Establish the review mandate
Begin with the editor’s invitation, the journal’s scope and the manuscript’s article type. Determine whether the editor expects a broad scientific assessment or a specialist review of statistics, methodology, clinical relevance or another defined component. A reviewer’s task is to advise the editor on the submitted work under the journal’s criteria, not to impose a different research agenda.
Before opening a detailed critique, confirm that you have the complete manuscript and the materials needed to assess it. If supplementary files, protocols, reporting checklists or revised versions are referenced but unavailable, ask the editorial office. Do not silently judge a paper as incomplete because a file was omitted from your reviewer access.
Keep the distinction between assessment and decision visible throughout the process. Your recommendation is advice supported by reasons; the editor is responsible for weighing all reports, journal criteria and other relevant information.
Protect the manuscript and prepare a secure workspace
Treat every submitted file, editorial message and unpublished result as confidential. Read and annotate the manuscript only in environments consistent with the journal’s confidentiality requirements. Avoid shared folders, unsecured devices or tools that automatically synchronize confidential content to services whose handling conditions you have not checked.
Do not forward the manuscript to a colleague, trainee or specialist without editorial permission. Co-reviewing may be valuable when the journal authorizes it, but contributors should be identified according to journal rules and bound by the same confidentiality expectations.
Do not upload confidential manuscript text, figures or datasets to external generative-AI systems unless the journal explicitly permits the specific use and appropriate safeguards are in place. The reviewer remains responsible for every statement in the report.
Make a first-pass scientific map
Read the title, abstract, introduction, principal figures and conclusion once without trying to write every comment. State the research question in your own words, identify the primary claim and determine which evidence the authors use to support it. Note the population or material studied, the design, the primary outcomes and the main analytical strategy.
This first pass should reveal whether the manuscript’s sections describe the same study. A mismatch between the abstract’s claim and the Methods, or between the stated primary outcome and the headline figure, may be more important than dozens of wording problems.
Write a provisional one-paragraph summary before beginning detailed review. This prevents the final report from becoming a disconnected collection of marginal notes and gives the editor a concise account of what you believe the manuscript actually shows.
Identify the article type and reporting framework
Match your expectations to the submitted design. Randomized trials, observational studies, diagnostic accuracy studies, systematic reviews, qualitative studies, case reports, protocols, data papers and software articles answer different questions and require different evidence.
Use relevant reporting guidance to locate essential missing information—for example, a suitable CONSORT, STROBE, PRISMA, STARD, TRIPOD, CARE, COREQ or ARRIVE resource when applicable. Confirm that you are using the appropriate guideline and extension rather than mechanically applying a familiar checklist to an unrelated design.
Reporting guidance is an aid to transparent assessment, not a numerical quality score. A fully reported study can still be methodologically weak; an incompletely reported study may require clarification before its quality can be judged.
Test the importance and precision of the research question
Ask whether the introduction identifies a coherent knowledge gap and whether the stated objective follows from it. Novelty is not the only source of value: replication, confirmatory work, negative findings and rigorous local evidence can also contribute to the record.
Check whether the manuscript distinguishes its primary question from secondary or exploratory aims. A study that changes emphasis after seeing results may still contain useful findings, but interpretation and disclosure must reflect that distinction.
Do not demand that authors reshape a legitimate question merely because you would have studied a different population or chosen another endpoint. Criticism should address whether the submitted work answers its own defensible question.
Evaluate design and threats to validity
Determine whether the design can support the principal claim. For intervention studies, consider allocation, comparators, masking where relevant, adherence and outcome ascertainment. For observational studies, examine selection, confounding, measurement, temporality and missingness. For qualitative work, examine sampling, data generation, analytic transparency and reflexivity as appropriate.
Distinguish fatal limitations from remediable reporting omissions. Missing information about recruitment may be clarified; a design that cannot establish the causal claim may require substantially narrower conclusions rather than an additional paragraph.
Describe threats to validity concretely. A useful comment identifies the design feature, the plausible direction or nature of its effect, and why the concern matters for the paper’s principal inference.
Check participants, samples and study flow
Examine eligibility criteria, recruitment or sampling, setting, dates, exclusions and the path from eligible units to the analyzed sample. Check whether numbers reconcile across the text, flow diagram, tables and supplementary material.
Assess whether selection or attrition could distort the result. Missing participants and missing measurements are different problems; both may require explanation. When relevant, ask how withdrawals, protocol deviations and excluded observations were handled.
Do not assume a larger sample is always necessary. Evaluate whether sample-size planning, information power or precision is appropriate to the design and the claims the authors make.
Examine interventions, exposures and measurements
Check that the intervention, exposure, comparator or phenomenon of interest is described sufficiently to interpret the study. Important details may include timing, dose, implementation, instruments, diagnostic thresholds, operational definitions and the reliability of measurements.
Ask whether outcomes are meaningful, consistently defined and measured in a manner that could introduce bias. A seemingly impressive effect may depend on a poorly defined outcome or a measurement method applied differently across groups.
Request details that affect interpretation or reproducibility. Avoid demanding exhaustive procedural descriptions unrelated to the study’s central evidence.
Audit protocols, registration and analysis plans
For studies where registration or protocols are relevant, compare reported objectives, outcomes and analysis choices with the accessible prospective record when available. Distinguish legitimate documented amendments from unexplained post hoc changes.
Do not equate every difference with misconduct. A protocol may evolve for sound reasons, but material changes should be transparent and their implications for confirmatory interpretation should be considered.
If the protocol or registration record is essential but inaccessible, ask the editor how to proceed. Do not disclose confidential manuscript information to third parties while attempting to verify it.
Evaluate statistical and analytical choices
Review whether the analysis matches the design, unit of analysis, data structure and stated estimand or research question. Consider clustering, repeated measurements, confounding adjustment, model assumptions, missing data and the treatment of multiple outcomes where relevant.
Look beyond threshold-based significance. Effect sizes, uncertainty intervals, clinical or practical relevance, sensitivity analyses and consistency across related outcomes often matter more than a single P value.
When the central method is outside your competence, state that limitation to the editor and recommend specialist assessment when justified. Do not present a speculative statistical objection as established fact.
Separate prespecified, exploratory and post hoc findings
Identify which analyses were planned and which appear to have emerged during exploration. Exploratory work can be valuable, but it should not be framed as if the exact hypothesis and analysis were specified before the data were examined.
Pay attention to subgroup claims, numerous secondary endpoints, repeated modeling choices and selective emphasis on positive results. Ask for transparency about the analytic path rather than automatically demanding every possible alternative analysis.
Where a claim depends on a post hoc result, consider whether the conclusion should be framed as hypothesis-generating and whether replication or independent validation is needed.
Reconcile results across every representation
Check the abstract, main text, tables, figures, supplements and conclusions against one another. Track sample sizes, denominators, event counts, units, estimates, confidence intervals and reported directions of effect.
Prioritize discrepancies that alter the inference. A mislabeled axis and an inconsistent primary outcome count are not equivalent problems, although both may require correction.
Quote or identify the exact location of a discrepancy and describe the needed clarification. A report that merely says 'numbers are inconsistent' creates unnecessary work for authors and editors.
Assess figures, images and visual evidence
Determine whether each figure represents the stated data accurately, with readable axes, appropriate scales, clear legends and transparent handling of uncertainty. Evaluate whether image processing, grouping or illustrative examples could mislead readers.
If a scientific image appears irregular, document the specific observation and notify the editor confidentially where warranted. Do not assume that an unusual appearance proves manipulation or attempt a public investigation.
Distinguish presentation improvement from requests for new experiments. The latter should be scientifically necessary and feasible in relation to the paper’s claim.
Evaluate research data, code and reproducibility
Read the data-availability and code-availability statements rather than treating them as administrative formalities. Ask whether the materials and methodological details are sufficient for an appropriately qualified reader to understand or, where feasible, reproduce the analysis.
Privacy, consent, intellectual property and contractual restrictions may justify controlled access rather than unrestricted public release. Assess whether restrictions are explained and whether the proposed access pathway is credible.
If data or code are provided to reviewers, preserve their confidentiality. Do not use the materials for your own research or redistribute them outside the editorial process.
Check human and animal research safeguards
Assess whether the manuscript provides coherent information about independent ethics review, approval or a documented determination that approval was not required, informed consent where applicable, participant privacy and appropriate animal oversight where relevant.
Research participation consent and consent to publish identifiable material are different matters. For identifiable clinical images or narratives, flag publication-consent concerns without requesting unnecessary personal information.
If an ethical statement is contradictory or the reported conduct raises a concrete concern, alert the editor through confidential comments. Reviewers should identify signals; editors and relevant institutions coordinate any investigation.
Read the discussion as an inference audit
Ask whether the discussion distinguishes results from interpretation and acknowledges limitations that could materially change the reader’s understanding. Check whether the authors overstate causality, generalizability, diagnostic performance or clinical readiness.
Consider whether alternative explanations are fairly addressed and whether the literature is represented proportionately. Reviewers should identify important omitted context, not demand allegiance to their preferred interpretation.
Conclusions should match the study’s design, evidence and uncertainty. When the evidence is narrower than the language, recommend precise reframing instead of vague instructions to 'tone down' the paper.
Assess originality and the use of literature
Check whether the manuscript appropriately situates its contribution and cites essential prior work. Novelty claims should be evaluated against relevant evidence rather than prestige, geography or the reviewer’s personal research agenda.
Recommend citations because they materially improve context, methods or interpretation. Do not require authors to cite your publications, collaborators or the journal merely to increase citation counts.
Potential plagiarism, redundant publication or suspicious overlap should be reported with specific evidence to the editor. Similarity alone does not establish wrongdoing.
Distinguish essential revision from optional enhancement
Before writing a request, ask whether resolving it is necessary to assess validity, meet ethical obligations, correct misleading interpretation or achieve adequate reporting. If not, label it as optional or omit it.
Requests for additional experiments, larger cohorts or entirely new research questions can impose disproportionate burdens and may effectively demand a different study. Explain why any substantial new work is essential to the manuscript’s existing claim.
Where the design cannot support the claim, a transparent limitation or narrower conclusion may be more appropriate than asking for infeasible new data.
Construct a prioritized reviewer report
Begin with a concise, neutral summary of the manuscript’s question, design and principal result. Follow with a brief overall assessment that identifies the central scientific issues without prematurely announcing an editorial decision to the authors.
Organize major comments by importance. For each, state the concern, point to the relevant location, explain its consequence and describe the information or change needed. Use minor comments for clarity, labeling and correctable presentation issues.
Make the report usable. A short set of precise, high-impact comments can be more valuable than a long unprioritized list that hides the decisive scientific problems.
Write comments that authors can answer
Frame comments as testable or answerable requests. Instead of 'the statistics are weak,' identify the mismatch between the analysis and design, the missing assumption check or the unexplained handling of missing observations.
Separate what you observed from what you infer. If you suspect a problem but cannot establish it, ask for clarification rather than asserting that an error occurred.
Use professional language and avoid remarks about authors’ identities, presumed motives or competence. Firm criticism is compatible with respectful communication.
Use confidential comments to the editor correctly
Provide the editor with your recommendation and a concise explanation of the issues most relevant to the decision. Do not assume that the recommendation category communicates your reasoning.
Reserve confidential comments for matters that legitimately require editorial handling: potential integrity concerns, conflicts, confidentiality issues or sensitive context. Where appropriate, make scientific concerns visible to authors as well so they have a fair opportunity to respond.
Do not use the confidential field to make unsupported accusations or to conceal a decisive criticism that could appropriately be shared with the authors.
Handle suspected integrity problems without becoming an investigator
Possible fabrication, falsification, image manipulation, inappropriate text overlap, undisclosed conflicts or peer-review interference should be reported through the editorial process with specific observations.
Do not contact the authors’ employer, institution, funder or social media audiences on your own initiative. Do not circulate the confidential manuscript to gather informal opinions.
A concern is not a verdict. The editor determines appropriate verification and whether institutional or other formal procedures are required.
Perform a final quality check before submission
Reread your report for factual accuracy, internal consistency, prioritization and tone. Confirm that each major request is scientifically justified and that minor suggestions have not been presented as acceptance conditions.
Check that your recommendation aligns with the concerns described. Verify that no confidential or identifying material has been placed in an inappropriate field and that any permitted AI assistance has been verified and disclosed as required.
Submit within the agreed timeline. If the report cannot be completed responsibly, communicate with the editor rather than sending an unexamined assessment.
Review revised manuscripts with continuity
On a later round, read the decision letter, your prior report, the authors’ point-by-point response and the revised manuscript together. Judge whether the scientific concern was resolved, not simply whether the authors accepted your wording.
Do not add an unrelated new wish list at each round. New serious issues may emerge from revisions, but reviewers should support a convergent process when earlier concerns have been adequately addressed.
An evidence-based disagreement is a legitimate response. Explain why the response does or does not resolve the concern and leave the editorial decision to the editor.
Treata standard for a complete manuscript review
A Treata reviewer should provide an independent, confidential, proportionate and evidence-based assessment that distinguishes validity, reporting completeness, interpretation and optional presentation improvements.
The report should communicate its central concerns clearly, avoid coercive citation and inappropriate demands, respect authors, disclose relevant conflicts and permitted tool use, and route integrity concerns through the editor.
Journal-specific instructions may prescribe report fields, recommendation categories or specialist assessments. Follow those instructions alongside the publisher-wide Peer Review, Competing Interests, Research Integrity, Research Data and Artificial Intelligence policies.
Continue learning
Related reviewer guides
Related standards