How to Peer Review a Research Paper

To peer review a research paper well, assess whether the question matters, the methods can answer it, the results are reported faithfully, and the conclusions stay within the evidence. Then translate that assessment into comments an editor can act on and authors can use.

The checklist below applies broadly to empirical manuscripts. Adapt the technical questions to the study design and follow the journal’s instructions. The COPE ethical guidelines for peer reviewers are a useful baseline for confidentiality, competing interests, expertise, and accountability.

Before accepting the review

Do three checks before opening a detailed review file:

1. Expertise: Can you assess the central method and claims? State any important boundary in your expertise.
2. Competing interests: Could a personal, financial, intellectual, or professional relationship impair—or appear to impair—your judgment?
3. Capacity: Can you deliver a careful review within the requested time?

Declining promptly is better than accepting a review you cannot complete fairly. Never use confidential manuscript content for your own work, and do not upload it to an external system unless the journal explicitly permits that use.

First pass: identify the manuscript’s claim

Read the title, abstract, figures, and conclusion to answer four questions:

• What is the primary research question?
• What design is used to answer it?
• What is the main claimed contribution?
• Which result carries that claim?

Write the answers in your own words. If the manuscript’s objective remains unclear, that is a review finding. A focused PICO question can help parse intervention and clinical manuscripts.

Do not draft detailed comments on the first pass. Note potential concerns, then test them against the full methods and results.

Review the methods before judging the conclusion

The methods define the ceiling on the manuscript’s claims. Work through these domains.

Study design and protocol

• Is the design appropriate for the question?
• Was a protocol or registration available before analysis?
• Are primary and secondary outcomes prespecified?
• Are deviations from the protocol identified and explained?

Use a design-specific reporting guideline when relevant. The EQUATOR Network library routes reviewers to CONSORT for randomized trials, STROBE for observational studies, PRISMA for systematic reviews, STARD for diagnostic studies, and other guidelines.

Participants, data, and setting

• Are inclusion and exclusion criteria justified?
• Does recruitment create selection bias?
• Is the sample size rationale reported?
• Are the setting and time period clear?
• Are missing observations and exclusions accounted for?

Track denominators across the manuscript. Differences between enrolled, randomized, treated, followed, and analyzed samples should be visible and explained.

Intervention, exposure, and comparator

• Is the intervention or exposure defined well enough to reproduce?
• Is the comparator appropriate and implemented as described?
• Could co-interventions, contamination, or adherence explain the result?
• For observational work, are likely confounders measured and handled credibly?

Outcome measurement

• Does the outcome correspond to the stated question?
• Are measurement tools valid for the population and context?
• Were outcome assessors blinded when feasible?
• Are time points and thresholds prespecified?
• Are surrogate outcomes presented as if they were patient-important outcomes?

Review the statistical analysis

You do not need to reproduce every calculation to detect structural problems. Ask whether the analysis matches the data and design.

Check:

• unit of analysis and independence assumptions;
• treatment of repeated measures or clustered data;
• missing-data assumptions and sensitivity analyses;
• model covariates and whether they were selected before seeing results;
• effect estimates with uncertainty, not only p-values;
• multiplicity across outcomes, subgroups, and time points;
• subgroup interaction tests rather than separate significance claims;
• consistency between planned and reported analyses.

Our articles on statistical flaws in research papers and the multiple comparisons problem provide focused checks for common failure modes.

If a method lies outside your expertise, identify the concern precisely and suggest specialist statistical review. Do not disguise uncertainty as a definitive accusation.

Verify the results claim by claim

For each central claim, locate its source in a table, figure, or analysis. Confirm:

1. the population and denominator;
2. the outcome definition and time point;
3. the comparator;
4. the effect estimate and uncertainty;
5. whether the analysis is primary, secondary, or exploratory.

Look for results that are mentioned in the abstract but difficult to find in the main paper, and for prespecified outcomes that disappear. Check whether tables, figures, text, and supplement agree.

A structured research paper summary is a useful neutral layer here: record what the manuscript reports before evaluating what it means.

Test the discussion and conclusion

The discussion should interpret the results without repairing weaknesses in the methods through rhetoric. Flag conclusions that:

• imply causality from an observational design;
• equate non-significance with equivalence or no effect;
• focus on within-group change when the relevant contrast is between groups;
• generalize beyond the studied population or setting;
• ignore imprecision, attrition, harms, or inconsistent outcomes;
• present exploratory subgroup findings as confirmation;
• claim novelty without a transparent comparison to prior evidence.

Also note when the authors have calibrated the conclusion appropriately. A review should recognize strengths as well as problems.

Check reporting, ethics, and transparency

Confirm that the manuscript reports ethics approval or an appropriate explanation, consent where required, funding, competing interests, data availability, and trial registration or protocol information when applicable.

Plagiarism, image manipulation, duplicated publication, or fabricated data require careful handling through the editor. Describe the observable evidence; do not make unsupported allegations to the authors.

Write the report in three layers

A clear peer-review report separates editorial judgment from revision detail.

1. Brief summary

In two or three sentences, state the question, design, and main contribution. This shows the editor and authors how you understood the manuscript.

2. Overall assessment

Name the manuscript’s main strengths and the few issues that determine whether its central claim is supportable. Keep confidential publication recommendations in the editor-only field if the journal requests that separation.

3. Prioritized comments

Order comments by importance:

• Major: validity, missing analysis, unsupported central claim, or essential reporting gap.
• Minor: clarification, presentation, terminology, or a local inconsistency that does not change the conclusion.

For each major comment, use this structure:

Location: where the issue appears.
Observation: what the manuscript currently reports.
Why it matters: how it affects validity or interpretation.
Requested action: a proportionate analysis, clarification, correction, or limitation.

Replace “the statistics are weak” with a locatable concern and a feasible response. Separate essential changes from optional suggestions.

Common peer-review mistakes

• Reviewing the topic you wish the authors had studied.
• Spending more space on grammar than on validity.
• Asking for citations to the reviewer’s own work without necessity.
• Treating a reporting omission as proof that a method was not performed.
• Demanding new experiments that are outside the paper’s central claim.
• Listing concerns without explaining their consequence.
• Using sarcastic, dismissive, or personal language.
• Letting an AI tool invent flaws or expose confidential text.

Using AI as a second reader

An AI critique can help create a checklist, locate potentially inconsistent claims, or surface missing reporting. It should not replace domain judgment or journal policy.

If AI assistance is permitted:

1. use only material you are authorized to process;
2. ask for evidence locations behind each concern;
3. verify every claim against the manuscript;
4. delete generic or speculative criticism;
5. rewrite the final report in your own accountable judgment;
6. disclose use when the journal requires it.

SinaPilot’s AI Review workflow analyzes readable papers for strengths, limitations, statistical issues, conflicts, and open questions. Its output is best used as a structured second pass, not an autonomous editorial decision.

How to peer review a research paper: final checklist

Before submitting, confirm that your review is confidential, unbiased, evidence-based, prioritized, respectful, and proportionate. The editor should know what drives your assessment. The authors should know what they can do next. Neither should have to infer the core problem from a long list of disconnected comments.

Related guides

• Establish the factual record with an accurate scientific paper summary.
• Evaluate trial validity with the clinical trial critical-appraisal guide.
• Use a formal, design-specific risk-of-bias assessment when the review requires domain judgments.
• Compare the manuscript’s claims against other studies using an evidence matrix.
• Place the paper inside a transparent systematic review workflow.