What peer review is for
Peer review is the process by which journals ask independent experts to assess a manuscript before publication. Reviewers advise the editor on whether the work is sound, important and clearly reported, and they suggest how it could be improved. The editor makes the decision. For an early-career researcher, reviewing is also a training ground: assessing someone else's design and reporting makes the weaknesses in your own work easier to see.
What reviewers are expected to do is broad. A scoping review of guidance from biomedical journals and publishers identified dozens of distinct tasks, most of them about assessing the content of the manuscript (Glonti et al., 2019). Not all tasks are equal. When experienced reviewers of randomised trials were asked which tasks mattered most, they placed evaluating the risk of bias first, although few journals asked for it explicitly (Chauvin et al., 2015). Your first job is therefore to judge whether the study can answer its question; style and formatting come later.
Models of peer review
Journals organise review in different ways, and the model affects what you may say and who will see it. In single-anonymous review the reviewers know who the authors are but the authors do not know the reviewers. In double-anonymous review neither side is named, so avoid comments that reveal your identity, such as repeated requests to cite your own papers. In open models the reviewers' names, and sometimes their reports, are published with the article. Some journals also invite comments on preprints or run review after publication. The invitation letter or the journal's reviewer guidance tells you which model applies; write every report as if the authors might one day read it with your name attached.
Plan the time realistically. A careful first review of a research article usually needs several hours spread over more than one sitting: one to read and understand, one to check the methods, results and statistics, and one to write and revise the report. If the journal's deadline does not allow that, say so when you reply to the invitation rather than after you have accepted.
Your obligations before you accept
The Committee on Publication Ethics sets out the basic principles that reviewers are expected to follow (Committee on Publication Ethics, 2017). In summary:
- Expertise. Accept only if you can assess the core of the work, and say which parts you cannot judge.
- Conflicts of interest. Declare any financial, personal or academic interest, and decline if you work closely with the authors, have recently supervised or collaborated with them, or are preparing a competing manuscript.
- Confidentiality. Treat the manuscript and your review as confidential during and after the process, and do not use unpublished information for your own work.
- Permission. Ask the editor before involving anyone else, including a co-reviewer such as a supervisor or a student you are training.
- Timeliness. Respond promptly to the invitation and return the report within the agreed time, or ask for an extension early.
- Tone. Be objective and constructive, and avoid hostile or personal comments.
Generative AI tools raise a specific risk. Uploading a manuscript to a public tool can breach confidentiality, and medical journal editors ask reviewers not to upload manuscripts to software where confidentiality cannot be assured and to seek the journal's permission before using AI to help with a review (International Committee of Medical Journal Editors, 2024). Read the journal's own policy, because publishers differ, and remember that you remain responsible for every statement in your report.
A ten-step process for reviewing a manuscript
The steps below follow the order in which most reviewers work. Read the journal's reviewer guidance first: it may ask specific questions or use a scoring form.
- Read the whole manuscript once. Without making notes, read for the argument. Then write a short summary in your own words: the question, the design, the main findings and the claimed contribution. If you cannot write it, that is itself a finding.
- Judge the question and contribution. Is the question clear, important and new relative to the literature? Problems that are important or timely are a strength reviewers value (Bordage, 2001).
- Check the title, abstract and keywords. Do they describe the study accurately, and does the abstract stand alone? Is the justification in the introduction consistent with what the study actually did?
- Assess the design and methods. Can this design answer this question? Consider sampling, measurement, controls or comparators, and sources of bias. Weak design is the reason most often given for outright rejection (Pierson, 2004).
- Check reporting against the relevant guideline. For a randomised trial use CONSORT (Schulz et al., 2010); for an observational study use STROBE (von Elm et al., 2007). Missing items often reveal missing information rather than poor writing.
- Examine the results, tables and figures. Can each table and figure be understood on its own? Are numbers consistent across text, tables and abstract? Are measures of uncertainty reported, and are the same observations counted only once (for example, repeated measurements on one person treated as independent data)?
- Review the statistics. Check that each analysis matches the design and data, that assumptions are addressed and that effect sizes and confidence intervals are given, not only p-values (Lang & Altman, 2015; Wasserstein & Lazar, 2016). If this is outside your expertise, say so to the editor.
- Weigh the discussion and conclusions. Do the conclusions follow from the results? Are findings interpreted in the light of earlier work, and are limitations stated honestly? Watch for causal language that the design does not support.
- Check the references. Are key or recent studies missing? Are claims supported by appropriate sources: primary studies for specific findings, and systematic reviews for summary statements?
- Write the report and recommendation. Organise your comments as described below and give a recommendation in the form the journal asks for.
Reviewing a review article
Review articles are judged differently. Ask whether the search and selection were systematic and reported, whether the synthesis adds understanding beyond a list of studies, and whether the conclusions and suggestions for future research follow from the evidence. Specialised guidance exists for some review types; for bibliometric reviews, for instance, Hoang (2025) offers a practical guide for reviewers and readers.
Writing the reviewer report
Editors need a report they can act on, and authors need comments they can answer. Authors are usually advised to respond point by point (Noble, 2017), so numbered, specific comments make the next round easier for everyone.
| Part of the report | What to include | Tips |
|---|---|---|
| Summary | Two to four sentences restating the question, design, main findings and claimed contribution | Shows the editor and authors that you understood the work |
| Overall assessment | Main strengths and the key issue(s) that determine your recommendation | Be explicit about what would change your view |
| Major comments | Numbered problems affecting validity or conclusions: design, bias, analysis, interpretation | Explain why each matters and suggest a fix |
| Minor comments | Numbered points on clarity, presentation, references and small errors | Give page or line numbers |
| Confidential comments to the editor | Concerns about ethics, overlap with other work, or parts you could not assess | Do not repeat the recommendation in a hostile way |
| Recommendation | Accept, minor revision, major revision or reject, as the journal's form defines them | Keep it consistent with your comments |
A worked example: turning criticism into a useful comment (illustrative)
The comments below are invented for this guide and refer to a hypothetical survey study of smartphone use and sleep in university students.
| Unhelpful comment | Constructive comment |
|---|---|
| The methodology is weak. | Major 1. The study uses a cross-sectional survey, but the discussion states that smartphone use "reduces" sleep quality. Because exposure and outcome were measured at the same time, the design supports an association only. Please revise the causal language in the abstract and discussion, or explain why a causal interpretation is justified. |
| Statistics are not clear. | Major 2. Table 3 reports p-values for the regression but no coefficients or confidence intervals. Please report the estimates with 95% confidence intervals and state which covariates were included and why. |
| References are old. | Minor 4. The introduction cites no studies after 2015 on screen use and sleep. Please consider whether more recent work, including any systematic reviews, changes the rationale. |
Each constructive comment names the problem, explains why it matters for the conclusions and suggests what the authors could do. None comments on the authors themselves.
Common mistakes
- Accepting a review outside your expertise without telling the editor which parts you cannot assess.
- Sharing the manuscript with colleagues or AI tools without the journal's permission.
- Rewriting the paper as you would have done it, rather than judging whether it answers its own question.
- Spending most of the report on language and formatting while missing design problems.
- Asking for many extra analyses without explaining how they bear on the conclusions.
- Giving a recommendation that does not match the comments (for example, "minor revision" with fatal flaws listed).
- Using sarcastic or personal language.
- Returning the report late without warning the editor.
Reviewing will also change how you write. For the author's side of the process, see Why biomedical PhD manuscripts are rejected at peer review, How to write a journal manuscript section by section and How to select a journal for publication. For statistical reporting standards, see Reporting statistical results in academic manuscripts.
Key takeaways
- Reviewers must keep manuscripts confidential, declare conflicts, seek permission before involving others and report on time (Committee on Publication Ethics, 2017).
- Do not upload a manuscript to an AI tool unless the journal allows it and confidentiality is assured (International Committee of Medical Journal Editors, 2024).
- Judge design and risk of bias first; reviewers rate it the most important task (Chauvin et al., 2015).
- Use reporting guidelines and statistical reporting standards to check completeness (Schulz et al., 2010; von Elm et al., 2007; Lang & Altman, 2015).
- Write numbered, specific and constructive comments that authors can answer point by point (Noble, 2017).
