Artificial Intelligence–Assisted Peer Review Policy

1. Purpose

This policy governs the use of artificial intelligence (AI) tools in the editorial and peer review process of Macula: The Journal of Macular Diseases ("the Journal"). It defines what AI assistance may and may not do, how it is disclosed, and how it is reconciled with human editorial judgment. It applies to every manuscript submitted to the Journal from the effective date above.

This policy is written to be consistent with the current guidance of the Committee on Publication Ethics (COPE) on authorship and AI tools, COPE's discussion document on AI in editorial decision-making, and the position adopted by major medical publishers that AI may assist review but must never determine its outcome.

2. Scope

This policy covers the use of AI in the review of submitted manuscripts. It does not cover, and is addressed separately from:

• Authors' use of AI in drafting or analyzing their own manuscripts (see the Journal's Author Guidelines, which require disclosure of any generative AI use in the Methods or Acknowledgments section);
• Administrative or production uses of AI unrelated to scientific judgment (e.g., formatting checks, reference validation).

3. Guiding Principles

1. Human accountability is non-negotiable. Every editorial decision is made and signed by a named human editor. No AI tool is authorized to accept, reject, or request revision of a manuscript.
2. AI is an assistant to reviewers and editors, not a replacement for them. Its output is one input among several, weighed by people with clinical and methodological expertise in macular and retinal disease.
3. Transparency to authors and readers. Authors are informed that AI-assisted review is part of the process at submission. The consolidated review report is published alongside the accepted article.
4. Confidentiality of unpublished work is preserved, whether a manuscript is read by a human or processed by an AI tool.
5. Tone is an editorial responsibility, not an AI feature. Non-constructive or pejorative language is removed from any review — human or AI-generated — before it reaches an author or is published. The substance of a criticism is preserved; the manner in which it was expressed is not.

4. Where AI Is Used in the Review Workflow

AI-assisted review (currently performed using Claude, provided by Anthropic) is applied at a defined stage, after independent human review has begun:

Function	Description
Methodological consistency check	Flags internal inconsistencies between stated methods, reported results, and figures/tables.
Reporting-guideline adherence	Checks conformance to the applicable standard (CONSORT, STROBE, CARE, PRISMA) and flags omissions.
Statistical clarity check	Flags statistical claims that are ambiguous or inconsistent with the data presented, for a human statistical editor to evaluate.
Prior-literature contextualization	Surfaces closely related published work for the editor's consideration — not a novelty or plagiarism determination.

The AI's output is delivered only to the handling editor, never directly to authors, and never shown to human reviewers before they submit their own independent assessment — this sequencing prevents the AI's read from anchoring the human review.

5. Where AI Is Not Used

To be explicit about the boundary, AI is not used to:

• Determine or recommend an accept / reject / revise decision;
• Assess a manuscript's novelty, clinical significance, or fit with the Journal's scope;
• Communicate directly with authors at any stage;
• Screen or select peer reviewers, or adjudicate disagreements between them;
• Serve as one of the two or more required independent reviewers.

Consistent with COPE's position on authorship, the AI tool is never listed as an author, reviewer, or contributor, and holds no accountability for a review's content — accountability rests with the named editor who approves the final report.

6. Confidentiality and Data Handling

• Unpublished manuscripts processed through the AI tool are handled under terms that preclude use for model training and limit retention to the review cycle, consistent with obligations placed on human reviewers.
• Identifying information about authors and reviewers is minimized in what is submitted to the AI tool wherever the check being performed does not require it.
• Any change to the AI vendor, model, or data-handling terms materially affecting confidentiality will be disclosed on this page and communicated to the Editorial Board before implementation.

7. Editorial Discretion and Report Consolidation

The handling editor is responsible for reviewing AI-generated notes alongside the independent human reviews; deciding which AI-flagged items are substantive enough to raise with authors; rewriting or removing any dismissive or non-constructive language from either source while preserving the underlying scientific point; and producing one consolidated report, sent to authors and, upon acceptance, published as part of the Journal's open review record.

8. Limitations and Safeguards

AI tools can produce incomplete, overconfident, or mistaken assessments, and are not a substitute for domain expertise in retinal and macular disease. Accordingly, AI-flagged items are treated as prompts for human verification, not findings of fact; reviewers and editors are instructed to disregard any AI note that is not independently verifiable against the manuscript; and the Journal periodically audits a sample of AI-assisted reviews against final editorial outcomes, suspending any function found to be unreliable, pending correction.

9. Disclosure to Authors and Readers

• The submission system informs authors, before submission, that manuscripts will undergo AI-assisted review as described in this policy.
• Each published article is accompanied by its consolidated review report, noting in general terms where AI-assisted analysis contributed.
• This policy is published on the Journal's website and linked from the Peer Review and Author Guidelines pages.

10. Appeals and Author Recourse

Authors who believe an AI-derived observation in their review report is factually incorrect may request reconsideration in writing to the handling editor. Such requests are evaluated by a human editor — and, where relevant, an independent additional reviewer — without reference back to the AI tool, and are resolved on the scientific merits.

11. Policy Governance

This policy is approved and owned by the Editorial Board of Macula: The Journal of Macular Diseases. It will be reviewed at least annually and upon any material change to COPE, ICMJE, or DOAJ guidance on AI in scholarly publishing, or to the Journal's review workflow or AI vendor.

Questions about this policy may be directed to the editorial office at maculajournal.editor@gmail.com.