What Peer Reviewers Wish Authors Knew: 5 Habits

Reviewers are unpaid experts protecting the record, and these five habits make their job and your revision far easier.

Opinion · 10 min read · Directive Publications Editorial Team

1. Make the contribution obvious early

If the reviewer cannot find the “so what?” by page two, frustration rises.

The contribution is not your topic. A reviewer already has the topic from the title. What they are looking for in the abstract and in the last paragraph of the introduction is narrower: what was not known before this paper, what you did about it, and what changes now that you have.

State it in one sentence, in that shape. Something is well established; something adjacent has not been examined in your setting or population; you did a specific thing; here is what you found; here is what a reader can now do, or stop assuming. Put that sentence in the abstract, then again — not word for word, but recognisably the same claim — at the end of the introduction. If it appears for the first time in the discussion, it has arrived too late to shape how the paper is read.

The test is cheap. Give your abstract and first two pages to a colleague from a neighbouring subfield and ask them to write down, in one sentence, what is new here. If what comes back is your topic rather than your finding, a reviewer will land in the same place.

What goes wrong when the contribution is buried is that the reviewer builds one for you. They read to page six, form their own version of what the paper claims, and review that version. It is almost always smaller than the one you had in mind, and the report then argues against the smaller version — which is how the word incremental ends up in a decision letter about work that was not.

2. Do not overclaim

Narrow conclusions read as confidence. Sweeping claims read as risk.

Overclaiming is rarely dishonesty. It is a gap between what the design can support and the verb you reached for. Three forms account for most of it: causal verbs on data that cannot establish causation (improves, reduces, leads to, where the design shows association); generalisation past the sampling frame, so a result from one site, one species, one platform or one time window is written as a property of the world; and priority claims — the first study to show — which a reviewer can only verify or resent.

Apply one rule to every sentence in the abstract, the conclusions and the highlights. Name the specific result that supports the claim, then name the population, setting and period the design actually covers. Rewrite the claim to cover exactly that and nothing beyond it. In practice this turns "X reduces Y" into "in this cohort, higher X was associated with lower Y", and it turns "these findings apply to clinical practice" into a sentence about the specific patients you studied.

The narrower claim is not the weaker paper. It is the defensible one, which matters years later when somebody tries to build on your result and finds it does not extend as far as your abstract implied. Reviewers know that, and a paper written at the right scope reads as work by someone who understands their own design.

Two related habits help. Let your limitations section say what each limitation prevents you concluding, rather than listing constraints as a ritual. "Because the sample was drawn from a single site, these estimates should not be read as national" tells a reviewer you have already thought about the objection they were drafting. And drop the adjectives — novel, groundbreaking, unprecedented carry no information inside a manuscript. State what you found and let the reader supply the word.

Overclaiming compounds, too. A reviewer who catches one claim your data cannot carry starts reading the rest of the paper looking for more, and trust in a manuscript is global rather than local. A single unsupported sentence in an abstract can turn a routine review into a suspicious one.

3. Make figures self-contained

Axes, sample sizes, and definitions should not require a treasure hunt.

Reviewers do not read straight through. They read the abstract, jump to the figures, and only then go back for the methods. Assume every figure will be looked at by somebody who has not yet read your methods section, and build it so that person is not stranded.

Before you submit, check each figure for:

• Axis labels with units, and a note when an axis is log-scaled or does not start at zero
• The number of observations behind each group, in the caption or on the panel itself
• What the error bars are — standard deviation, standard error or confidence interval — stated explicitly, because all three look identical and mean different things
• Every abbreviation defined in the caption, even where it is already defined in the text
• Scale bars on images, and panel letters running in the order the caption describes them
• The test behind any significance marker, and what each asterisk stands for
• Colour that survives greyscale printing and does not carry meaning by red-versus-green alone

The check takes a minute. Cut the figure and its caption out, hand them to a colleague with no other pages, and ask what was measured, in what units, in how many samples, and what is being compared. Anything they cannot answer is something a reviewer will have to ask for in writing.

That is the real cost. Every trip from your figure back to your methods is another chance for a reviewer to notice something else, and every unanswerable question becomes a numbered comment. A report padded with requests for sample sizes and error-bar definitions is a report that had less room for the science, and it usually guarantees a further round.

4. Answer every comment in revision

Silence on a hard point is louder than disagreement. Be complete and civil.

Number every comment as the reviewer laid it out, and split any comment that bundles three requests into three numbered items — you will have to answer them separately anyway. Then answer in their order, not in the order that flatters you.

Give every item the same three parts: the comment reproduced verbatim, your response, and the revision. The revision line names the section and the line numbers from the final file, and quotes the new wording where it is short enough to quote. A reviewer who can approve a change without reopening your manuscript usually will, and sending them back into the file is where fresh objections are born.

Accepting a point is a complete response in itself. "Corrected, thank you" answers a typo. The letter does not need to be longer than the review; when it is, you are usually arguing rather than revising.

Where you disagree, disagree in three moves: state what you understood the reviewer to be asking, give the specific reason you are not doing it, and say what you did instead. That third move is what makes it land, because nearly every declined comment has a partial concession available. You cannot add the second cohort, but you can name it in the discussion as the study that would settle the question. Watch where the wording puts the fault, as well: "we did not make our design clear, and have rewritten the opening of Section 2" arrives at the same place as "the reviewer has misunderstood our design", without the cost.

If a comment is genuinely unclear, say so plainly to the editor, state your best interpretation, and answer that. Guessing in silence is what produces a third round. And the one comment you skip is the one the reviewer looks for first when the revision reaches them.

5. Suggest ethical reviewers (when asked)

Independent experts, institutional emails, no close conflicts. Guide: Suggesting Reviewers

Offer names only when the journal asks for them, and treat the field as a test of judgement rather than a favour box.

Rule out anyone who has co-authored with any of you recently, supervised or been supervised by any of you, works at your institution, holds a grant with you, collaborates with you now, or has read a draft of this manuscript. Family and close friends are obviously out. What remains — researchers whose work you cite and argue with, in other groups and ideally other institutions — is exactly the pool an editor wants.

Give each name on one line: full name, institution, institutional email address, a short clause on why they are qualified, and an explicit statement that you have no relationship with them. Use the address published on their institutional page. Never supply a personal webmail address for a suggested reviewer. Because reviewer-suggestion fields have been abused to route review requests back to the authors themselves, editors treat a non-institutional address as a warning sign even when your intentions are entirely ordinary.

You can usually ask for someone to be excluded as well. Keep the reason short and professional — a directly competing project, an unresolved dispute — and name one or two people rather than half the field. A long exclusion list reads as an attempt to pick your own verdict.

The failure here is expensive. A conflict the editor discovers costs you time at minimum, and one discovered after publication stops being an editorial matter and becomes an integrity one.

What these five have in common

Each habit removes a question a reviewer would otherwise have to ask you in writing. A review round is largely the cost of questions the manuscript could have answered by itself: what is new here, how far does the claim reach, what does this axis mean, did they address the hard comment, is this suggested reader independent.

Answer those in the file and the report that comes back is about your science. That is the review you actually want, and it is the one that reaches a decision fastest.

Bonus wish

Cite data/code when the paper depends on them. Integrity context: Directive Academy · Peer Review Process

Cite them as references, with persistent identifiers and version numbers, rather than as a line in the acknowledgements promising availability on request. A reviewer who cannot reach your dataset is reviewing your description of your dataset, and they will say so in the report.

Deposit before you submit, so the identifier exists at the moment a reviewer wants it. Where data are genuinely restricted, give the reviewer a route anyway — a repository's anonymous reviewer link, a de-identified or synthetic extract, or a clear statement of who controls access and how it is requested — rather than nothing at all. For software, name the versions of the packages you actually ran and point to the analysis code, because a method described in prose and a method as executed are not always the same thing.

A last pass before you submit

1. The contribution is stated in one sentence in the abstract and again at the end of the introduction
2. Every claim in the abstract and conclusions names the population, setting and period the design covers
3. No causal verb sits on a design that cannot support one
4. Each figure carries units, sample sizes, error-bar definitions and its own abbreviations
5. Limitations say what they prevent you concluding, not just what was constrained
6. Data and code are deposited, cited with identifiers, and reachable by a reviewer today
7. Suggested reviewers are independent, described in a line each, and listed with institutional emails
8. A colleague outside your subfield has read the first two pages and told you, unprompted, what is new

More: Directive Academy · For Authors