How to write a response to reviewers

A response to reviewers is the document accompanying a revised manuscript that reproduces every reviewer comment verbatim, numbered by reviewer (R1.1, R2.3), and answers each one immediately beneath — what changed, where in the manuscript, and the revised sentence quoted — opening with a half-page summary for the editor and leaving no comment unaddressed.

A response letter decides more of the revision’s outcome than the new experiments do: an editor reading a revision reads it first, and often reads little else closely. A letter that is complete, navigable and specific converts a borderline revision into an acceptance; one that is defensive or evasive converts a straightforward one into a second round.

That is not only a claim about editorial habit. Journals ask for the letter as a document in its own right, uploaded separately, and they say what it has to do. PLOS ONE’s instructions for revising a manuscript put it in one sentence: “Address the specific points made by each reviewer.” The same instructions ask for three files — the response to reviewers, a marked-up manuscript showing the changes, and a clean final version — which means the letter is the only one of the three written specifically for the editor rather than for the record.

The structure that works

Reproduce every comment in full, in the reviewer’s own words, and answer each immediately beneath it. Number them. Distinguish the reviewer’s text from yours by formatting — bold, italic, or a coloured block — consistently throughout.

For each comment, three things in this order: what you did, where it is in the revised manuscript with a page or line reference, and — where it matters — the text as revised, quoted.

Quoting the revised text is the highest-value habit in the whole document. An editor who can see the change without opening the manuscript is an editor who does not have to go looking, and reviewers re-reading a revision check the letter first.

Number by reviewer and position — R1.1, R1.2, R2.1 — rather than running a single sequence across all reviewers. The reason is arithmetic: a single sequence renumbers every downstream comment the moment you split one, and reviewer 2 reading “comment 14” cannot tell whether it is theirs. Reviewer-prefixed numbering also gives you a stable handle for the cross-references described below, and it survives a third round unchanged.

Anchor your location references to one named file and say which. “p. 7, lines 214–219 of the clean copy” is checkable; “p. 7, line 214” is not, because the tracked-changes copy and the clean copy have different line numbers on every page after the first insertion. Submit both files, name them in the letter’s first line, and key every reference to the clean copy.

Do not rely on colour alone to separate the reviewer’s words from yours. Response letters are forwarded, printed, converted to PDF and read on phones, and a blue-versus-black distinction disappears in at least two of those. Pair the colour with a structural marker — a bold Reviewer 1, comment 3 label and an indented block — so the document still parses in greyscale.

Quoting the revised sentence in the letter supplements a marked-up manuscript; it does not replace one. Some editorial offices bounce a revision back when the changes are shown only as manual bolding or highlighted text rather than the word processor’s own change-tracking, because a reviewer cannot accept or reject an individual change, or run a compare-documents check against the original submission, when the markup is decorative rather than structural. Generate the marked copy with Word’s or LibreOffice’s native track-changes feature, then quote from the clean copy in the letter itself — the two documents serve different readers and neither substitutes for the other.

Common questions about the letter’s format

How long should a response to reviewers be?

As long as the comments require, not a fixed length. Completeness matters more than brevity: a short paragraph — what you did, where it is, the revised sentence quoted — covers a substantive comment, and two or three sentences cover a minor one. A letter noticeably longer than the manuscript itself usually signals defensiveness rather than thoroughness.

Should reviewer comments be numbered straight through, or restart for each reviewer?

Number by reviewer and position — R1.1, R1.2, R2.1 — not with one sequence running across all reviewers. A single running sequence renumbers every downstream comment the moment you split one in two, and a reviewer reading “comment 14” cannot tell whether it addresses their own review or someone else’s.

Do I actually have to quote the revised sentence in my response, or is it fine to just say I clarified it?

Quote the sentence. “We have clarified this in the text” is the single most distrusted line in a response letter, because reviewers cannot verify it without opening the manuscript and searching. Quoting the exact revised text, keyed to a page and line in the clean copy, converts an assertion into a fact the reviewer can check in seconds.

The editor-facing summary at the top

A response letter opens with a summary written for the editor, not for the reviewers, and it is half a page at most. The editor is deciding whether to send the paper back out, and that decision is made before they reach comment R1.1.

Four things belong in it: what the substantive changes were, in three or four sentences; which requests you declined and the one-line reason for each; any point on which the reviewers contradicted one another, flagged explicitly for arbitration; and anything material that changed independently of the reviews, such as a new author, a corrected figure, or a data deposit that now has an accession.

Declaring the declines in the summary is counter-intuitive and it is the right move. An editor who discovers on page 9 that you refused a requested experiment reads the omission as concealment; an editor told on page 1 reads the same refusal as a position you are willing to defend. The cover letter that accompanied the original submission is not a template for this — that document argued the paper’s fit and significance, and the summary here argues only that the revision is complete.

Answer every comment

A complete response letter answers every comment in the review. Every one, including the ones you consider trivial, and including the ones you decline. An unanswered comment reads as an ignored comment, and it is the single most reliable way to earn another round.

Where two reviewers raise the same point, answer it fully in one place and cross-reference from the other: “Please see our response to Reviewer 1, comment 4.” Do not paste the same paragraph twice — it reads as padding.

The failure mode is not usually a comment you skipped on purpose. It is a comment you did not see, because reviewers do not write in numbered lists. A single paragraph frequently carries three requests: a statistical objection, a request for a citation, and an aside about the figure legend. Split that paragraph into R2.4a, R2.4b and R2.4c, reproduce the reviewer’s full sentence under each, and answer them separately. Extracting the requests is the step most authors skip and the step that produces most second rounds.

Audit completeness mechanically before you submit. Count the imperative and interrogative sentences in each review — every “please”, “the authors should”, “it is unclear whether”, “why did”, “I would suggest” — and check that each one has a numbered response. Counting that way reliably surfaces more requests than reading the review did, because the requests buried mid-paragraph are exactly the ones a reader’s eye skips.

Reviewers also write a “general comments” preamble that contains no request at all. Answer it anyway, in two sentences, acknowledging the summary and noting where the revision changed the picture they described. Silence there reads as the same silence as anywhere else.

Questions about answering every comment

Do I have to do everything the reviewers ask?

No — you have to respond to everything, not agree with everything. Declining a request with a specific reason and an offered alternative is normal scientific exchange; silence is not. The strongest declines name a verifiable obstacle — a closed cohort, a discontinued reagent, a consent clause — or cite a published result showing the request would not be informative.

Is it ever okay to just tell a reviewer no?

Yes, when the decline comes with a specific reason and, where possible, something offered in its place — a supplementary analysis, an added limitation, a substitute dataset. What reads badly is a bare refusal with no reasoning attached, or a decline that repeats “beyond the scope of this study” without saying what would put it back in scope.

A reviewer’s paragraph seems to bury three different requests in one comment — do I need to split it into separate responses?

Yes. A single paragraph frequently carries a statistical objection, a request for a citation, and an aside about a figure legend, and answering it as one comment usually means answering only the first request found. Split it into R2.4a, R2.4b, R2.4c, reproduce the reviewer’s exact sentence under each, and answer them separately — extracting the buried requests is the step most authors skip.

How to disagree

Disagreeing with a reviewer works when the response follows four moves in order: restate the concern in your own words, refuse it if you must, cite the specific reason — a source, an analysis, a design constraint — and substitute something in its place. Disagreement itself is normal and editors expect it; what fails is skipping straight to the refusal without the other three moves.

State the reviewer’s concern as you understand it, so it is clear you engaged rather than deflected. Give the specific reason you disagree. Offer something: the alternative analysis in supplementary material, a limitation added to the discussion, the data that let the reader judge.

We appreciate this suggestion. A post hoc power calculation computed from our observed effect would be a deterministic function of the reported p-value and would not provide independent information. We have instead added 95% confidence intervals for each primary comparison (Table 2, p. 14), which convey the range of effects our data can and cannot exclude.

That declines the request, explains why, and supplies something better. It reads as competence.

Restate, refuse, cite, substitute — the same four moves — carry the paragraph above and generalise beyond it. The citation is what does the work: Hoenig and Heisey’s “The Abuse of Power” (The American Statistician 2001;55:19–24) establishes that observed power is a monotone transformation of the p-value, so the refusal rests on a published result rather than on the authors’ preference. A disagreement anchored to a named source is a disagreement the editor can adjudicate without taking sides; see reviewer asked for a post hoc power calculation for the full argument and reviewer objections about statistics for the neighbouring comments.

Hoenig and Heisey’s result is narrower than authors often apply it, and the narrowness matters because the same citation gets pasted against a different, legitimate request. It rules out only observed power computed from the study’s own effect size, because that number is a fixed function of the p-value and adds no independent information. It does not rule out power computed the other way: a sensitivity analysis that takes the achieved sample size as given and asks what effect size the study could reliably detect, or an a priori calculation for a planned follow-up using an external effect estimate. A reviewer asking for the second kind is asking a different, answerable question, and citing Hoenig and Heisey against it is a citation misapplied rather than a citation used. The worked mixed-model example later in this article turns on a related but distinct problem, pseudoreplication — treating repeated measurements from the same animal or subject as independent observations, the design flaw Hurlbert named in ecological statistics in 1984 and that reviewers of preclinical and behavioural work now catch routinely.

The same shape works where the disagreement is about interpretation rather than method:

Reviewer 2 reads Figure 3 as showing that the intervention causes the change in expression. We agree the language in the original discussion supported that reading and have narrowed it: the design is cross-sectional and cannot separate the two directions. The revised text (p. 18, lines 502–507 of the clean copy) states the association and names the experiment that would test causation. We have not run that experiment, for the reason given in our response to R2.6.

That answer concedes the substance, fixes the manuscript, and declines the experiment in the same breath — which is the combination editors are looking for. Where the underlying objection is that association has been written as causation, reviewer says the result is correlative, not causal covers the specific rewrite.

What does not work: “the reviewer is mistaken”, answering a different question than the one asked, or conceding in the letter while changing nothing in the manuscript.

Where a reviewer is wrong on a verifiable fact — they state a control is absent when it is in Figure S3 — say exactly that, name the location, and stop. One sentence, no adjectives. The same test that governs when to appeal a rejection applies inside a revision: a factual error the editor can verify without re-forming a judgement is worth raising, and a difference of judgement is not.

When the reviewers contradict each other

Contradictory reviewer requests are an editorial problem, not an authorial one, and the correct move is to hand them back. Reviewer 1 wants the subgroup analysis removed as underpowered; reviewer 2 wants two more subgroups added. You cannot satisfy both, and choosing silently means the reviewer you disappointed reads the revision as non-compliance.

Write it out explicitly. Quote both comments side by side under a single numbered heading, state the conflict in one sentence, say which course you took and on what ground, and ask the editor to rule if they prefer the other. Editors expect this and prefer being asked to discovering it later.

Choose on a principle you can name, not on which reviewer sounded more senior. Reporting-guideline conformance is the strongest ground available — a CONSORT or STROBE item settles most conflicts about what must be reported in a clinical or observational study, and the ARRIVE 2.0 guidelines (Animal Research: Reporting of In Vivo Experiments, revised 2020) do the equivalent job for animal work — a reviewer asking you to drop a required item is asking you to violate the journal’s own submission requirements. Pre-registration is the second strongest: an analysis specified in the registered protocol survives a reviewer’s preference for a different one, and adding the requested analysis as clearly labelled exploratory work satisfies both.

Where the conflict is about scope rather than method — one reviewer wants the paper broadened, the other wants it narrowed — the editor’s original decision letter usually resolves it. Editors write “please address reviewer 1’s methodological concerns in particular” precisely to signal priority, and that sentence is the arbitration you already have.

Common questions about disagreeing with reviewers

How do I handle contradictory reviewer requests?

Say so explicitly: quote both comments side by side under one numbered heading, state the conflict in a single sentence, say which course you took and on what ground, and invite the editor to rule otherwise if they prefer. Choose on a principle you can name — reporting-guideline conformance or a pre-registered analysis — not on which reviewer sounded more senior.

My two reviewers want opposite things — one wants an analysis removed, the other wants two more added — what do I actually do?

Neither compliance nor silent choice: write the conflict out. State that reviewer 1 asked for the subgroup analysis to be dropped as underpowered while reviewer 2 asked for two more to be added, say which you did and why, and ask the editor to overrule you if they disagree. Choosing one silently reads to the disappointed reviewer as non-compliance, not compromise.

Can I just tell a reviewer they’re factually wrong?

Only when it is verifiable, not a matter of judgement. If a reviewer states a control is absent when it is in Figure S3, say exactly that, name the location, and stop — no adjectives. A difference of interpretation is a different case: state your reading, concede what is fair, and let the editor decide rather than declaring the reviewer mistaken.

When you cannot run the requested experiment

Declining a requested experiment is legitimate, and the response turns entirely on which of three reasons applies. Editors distinguish between them sharply, and authors routinely give the weakest available one.

The experiment is impossible. The cohort is closed, the cell line is no longer available, the reagent has been discontinued, the ethics approval does not cover it, the patients were recruited under a consent that forbids it. State the specific obstacle and, where you can, evidence it — a discontinued catalogue number, the date the biobank closed, the clause of the approved protocol. This is the strongest decline and it usually ends the exchange.

The experiment would not answer the question. Explain why, with the same restate–refuse–cite–substitute structure used above, and propose the design that would. This is a scientific argument and it can be won. One recurring version: a reviewer asks the authors to “prove there is no effect.” Ordinary null-hypothesis significance testing cannot do that — a non-significant result is absence of evidence, not evidence of absence — and the answerable substitute is equivalence testing (two one-sided tests, TOST, against a pre-specified smallest effect size of interest), not a rerun of the same test with a larger sample.

The experiment is disproportionate to the claim. Here the honest move is often to narrow the claim rather than defend the omission. A reviewer asking for an in vivo confirmation of a mechanism is telling you the paper claims more than it shows; scaling the conclusion back to what the data support is a smaller change than the experiment and answers the same concern. Reviewer says the mechanism is not shown and reviewer wants more controls both work through that trade.

Always offer the nearest available alternative — an orthogonal assay you already have, a re-analysis, a published dataset that speaks to the point, or a limitation naming the experiment as future work. A decline with nothing attached reads as refusal; a decline with a substitute reads as a decision.

How long you have, and what to do about it

The revision window is set in your decision letter, and there is no cross-journal standard to fall back on. PLOS ONE’s public instructions for revising a manuscript specify what the response must contain and state no fixed number of days; the day counts circulating on editing-service and author-advice sites are conventions those sites report, not a published policy you can cite. Read the letter, and if it names no date, write to the editorial office and ask for one in writing.

Ask for an extension before the window closes, not after. Editorial offices grant extensions routinely when the request arrives in time and names a specific new date and a specific reason; the request that arrives two weeks late is a different conversation, because the submission may already have been closed in the system. Where the reason is an experiment with a fixed duration — an eight-week animal protocol, a sequencing queue — say so, because a dated technical reason is easier for an editor to approve than “we need more time”.

A revision returned after the window may be handled as a new submission at the journal’s discretion, which resets the submission date and can send the paper to different reviewers. Treat the date in the decision letter as the operative constraint on scope: it determines which of the requested experiments you can actually run, and that determination belongs in your scoping of the revision before you draft a single response.

The four mistakes

Four failures account for most avoidable second rounds, and each has a signature an editor recognises immediately.

Silence on a comment. Covered above, and worth repeating because it is the most common. The reviewer’s version of this comment is “the authors have not addressed my point 3”, and it arrives with the credibility of the whole letter already spent.

Fixing the analysis and leaving the claims. If the reanalysis narrowed the result, the abstract and discussion must narrow with it. Reviewers check this specifically, and finding the original claim intact reads as bad faith rather than oversight. Run the overclaim check over the revised manuscript before submitting the response. The check is mechanical: list every claim in the abstract, and for each one name the figure or table that now supports it at the revised effect size. Any claim whose support moved is a claim whose wording must move.

Defensiveness. Long arguments about why criticism is unfair, however deserved the feeling. Editors read tone. The visible version of this defect is disproportion: an answer many times longer than the comment that prompted it reads as defensive whatever its content, and an editor scanning for the substance will not find it.

Vague compliance. “We have clarified this in the text” without saying where or how. Reviewers read it as having done nothing, and half the time they are right. The repair is the quoted revision described above — the sentence as it now stands, with its page and line, which converts an assertion the reviewer must verify into a fact they can see.

Tone

Tone in a response letter is one decision, made once: brief thanks at the top, then straight into the work. No effusiveness, no injury. Reviewers are usually anonymous colleagues who spent unpaid hours on your paper, and a letter that treats their comments as a technical problem to solve reads better than one that treats them as an attack to survive.

Where a reviewer was genuinely unfair or made an error of fact, say so plainly and factually once, and let the editor weigh it. Escalation beyond that rarely helps.

The specific pattern to delete is the per-comment compliment. “We thank the reviewer for this insightful comment” pasted above every answer adds length without information and makes the one genuine acknowledgement worthless. Thank once, then use the space for page numbers.

Two register failures are worth naming because they are invisible to their authors. The first is the passive concession — “the wording has been amended” — which hides who decided what and leaves the reviewer unsure whether the substance changed. Write “we have narrowed the claim to”. The second is the pre-emptive rebuttal, where the letter argues against a criticism the reviewer did not make; it signals that the authors know the weakness and expects the editor to go looking for it.

What the letter looks like on the page

A complete response letter is three files and a predictable shape. The files are the point-by-point response, a marked copy of the manuscript showing every change, and a clean copy. The shape runs: a two-line note naming the manuscript number and the three files; the editor-facing summary; then reviewer by reviewer, comment by comment.

Each entry has the same four parts, in the same order:

Reviewer 1, comment 2. The statistical analysis does not account for repeated measurements from the same animal.

Response. We agree. The original analysis treated each measurement as independent, which inflates the effective sample size.

Change. We refitted the primary model as a linear mixed model with a random intercept for animal (Methods, p. 6, lines 158–166 of the clean copy). The revised estimate is −0.34 (95% CI −0.61 to −0.07), compared with −0.34 (95% CI −0.52 to −0.16) previously; the interval widens and the conclusion is unchanged.

Revised text. “Differences between groups were estimated using a linear mixed model with a random intercept for animal to account for repeated measurements (n = 8 animals, 3 measurements each).”

That entry runs well under 150 words and answers everything: what was wrong, what was done, where it is, what changed numerically, and the sentence a reader will now encounter. Reviewer says the n is not independent covers the underlying statistical point; a filled-out response to reviewers sample shows the same structure across a full letter.

Your response letter may be published

Response letters are published documents at a growing number of journals, and authors write them as though they were private correspondence. PLOS ONE tells authors plainly that “if you choose to publish the peer review history of your manuscript, your response to reviewers will appear online alongside the final article.” eLife publishes the letter as part of the permanent record wherever authors supply one: a Reviewed Preprint carries the “eLife assessment, public reviews and a response from the authors (if available).”

Write for that audience from the first draft. A sentence you would not want a future reader — a hiring committee, a competitor, the reviewer’s own students — to see attached to your name permanently should not be in the letter. The practical filter is simple: nothing about the reviewer’s competence, nothing about their likely identity, and no complaint that depends on the reader taking your word for a tone they cannot see.

Publication cuts the other way too, and usefully. A letter written to be read publicly tends to be better: the arguments are made explicitly rather than gestured at, the citations are real, and the declines are defended rather than buried. It also becomes a durable statement of what the paper does and does not claim. If a reader later asks why the in vivo experiment was not run, or why the causal language was narrowed, the published letter answers in your words rather than leaving the question to inference — which is a reason to write the declines carefully rather than a reason to hide them.

Frequently asked questions about tone and publication

Should I thank the reviewers?

Once, briefly, at the top — not before every answer. “We thank the reviewer for this insightful comment” pasted above each response adds length without information and devalues the one thank-you that was genuine. Use the space that per-comment gratitude would take for what actually helps the reviewer: the page number and the revised sentence.

Will my response to reviewers be published?

At a growing number of journals, yes. PLOS ONE tells authors directly that choosing to publish the peer review history puts the response to reviewers online alongside the final article, and an eLife Reviewed Preprint carries a response from the authors alongside the public reviews when one is supplied. Write every letter as though that future reader already exists.

Is it bad if my response letter sounds a little annoyed?

Yes — irritation reads to an editor as defensiveness, and defensiveness is one of the four failures that most reliably earns a second round. The tell is disproportion: an answer many times longer than the comment that prompted it signals annoyance regardless of the words used. Where a reviewer was genuinely unfair, say so once, plainly, and let the editor weigh it.

Before you submit the revision

Before you submit the revision, read the reviews once more for the objection you decided not to address. If it would recur at another journal, it will recur here. Deciding to leave it is legitimate; leaving it by accident is not.

Then run four checks. Confirm every numbered comment has a numbered response, by count, not by impression. Open the clean copy at each page and line reference in the letter and confirm the quoted text is actually there — line numbers move when you edit, and a letter full of stale references is worse than one with none. Re-read the abstract against the revised results, which is where the overclaim survives. And check that anything you promised in the letter, such as a data deposit or a supplementary file, exists and is accessible; data availability and accession numbers is the specific item reviewers verify.

If the revision is substantial enough that the paper now makes a different argument, read it once as a whole rather than as a sequence of patches. Revisions assembled comment by comment develop internal contradictions — a limitation added for reviewer 2 that the discussion then contradicts, a narrowed result that the introduction still promises. A single continuous read is the only step that catches a contradiction the revision itself introduced, and no comment-by-comment check will substitute for it.

What editors say when the letter fails

Editors and reviewers write a small, recognisable set of comments about response letters, and each maps to a defect above.

“The authors have not responded to my comments 3 and 7.” “The response states that the text has been clarified, but I am unable to locate the change.” “The revised analysis is appropriate, but the abstract still reports the original effect size.” “The authors decline the requested experiment without explaining why it is not feasible.” “The response letter argues at length that the original criticism was unwarranted; I would prefer to see the manuscript changed or the claim withdrawn.” “The line numbers cited in the response do not correspond to the submitted file.” “Reviewer 1 and reviewer 2 make incompatible requests and the authors have followed one without acknowledging the other.”

The last is the one authors most often earn without deserving it, because they resolved the conflict sensibly and never said so. Naming a contradiction takes two sentences and removes the entire class.

More in this family: why papers get rejected, when to appeal a rejection, and rejected after review, what next.

PerfectPaper reads the revised manuscript against the claims it makes, and reports where a narrowed result is still described in its original terms — the defect a reviewer checks first when a response letter says the analysis was corrected.

Review my manuscript