How to write a peer review

A bad review is easy to write in two ways: you can be vague and encouraging, which wastes everyone's time, or you can be dismissive and detailed, which is worse. A good review is specific, ordered by importance, and separates what the work must fix from what you would personally have done differently.

That last distinction is the hardest and the most important. Reviewers routinely ask authors to write the paper the reviewer would have written. That is not the job.

Before you accept

Decline if you cannot be fair or competent. Say so quickly — a review declined in a day costs the author nothing; one declined in three weeks costs them a month.

Declare a conflict of interest if you have one: collaboration, competition, a personal relationship, or work of your own that would be affected by the outcome.

Read the journal's or the client's criteria before you read the paper. You are assessing against their standard, not your own.

Read it twice

First pass, no notes. Read it through as a reader would. What is the claim? Did you believe it? Where did you get lost? Note only the general impression at the end.

Second pass, with notes. Now check the argument against the evidence. Does the method support the conclusion? Are the limitations acknowledged? Is the literature engaged with honestly? Are the figures readable and the statistics appropriate?

The structure of the review

A short summary in your own words. Two or three sentences on what the paper does and claims. This is not padding — it lets the author see how their work was understood, and a mismatch here is itself the most valuable feedback in the review.

Your overall assessment. One paragraph. What is the contribution, and is it sound?

Major points, numbered. Things that must be addressed for the work to stand: a claim the evidence does not support, a missing control, an alternative explanation not ruled out, a structural problem. Three to six is normal. For each, say what the problem is, why it matters, and — where you can — what would resolve it.

Minor points, numbered. Clarity, figures, terminology, missing citations, inconsistencies. Reference page and line numbers.

Confidential comments to the editor, if the process has them: your recommendation, and anything you cannot say to the author, such as a suspicion of misconduct.

Tone

Write about the work, never the author. "The conclusion overstates the evidence" is a review; "the authors do not understand their own data" is not, and it is also less useful.

Be direct about serious problems. Softening a fatal flaw into a suggestion does the author no favours; they will discover it later at greater cost.

Say what works. Not as a courtesy sandwich but because authors need to know which parts to protect during revision — a great deal of good material gets cut in response to vague criticism.

Assume competence and good faith. Most apparent errors are unclear writing rather than incompetence, and asking "do you mean X?" is more accurate and less costly than asserting a mistake.

Distinguishing must from would-prefer

Label it explicitly. "Required: the sample selection needs justifying." "Optional: I would have found a table clearer than the current prose, but this is a preference."

Authors cannot tell the difference on their own, and editors are grateful for reviewers who make it obvious.

Common mistakes

A review that is three lines long. Demanding a different paper. Insisting the author cite your own work — reviewers do this and it is visible. Vague criticism with no location. Rewriting sentences into your voice. Reviewing the topic rather than the paper. Missing the deadline and never saying so. Being harsher because the review is anonymous.

Practical tips

• Number everything. Authors respond point by point and unnumbered reviews make that painful.
• Give page and line references for every minor point.
• Write the summary paragraph before your criticisms; it keeps the review fair.
• Leave it overnight and reread your own tone before submitting.
• Target 500–1,200 words for a journal article. Longer is not better.