Reviewer Guidelines: What to Look For, How to Structure Comments, and Common Mistakes

This guide is about the mechanics of writing a peer review once you’ve agreed to do one: what to actually evaluate as you read a manuscript, how to organize your comments so they’re useful to an editor and workable for an author, and the recurring mistakes that make an otherwise sound review harder to act on. It assumes you already know what peer review is for and how the editorial process works — see Peer Review: What It Is, How It Works, and Why It Matters for that background, and Peer Reviewer: Role, Responsibilities, and How to Become One for how reviewers are selected, recognized, and held to COPE’s ethical obligations. This page picks up from there — the practical craft of producing the report itself.

Before you start writing

Two checks belong before you open a blank document. First, confirm you can actually do the review in the window the journal asked for. The Committee on Publication Ethics’s Ethical Guidelines for Peer Reviewers (v2, September 2017) treats declining a review you can’t complete on time as more constructive than accepting and delivering late — a late review is not fair to the author or the editorial office. The same point is rule one of the widely-cited PLOS Computational Biology piece “Ten Simple Rules for Reviewers” (Bourne & Korngreen, 2006): decline rather than sit on a manuscript.

Second, read the journal’s own reviewer instructions before you read the manuscript. Journals differ on what they want in a review — some want a structured form with numbered fields, some want free text; some want you to recommend a decision, others reserve that explicitly for the editor. ICMJE’s Responsibilities in the Submission and Peer-Review Process is explicit that editors are not bound by reviewer recommendations (“a peer-reviewed journal is under no obligation to follow reviewer recommendations”) — your job is to give the editor the information needed to decide, not to render the decision yourself. Check, too, whether the manuscript is under double-anonymized or single-anonymized review, since that affects what you can reference in your comments (a single-anonymized reviewer referencing an author’s prior identifiable work is fine; the same move under double-anonymization can compromise the process).

What to actually look for, section by section

A review that just says “methods are sound, results are interesting” gives an editor nothing to act on. Work through the manuscript with specific questions in mind for each section:

• Significance and novelty. Does the introduction accurately represent what’s already known, and does the study add something to it? Overclaiming novelty is one of the more common problems reviewers are asked to catch — check whether the cited prior work actually supports the gap the authors claim to be filling.
• Methods. Is there enough detail for someone else to replicate the work? Are the methods appropriate to the question being asked, not just internally consistent? For clinical and health research specifically, check whether the manuscript follows the relevant reporting guideline for its study design — CONSORT 2010 for randomized trials, or the equivalent for the design in question — and flag missing required elements rather than treating the checklist as optional.
• Results and statistics. Do the reported results match what the methods described? Are statistical tests appropriate for the data type and sample size, and are effect sizes reported alongside p-values rather than p-values alone? If statistical review is outside your expertise, say so explicitly in your comments to the editor rather than passing silently over a section you can’t actually evaluate — that’s a documented gap, not a weakness in your review.
• Discussion and limitations. Do the authors’ conclusions follow from what the data actually show, or do they overreach? Is there a genuine limitations section, or has it been reduced to boilerplate?
• Ethics, data, and disclosures. Is there an ethics approval statement where one is needed, and does it match the study design? Are competing interests disclosed? Is the underlying data available or its availability explained? These are frequently the fastest things to check and the easiest to miss if you go straight to the science.

How to structure the review report

Most journals expect some version of the same three-part shape, whether or not their submission form labels it this way:

1. A short summary. One paragraph, in your own words, describing what the manuscript reports and its overall contribution. This does double duty: it tells the editor you understood the paper, and it tells the authors how a reader actually interpreted their work — often a valuable data point on its own, since a misunderstood point is sometimes as much a clarity problem as a reviewer error.
2. Major comments. Substantive issues that affect whether the paper’s conclusions are supported — methodological gaps, missing controls, claims the data don’t support, a reporting-guideline element that’s absent. Number them. Reference specific line numbers, figures, or table numbers rather than describing a problem in the abstract; “the statistics section needs work” is not actionable, “the effect size reported in Table 2 does not match the raw values in Table 1” is.
3. Minor comments. Typos, unclear phrasing, small formatting issues, suggestions that would improve but not fix the paper. Keep these separate from major comments so an editor and author can immediately see which issues gate acceptance and which don’t.

Most systems also give you a field for confidential comments to the editor, not shown to the authors. Use it for exactly what it’s for: context the editor needs but the authors don’t — a disclosed conflict of interest you want on record, a suspicion of duplicate or overlapping publication you’re flagging rather than personally investigating, or your honest confidence level in a section outside your core expertise. COPE’s guidelines are explicit that this channel is not a venue for unfounded accusations, and that whatever you tell the editor confidentially should be consistent with, not contradictory to, what you told the authors directly — a private aside that undercuts your own public comments is a documented form of reviewer misconduct, not a normal use of the field.

Tone: writing comments you’d want to receive

Rule three of Bourne and Korngreen’s “Ten Simple Rules for Reviewers” is blunt about the standard: write the review you would be satisfied with as an author. A 2026 companion piece in the same PLOS Computational Biology series, “Ten simple rules for writing a peer review” (Rahman & DeBlasio, 2026), frames the same idea slightly differently: be the reviewer you want to get a review from, avoid dismissive or sarcastic language, and stay alert to implicit bias toward an author’s institution, country, gender, or subfield — bias that’s easiest to introduce unconsciously in phrasing, not just in the substantive judgment. In practice this means: address the manuscript, not the authors (“the analysis in Section 3 doesn’t account for X,” not “the authors clearly didn’t consider X”); be decisive rather than hedging every comment into vagueness; and separate a genuine flaw from a stylistic preference you’d have written differently yourself.

Common reviewer mistakes

• Comments without a location. A comment that doesn’t say where in the manuscript the problem is forces the author to guess, and often forces the editor to ask you to clarify before the review is even usable.
• Rewriting instead of reviewing. Suggesting a full alternative framing or prescribing exact replacement text crosses from evaluation into ghost-authorship of someone else’s paper. Identify the problem and let the authors solve it.
• Suggesting your own citations without a genuine academic reason. COPE names this specifically as a form of reviewer misconduct, distinct from plagiarism or fabrication — a citation suggestion should be there because it strengthens the manuscript, not because it raises the reviewer’s own citation count.
• Treating the confidential-to-editor field as a separate, harsher review. If a concern is serious enough to raise privately, it belongs in the comments the authors see too, framed constructively — not held back and only surfaced to the editor.
• Silence on sections outside your expertise. Skipping a section without saying so reads, to an editor, as though you evaluated it and found nothing wrong. Say explicitly what you did and didn’t assess.
• Accepting the assignment, then going quiet. Missing the deadline without flagging it early is worse for the author than declining up front would have been — both COPE’s guidelines and Bourne and Korngreen’s rule one treat this as the most basic obligation a reviewer takes on.
• Letting recommendation stand in for reasoning. “Reject” or “major revision” with no comments to support it gives the editor a vote with no evidence behind it, and gives the author nothing to act on even if a resubmission is invited.

Common questions

Should I write my review as a numbered list or as prose?

Follow the journal’s own submission form where it dictates a structure. Where it doesn’t, a numbered list for major and minor comments (see the structure above) is easier for an editor to weigh and an author to respond to point-by-point than a single block of prose, even though the opening summary paragraph works better as prose.

How long should a peer review be?

Long enough to justify the recommendation with specific, referenced evidence, and no longer. A one-paragraph review recommending rejection of a complex manuscript rarely gives an editor enough to act on; padding a review with restated summary rather than substantive comment doesn’t add value either.

What if I disagree with a co-reviewer or the editor’s eventual decision?

That’s a normal, expected outcome — ICMJE is explicit that editors are not obligated to follow reviewer recommendations, and reviewers routinely disagree with each other on the same manuscript. Your job is to make your own reasoning clear and evidence-based, not to guess what other reviewers will say or defend a recommendation past the point your evidence supports it.

Can I use AI tools to help draft my review?

Check the journal’s policy before you do — publishers have moved quickly on this specifically because reviewer confidentiality is at stake: pasting an unpublished manuscript into a third-party AI tool can violate the confidentiality obligation COPE and ICMJE both place on reviewers, independent of whatever the tool’s own output policy says. See AI in peer review for how this is currently being handled across major publishers.

Does my review get any formal recognition?

Not by default. A completed review only becomes a citable, verifiable credit if you or the journal push it to a service designed for that — see Peer reviewer contribution for how that recognition actually gets recorded.

Related CASRAI resources

• Peer Review: What It Is, How It Works, and Why It Matters — the process overview this guide assumes.
• Peer Reviewer: Role, Responsibilities, and How to Become One — becoming a reviewer, COPE’s ethical obligations, and how contributions are recognized.
• Post-Publication Peer Review: From Concern to Correction — what happens when a concern is raised after publication.
• COPE Guidelines Explained — the wider set of COPE Core Practices peer review guidance sits within.
• Single-blind vs double-blind vs open peer review — how anonymization affects what a reviewer can and can’t say.
• Reviewer role typology and Journal Reviewer — related dictionary terms.
• Conflict of Interest (COI) in Research — what reviewers are required to disclose before accepting an assignment.
• Author Guidelines: What Journals Require and How to Meet Them — the author-facing counterpart to this page.