific information.
Such assistance may avoid certain confidentiality risks, but it can still fall within the journal's AI-use policy.
The reviewer should distinguish permission to use AI generally from permission to process confidential manuscript content.
Neither should be inferred solely from the absence of an explicit prohibition.
What If the Journal Has No Published AI Policy?
A silent policy creates uncertainty rather than unrestricted permission.
The ICMJE recommends requesting permission when a journal has not established an applicable AI policy.
Reviewers may contact the handling editor to describe the intended assistance and ask whether it is permitted and how it should be disclosed.
For example, a reviewer could ask whether an approved tool may be used solely to improve the grammar of independently written comments without disclosing manuscript information.
If the intended activity requires confidential material and authorization remains unclear, the reviewer should not proceed with that processing.
Should Authors Be Told That AI Assisted the Review?
This is distinct from disclosure to the journal.
The journal may require reviewers to declare AI assistance to the editor without necessarily including that information in the report shared with authors.
Whether authors are informed depends on the journal's policies and editorial transparency arrangements.
Reviewers should not independently decide to disclose confidential editorial information beyond the authorized process.
The immediate responsibility is to comply with the journal's instructions and provide an accurate account of the assistance to the appropriate editorial authority.
Why Disclosure Does Not Reduce the Reviewer's Responsibility
Disclosure tells the journal that AI was involved. It does not establish that the generated content was correct or that the reviewer's scientific assessment was adequate.
Reviewers remain responsible for identifying the evidence supporting their criticisms, distinguishing major weaknesses from minor concerns, and avoiding irrelevant methodological demands.
A declaration that AI was used cannot substitute for these responsibilities.
Ultimately, the question of which parts of peer review require human judgment remains separate from the administrative requirement to disclose technological assistance.
