Most researchers read their first peer review report on their own manuscript, which is the worst possible moment to learn the genre. The comments arrive numbered and polite, the verdict hides in a single sentence, and nobody tells you which three points decide whether the paper survives. This page shows a complete sample of journal reviewer comments, one by one, with what each comment means and how the authors could answer it.
The report is about a real paper: a randomized controlled trial published in BMC Nursing in December 2025 under a CC BY 4.0 licence, which allows anyone to reuse and discuss it with attribution. Using a published, openly licensed paper means you can open the article next to this page and check every sentence the reviewer quotes. It also means nothing here comes from anyone’s confidential manuscript.
It is not the journal’s own report, and it is not a verdict on the authors. The paper passed peer review, and problems like the ones below survive peer review all the time. That is exactly why they are worth learning to spot before you submit.
Hemati M, Ghanbarzadeh S, Mohebbi Z. The effect of multimedia-based education on nurses’ clinical capability in radiation protection in operating rooms: a randomized controlled study. BMC Nursing. 2025;24:1479.
The trial enrolled 100 operating room nurses from two hospitals affiliated with Shiraz University of Medical Sciences and allocated them 50 to each arm. The intervention group received a two-week multimedia programme on radiation protection through WhatsApp: five slide presentations, three brochures and two posters. The control group continued with the hospital’s routine training. The outcome was a 32-item self-report questionnaire with four subscales (knowledge, attitude, performance and personal commitment), completed at baseline, one month and two months after the intervention.
The central claim is that the programme produced large improvements on every subscale, with effect sizes between 0.79 and 2.73, that the gains held at two months, and that a group by time interaction in an ANCOVA confirms it. The authors recommend implementing the programme in operating room training.
Hemati et al. (2025), BMC Nursing, doi:10.1186/s12912-025-04108-3 · Full text in Europe PMC (PMC12706883) · Licence: CC BY 4.0
Almost every journal report has the same skeleton. It opens with a short summary in which the reviewer restates what the paper claims, which tells you whether they understood it. The verdict usually sits in the sentence that follows the first “however”. Then come the major comments, the changes without which the reviewer would not recommend publication, and the minor comments, which are corrections and clarifications. The report ends with a recommendation to the editor: accept, minor revision, major revision or reject. Many systems also let the reviewer write confidential comments that only the editor sees.
Read the recommendation first, then the major comments of every reviewer, and only then the minor ones. One major comment that asks for data you do not have weighs more than ten minor ones about wording.
Where these comments come from: they are based on the report that Previum produced for this paper on 11 September 2026, which you can read unedited on the nursing example page. Here it is rewritten in the format a methods reviewer would send to a journal and checked by hand against the published article; the minor comments were added during that check. Every number quoted below is in the article.
Each comment appears as the reviewer would send it, followed by what it really means and how the authors could answer it.
“This randomized trial tests a low-cost, scalable way of training operating room nurses in radiation protection, a relevant occupational safety question. The trial was registered, reports a CONSORT flow diagram and used a blinded assessor for outcome data. However, the statistical analysis as reported does not support the main conclusion: the ANCOVA table is internally inconsistent, the analysis does not account for the repeated measurements, and the allocation procedure needs clarification. I recommend major revision.”
What it means: the first three sentences are not flattery, they show that the reviewer read the paper and found real strengths. Everything that matters is after “However”, and the closing sentence is the recommendation. Major revision means the paper is alive, on condition that the next version solves what follows.
“Table 3 cannot be correct as printed. For Group, F(1, 294) = 4.343 corresponds to p = .038, not p = .173. For Time, F(2, 294) = 0.810 corresponds to p = .446, not .552, and a partial eta squared of .448 is impossible with that F: it would be about .005. The Group row reports ηp² = .685 in the table but η² = .015 in the text. Finally, the three effects share the same 294 error degrees of freedom, so they should share one error mean square, yet dividing each mean square by its F gives about 8,506 for Group and Time and about 259 for the interaction. Please re-run the analysis and report the output exactly as the software produces it.”
What it means: this is arithmetic, not opinion. A p value follows from its F and its degrees of freedom, and partial eta squared follows from the same numbers (F × df1 divided by F × df1 + df2). When the numbers in a table do not reconcile, a reviewer stops trusting every other number in the paper, including the significant interaction on which the main claim rests. This single comment is often what turns a minor revision into a major one.
How to answer: re-run the model, replace the table with the software output, say plainly where the error came from (a transcription slip is a perfectly acceptable explanation) and state whether any conclusion changes. Never defend the original numbers.
“With 100 participants measured three times, an error term with 294 degrees of freedom (300 observations minus 6 cells) implies that the 300 measurements were analysed as if they were independent. They are not: each nurse contributes three correlated scores. The text calls this an ANCOVA adjusting for baseline, yet baseline also appears as one level of the time factor. Please analyse the follow-up scores with a model that respects the repeated measures, either an ANCOVA of the one- and two-month scores with the baseline score as covariate, or a linear mixed model with a random intercept for each participant, and report the group effect with its 95% confidence interval.”
What it means: treating repeated measurements as separate people inflates the degrees of freedom and shrinks the standard errors, so p values come out smaller than they should. It is one of the most common errors in pre-post designs, and it usually happens when the data are stacked in long format and run through a factorial ANOVA.
How to answer: run the model the reviewer asks for, present it as the primary analysis and explain in one sentence why it replaces the previous one. If the effect survives, the paper is stronger than before. If it shrinks, say so: the reviewer will check.
“The Methods state that participants were allocated 1:1 by block randomization with a block size of 4, using opaque, sequentially numbered envelopes opened at enrolment. The same paragraph states that, to minimize contamination, the control group was recruited first and completed both pre- and post-test assessments before enrolment of the intervention group started. Both sentences cannot describe the same procedure. If the groups were enrolled one after the other, allocation was not concealed and the groups were measured at different times of the year, so any change in the hospitals over those months is confounded with the intervention. Please describe exactly how and when each participant was allocated.”
What it means: allocation is the backbone of a trial, and an ambiguous description makes the editor ask whether the study is randomized at all. Reviewers who know the CONSORT checklist look for the sequence generation, the concealment mechanism and who enrolled and assigned participants, and they compare those sentences with the rest of the Methods.
How to answer: describe what actually happened, in order. If enrolment was sequential, the honest fix is to relabel the design as a non-randomized controlled study, change the title and add the limitation. It is far better to find this yourself than to have a reviewer find it.
“The sample size was calculated for a paired-sample t-test using pre and post means from a single-group study (59 [13] and 71 [20]), which gave 44 participants, and was then set at 50 per group. A paired calculation estimates the power to detect change within one group; the primary question here is the difference between two groups over time. Please justify the sample size for the design actually used, or report a sensitivity analysis stating the smallest between-group effect the study could detect with 50 participants per group.”
What it means: reviewers check that the power analysis matches the primary analysis. A mismatch is not fatal once the data are collected, but the answer is never a post hoc “observed power”, which is a restatement of the p value.
How to answer: add a sensitivity analysis. With 50 per group, a two-sided alpha of .05 and 80% power, a two-group comparison detects standardized differences of about d = 0.57 or larger. State it, and discuss what smaller effects the trial could have missed.
“All outcomes come from one self-report questionnaire, including the performance subscale, and participants knew whether they had received the training. The original validation reported Cronbach’s alpha between .50 and .76 for the subscales, and reliability in the present sample is not reported. Effect sizes up to d = 2.73 for a two-week messaging intervention are unusually large and may partly reflect expectations or social desirability. Please report reliability in this sample, discuss these sources of bias and temper the recommendation to implement the programme.”
What it means: reviewers weigh plausibility. A very large effect on a self-reported outcome in a trial where participants know their group invites an alternative explanation. The comment does not ask for new data; it asks for honesty in the Discussion.
How to answer: report alpha or omega for each subscale in your own data, add the limitation in plain words and rewrite the conclusion so that it claims what one trial in two hospitals can support.
Minor comments are conceded, all of them, without negotiation. Arguing over a minor point always costs more than fixing it.
“Major revision. The question is relevant and the trial has real strengths, but the main analysis must be re-run and the allocation procedure clarified before the conclusions can be assessed.”
What it means: the reviewer is willing to read the paper again. The editor will usually send the revised version back to the same reviewer, who will open the response letter with this report beside it and check each point.
Reviewer language is polite by convention, and the severity is coded in a handful of formulas. Knowing them prevents the two classic misreadings: treating a firm objection as optional, and treating a routine request as an attack.
The answer is a point-by-point document: every comment quoted in full, followed by your response and the exact change in the manuscript, with page and line. The order of work matters more than the order of the letter. Start with the numbers (comment 1), because if the re-analysis changes a result, everything downstream changes too. Then the model (comment 2), the description of the design (comment 3), the sample size (comment 4) and the Discussion (comment 5). Leave the minor comments for last and concede them all.
An answer to the first comment could read: “We thank the reviewer for checking Table 3. The values were transcribed incorrectly from the software output. We re-ran the analysis as recommended in comment 2 and replaced Table 3 with the new model (page 9, lines 210 to 224). The group effect at two months is now reported with its 95% confidence interval; the conclusion about the direction of the effect is unchanged, but its size is smaller, and we have revised the abstract and Discussion accordingly.” It concedes, it shows the change, and it gives the reviewer something to verify.
All five major comments were visible in the submitted text. The first one is pure arithmetic: recompute each p value from its test statistic and degrees of freedom and compare. That is one of the automatic checks Previum runs on every manuscript, and on this paper it flagged both inconsistent p values in Table 3.
The other four need a reader who knows design and statistics and reads adversarially: someone who asks whether the model matches the data structure, whether the allocation paragraph matches itself, whether the power analysis matches the primary analysis, and whether the effect is plausible for the outcome measure. That reading takes an hour before submission, or three months after it.
The Basic level of Previum shows the two findings that decide the verdict, each with the passage it quotes and a fix, together with the automatic checks: recomputed p values, figures that contradict each other and references verified one by one. The Professional level returns the full report from two reviewers. Your manuscript is not stored on any server.
The same approach is available for six published, CC BY licensed papers from BMC journals: nursing, physiotherapy, education, sports science, dentistry and psychology. Each page shows the report Previum produced on that paper, unedited, next to the article it reviews. There is also a full example of a Professional report on a fictional manuscript, written to show every section of the report.
It is a report on a real, published paper, but it is not the report written by the journal’s reviewers. It was built from the report Previum produced on the paper, rewritten in the format of a journal report and checked by hand against the article. Every quoted number can be verified in the published version.
A short summary of the paper, a list of numbered major comments, a list of minor comments and a recommendation to the editor. Major comments ask for changes in analysis, design description or interpretation; minor comments ask for corrections and clarifications. Most are polite in tone, which is why the formulas matter.
There is no standard number. It depends on the reviewer, the field and the paper. What matters is the type: comments that ask for clarification or reanalysis can be answered; comments that question the design or the contribution are the ones that decide the outcome.
A minor revision asks for small corrections that the editor can often check without sending the paper back to reviewers. A major revision asks for substantial changes, such as new analyses or a rewritten Discussion, and the revised version usually goes back to the same reviewers.
Yes. It is your manuscript. The restrictions many publishers place on AI tools in peer review are aimed at reviewers handling someone else’s confidential submission. Previum does not store your manuscript on any server, and the report stays in your account.
Journals with open or transparent peer review publish the reports next to the article. Look for a peer review history or open peer review link on the article page. Reading a few of them in your field is the best training there is for your own first report.