---
name: peer-review-response
description: >
  Builds an author's response to peer review: sorting the comments, structuring
  the response document, disagreeing professionally, and deciding whether to
  revise, appeal or withdraw. Use for "respond to reviewer comments", "write a
  response to reviewers", "reviewer 2 misunderstood my paper", "the reviewers
  contradict each other", "how do I disagree with a reviewer", "should I appeal
  this rejection", "revise and resubmit", "the reviewer wants a different paper".
  Read the confidentiality warning in Section 2 before using any part of it as a
  reviewer.
category: 15 Academic University Research
ref: "15.26"
tier: 3
inherits: [K2, K3, K4, K5]
---

# Peer Review Response

## 1. One-line description

A method for an author to read, sort and answer peer review: separating what must change from what should change, what is a misunderstanding and what is a genuine disagreement, and producing a response document that addresses every point in a form an editor can adjudicate.

## 2. What this skill is used for

### Read this first: confidentiality

**A manuscript under review is confidential.** It is an unpublished document shared with a reviewer under an expectation of privacy, and in most publishing systems reviewers agree explicitly not to disclose, distribute or use it. Entering someone else's manuscript, or the review of it, into an AI system is a disclosure. **Many publishers now prohibit this outright**, and treat it as a breach of reviewer confidentiality regardless of intent or of any assurance about how the text is handled. Some also prohibit AI-generated review text entirely, on the separate ground that a review is a professional judgement that must be made by the named reviewer.

Two consequences, both binding.

1. **This skill is built for an AUTHOR responding to reviews of their own work.** That is the use it is designed for and the use it supports. An author's own manuscript and the reviews they received are theirs to work with, subject to their venue's own AI disclosure rules.
2. **A reviewer must check their publisher's policy before using any part of this skill, and must not enter a manuscript under review into an AI system where that policy prohibits it.** Where the policy is silent, treat it as prohibited: the confidentiality obligation exists independently of whether the publisher has written a rule about this specific technology. Section 15 covers the ethics of being a reviewer, in a form that requires no manuscript to be disclosed.

**The research problem it solves.** Responding to review is a distinct professional skill and almost nobody is taught it. The characteristic failures are predictable: responding in the first two days, while the emotional reaction is still driving the sentences; treating every comment as an instruction and making changes that damage the paper; treating criticism as an attack and defending everything; missing comments, which editors notice; conceding the paper's contribution because a reviewer pushed; and writing a response document that an editor cannot adjudicate because it does not say what changed or where. This skill supplies the process: the reading discipline, the sort, the document structure, the register for disagreement, and the decision about whether this paper should be revised for this venue at all.

**Where it sits in the research lifecycle.** After submission and a decision letter, before resubmission. It also runs after a rejection, where the same sorting produces a revision plan for a different venue.

**Typical use cases.**
- Building a response document for a major or minor revision.
- Deciding which of a reviewer's requests to accept, negotiate or decline.
- Handling two reviewers who want opposite things.
- Responding where a reviewer has misread the paper.
- Deciding whether to appeal a rejection, revise for elsewhere, or stop.
- Managing a co-author group that disagrees about how much to concede.
- Extracting the useful revision plan from a rejection at one venue before submitting to another.

**Who uses it.** Authors at every career stage; doctoral researchers receiving their first reviews, for whom the emotional load is heaviest and the process least familiar; supervisors and mentors coaching a first response; research managers publishing from applied work.

## 3. When to use it

- A decision letter has arrived with reviews attached, whatever the decision.
- You have read the reviews once, reacted, and need a process rather than a reply.
- A revision is due and the response document has not been started.
- Two reviewers have asked for incompatible changes.
- A reviewer has criticised something the paper does not say.
- A reviewer is asking for a study you did not do.
- Co-authors disagree about whether to concede a point.
- A rejection has arrived and the decision between appeal, resubmit elsewhere and shelve has to be made.

## 4. When NOT to use it

- **You are a reviewer, and your publisher's policy prohibits AI use in review, or is silent.** See Section 2. Do not enter the manuscript, the review, or identifiable passages of either. Section 15 supports the reviewer role without any disclosure of the manuscript.
- **You have read the reviews within the last day or two.** This is not a delaying tactic; it is the method. The first reading is emotional and produces responses that have to be rewritten or, worse, sent. Step 1 covers this. Where a genuine deadline makes waiting impossible, say so and compensate with a co-author reading the response before it goes.
- **The paper has not been submitted yet, and what is wanted is pre-emptive criticism.** That is a different task: **15.19 Examiner-Proofing and Defence Strategy** reads work as a hostile examiner would, and **15.24 Journal Article Development** assesses publication readiness before submission. Anticipating reviewers is cheaper than answering them.
- **The reviews are consistent that the paper has no contribution, and you agree on reflection.** Revision will not fix this and a response document is the wrong instrument. Go back to **15.24 Journal Article Development**, Steps 1 to 4, and expect the contribution claim to change.
- **The decision is where to send it next, not how to answer.** That is **15.25 Journal and Venue Selection**. Note that a rejection still produces a revision plan here, and the plan should be executed before the next submission rather than after the next rejection.
- **The dispute is about authorship, data integrity, prior publication or an allegation of misconduct.** These are not review responses. They run through the venue's own process and, usually, an institution's. Do not answer an integrity allegation inside a response-to-reviewers document. `RESEARCHER DECISION REQUIRED` (K5 §2.4, §2.8).
- **A reviewer has made a comment that appears to identify them and creates a conflict, or the review is abusive.** That is a matter for the editor, raised separately and directly, not something to answer point by point. Answer the substantive comments; raise the conduct issue in a separate note to the editor.
- **What is wanted is a response document to send unedited under the author's name.** The response is a professional communication that the author signs and is accountable for. See §12.1.

## 5. Required inputs

**Required.**
- **The decision letter.** The editor's letter is the governing document and is routinely under-read. It states the decision, often prioritises among the reviewers' points, sometimes says which comments the editor considers essential, and occasionally contains an instruction that overrides a reviewer entirely.
- **The full reviews.** All of them, complete. A response built on the two comments the author remembers being stung by will miss half the requests.
- **The submitted manuscript**, as submitted, so that "the reviewer misread it" can be checked against what the text actually says.

**Optional, and what each one adds.**
- **The venue's guidance for revisions.** Deadlines, format for the response document, whether a tracked-changes version is required, and word limits that may have changed.
- **Any correspondence with the editor.** Including a presubmission enquiry, which sometimes contains a scope agreement a reviewer is now contradicting.
- **Co-authors' reactions, collected before drafting.** Cheaper to reconcile at the sorting stage than at the sign-off stage, and co-authors frequently see a concession the lead author is making unnecessarily.
- **The data, code and analysis files.** Determines which requested analyses are feasible, and how long each would take, which converts an abstract argument about scope creep into a schedule.
- **Time and resource constraints.** A deadline, a contract ending, access to a facility expiring, or a co-author leaving. Determines what can honestly be promised in the response.
- **Previous reviews of the same paper at another venue.** Where the same criticism appears twice from independent reviewers, it is evidence about the paper rather than about a reviewer.

## 6. Questions to ask before starting

1. **What did the editor actually decide, and what did the editor emphasise?** Reviewers advise; editors decide. Where the editor has prioritised, that priority governs the response. Default if unaddressed: re-read the decision letter line by line before touching the reviews.
2. **How long has it been since you first read the reviews?** Determines whether to start now or wait. Default: if under 48 hours, sort but do not draft.
3. **Which comments, if refused, would sink the paper at this venue?** Determines where the negotiating capital is spent. Default: assume any comment the editor repeated is in this category.
4. **Is the requested work feasible in the time allowed, and with the data and access you still have?** Determines whether the revision is a revision or a new study. Default: cost every requested analysis in days before agreeing to any of them.
5. **Do the co-authors agree on what to concede?** Default: collect reactions before drafting, not after.
6. **Is this still the right venue for this paper?** A revision is an investment, and sometimes the reviews reveal a scope mismatch that no revision fixes. Default: assume yes unless the reviews are asking for a different paper (Step 7).
7. **What does this venue's AI use policy require you to declare?** Applies to the revised manuscript and, at some venues, to the response document. Default: check, and disclose (K5 §7).

## 7. Step-by-step methodology

**Step 1. Read once, stop, and come back.**
Read the decision letter and all reviews once, completely, then put them away for at least two days. This is a method step, not self-care. The first reading is dominated by threat response: criticism of work you spent a year on is processed as criticism of you, attention narrows to the harshest sentences, and comprehension of the rest is poor. Authors who draft in that state produce responses that are defensive where they should concede, conceding where they should defend, and occasionally unsendable. On the second reading, days later, three things reliably change: comments that read as attacks read as reasonable requests; at least one that seemed devastating turns out to want a clarifying sentence; and the reviewer's point, as distinct from its tone, becomes visible. Writing the angry response and deleting it unsent discharges the reaction without incurring the cost. On the second reading take no position: record what each comment asks for, in its own terms, without evaluating it. *Correct result: two readings separated by real time, and a numbered inventory of every discrete request. Note that a single review paragraph frequently contains two or three separate requests, and the second and third are the ones authors miss.*

**Step 2. Sort every comment into one of four categories.**
This is the analytical core of the method, and doing it before drafting anything prevents the whole class of failures where every comment is treated identically.

- **Must change.** The comment identifies a real error, a genuine gap, an unsupported claim, a missing piece of method reporting, or something the editor has flagged as essential. Accept these fully and quickly; there is no negotiating capital to be gained by resisting a correct criticism, and resisting one damages the credibility of every disagreement elsewhere in the document.
- **Should change.** The comment identifies a genuine improvement that is not strictly required: a clearer framing, an additional robustness check, a restructured section, an extra citation. Accept most of these. They are cheap, they signal engagement, and they are where a reviewer's expertise earns its keep. Decline only where the change would damage something else, and say what.
- **Misunderstanding to clarify.** The reviewer has read the paper as saying something it does not say. Treat this, almost always, as evidence that the text was unclear rather than that the reviewer was careless (Step 4).
- **Disagreement to defend.** The reviewer's criticism is substantive, you understand it, and you think it is wrong. These are legitimate and, well argued, are frequently accepted (Step 5).

Two disciplines make the sort work. First, every discrete request gets its own row, even where three sit in one paragraph. Second, sort before deciding cost: the category is about the nature of the comment, not about how much work it implies. A "must change" that takes three weeks is still a must change, and that fact belongs in the schedule, not in the sort. *Correct result: a table with one row per request, its category, and a one-line note of what answering it requires.*

**Step 3. Build the response document so an editor can adjudicate it in one pass.**
The editor reading your resubmission is deciding whether the reviewers' concerns were addressed, reading your response beside the reviews, often quickly. The structure that makes this possible is fixed and unglamorous: address every comment, in the order the reviewer made them, with three elements each. **The reviewer's comment**, quoted in full and visually distinguished, so the editor need not hold two documents side by side. **The response**, saying plainly what you did or why you did not. **The location**, naming exactly where the change appears (section, page, line, or the new text quoted). The third is the one authors omit and editors need most: "we have clarified this" without a location is unverifiable and reads as evasion. Open with a short covering note thanking the reviewers briefly, summarising the main changes, and stating any point on which you have respectfully not made the requested change, so the editor is not surprised by it on page nine. Number comments to match the reviews. Answer every one, including the trivial: an unanswered comment reads as an unaddressed comment, and editors check. *Correct result: a document in which the editor can find every comment, the response, and the location, without opening a third file.*

**Step 4. Answer a misunderstanding by fixing the text, not by correcting the reviewer.**
When a reviewer has misread the paper, the instinct is to point out that the paper already says the thing. This is almost always wrong, for a reason that is uncomfortable and usually true: a competent specialist, reading carefully enough to write a review, took a different meaning from the passage. If one expert reader misread it, ordinary readers will too, and the text is unclear even where it is technically correct. So accept the diagnostic value of the misreading, change the text so it cannot be read that way, and say so without either capitulating on substance or scoring a point: acknowledge that the point was not clear, state what the paper intends, and name the change made to prevent the misreading. Where a passage genuinely cannot be improved, and this does happen, quote the original in the response so the editor can see it and say what has been added to make it more prominent. Never write "as stated on page 6" as a complete response: it is technically accurate, answers nothing, and reads as contempt. *Correct result: a changed manuscript, and a response that names the change rather than the reviewer's error.*

**Step 5. Disagree professionally, and expect to be taken seriously.**
A well-argued disagreement is a normal part of peer review and is frequently accepted. Authors, particularly early-career authors, systematically underestimate this and concede points they should hold, which weakens the paper and occasionally destroys its contribution. The structure of an effective disagreement has four parts. **Acknowledge the point properly**, in a way that shows you understood it, rather than restating a weaker version of it. **Give the reasons**, which must be substantive: evidence, method, the literature, the scope of the claim. "We prefer our approach" is not a reason. **Name the cost of doing it the reviewer's way**, which is usually the strongest part of the argument: what would be lost, what the alternative analysis would not be able to show, what assumption it would require. **Offer something**, where you honestly can: an added limitation, a sensitivity analysis, a sentence acknowledging the alternative view, a reframed claim. Most disagreements can be resolved by narrowing the claim rather than defending it whole, and a narrowed claim that survives is worth more than a broad one that does not. The register matters: neutral, specific, unemotional, and short. Sarcasm, appeals to the author's own standing, and complaints about the reviewer's competence all fail, and they fail with the editor rather than with the reviewer. Where you decline a request, say so directly rather than appearing to comply while not complying, which editors detect and which costs far more than an honest refusal. *Correct result: a disagreement an editor could adjudicate on its merits, with the cost of compliance stated and something offered.*

**Step 6. Handle contradictory reviewers by putting the contradiction to the editor.**
Two reviewers asking for opposite things is common, is not a crisis, and is not something an author can resolve alone, because complying with one necessarily fails the other. Name the contradiction explicitly in the response document, quoting both comments; explain the trade the two positions represent; state which you followed and why, in terms of the paper's contribution and audience rather than preference; and address the other reviewer's underlying concern by another route where possible, since reviewers pulling in opposite directions usually share a worry that a third change would settle. Then leave the adjudication to the editor, which is their role and which they will usually exercise when asked directly. Do not try to satisfy both, which produces an incoherent paper that fails both reviewers and the editor. Do not silently follow the reviewer you agree with and hope the other is not consulted again; they usually are. *Correct result: the contradiction named, the choice made and justified, the other concern addressed by an alternative route, and the adjudication explicitly requested.*

**Step 7. Identify the review that asks for a different paper, and answer it as a scope question.**
Some requests are not revisions. "This would be more interesting with a longitudinal design", "the authors should extend this to three more countries", "a control group would strengthen the study" describe a different study, not an improvement to this one. Distinguish these from genuine gaps by a specific test: could the request be met with the data, access and time available, and does the paper's claim require it? If the claim does not require it, the request is scope expansion and the answer is to state the paper's scope explicitly, explain why the claim is supported within it, add the limitation the reviewer's point implies, and identify the request as future work. If the claim does require it, the reviewer has found a real hole and it belongs in "must change" whatever it costs, up to and including withdrawal. Scope creep across a whole review set is worth costing in aggregate: five individually reasonable additional analyses can amount to a second paper, and agreeing to all of them silently produces a revision that misses its deadline. Where the aggregate is unmanageable, say so to the editor and propose a prioritised subset with reasons, which is a normal and usually well-received negotiation. *Correct result: each expansion request classified as required or not, with the reason, and an aggregate workload estimate the author has actually costed in days.*

**Step 8. Decide between revising, appealing and withdrawing, on the evidence rather than on the sting.**
On a revision decision the answer is usually to revise, and the real question is whether the work is feasible in the time. On a rejection, three routes exist. **Revise and submit elsewhere** is the default and is right most of the time: the reviews are free expert criticism and the next venue sees a better paper. **Appeal** fits a narrow set of cases: a factual error in a review that was material to the decision, a reviewer with an evident conflict, or a decision resting on a misreading the editor can check. Not merely because the decision was harsh or the author disagrees with a judgement call. Appeals rarely succeed, take months, and a failed one costs the relationship with the venue; make one only where the ground is specific, factual and checkable, and put it in a short letter rather than a re-argument of the paper. **Withdraw**, where a paper is under revision, fits where the requested work is not feasible, where the revision would produce a paper the authors do not believe in, or where the venue has become a poor fit. Withdrawal is formal, in writing, and complete before any submission elsewhere (**15.25 Journal and Venue Selection**, Step 9). Whichever route, extract the revision plan from Step 2 first: a rejected paper still has a list of real defects, and submitting it unchanged wastes both the reviews and the next set of reviewers. *Correct result: a decision with its reason recorded, and a revision plan that exists regardless of route.* `RESEARCHER DECISION REQUIRED` on appeal and withdrawal (K5 §2.5, §2.8); both have consequences for the author's standing at a venue that no analysis of the reviews can weigh.

**Step 9. Check the revised manuscript against the response document before resubmitting.**
The last failure in the sequence is a response document that promises changes the manuscript does not contain, which happens easily across a long revision with multiple authors. Go through the response line by line with the revised manuscript open and confirm that every claimed change exists, is at the location stated, and does what the response says it does. Then read the revised manuscript once as a whole, because a paper revised comment by comment frequently loses coherence: new material inserted at eleven points does not automatically form an argument, and reviewers of the second round comment on exactly this. Confirm any required declarations, including AI use where the venue requires it, and confirm that any narrowed claim in the response is also narrowed in the abstract, which is the place a softened claim is most often left standing. *Correct result: every promise verified against the manuscript, the whole paper read once for coherence, and declarations current.* `RESEARCHER SIGN-OFF REQUIRED` before resubmission (K5 §2.8).

## 8. Analytical framework

Every comment passes through the same four-stage assessment before it is answered:

    Comment → What is it actually asking? → Is it right? →
    What would answering it cost? → Category → Response and location

The critical stage is the second, and it is the one that is skipped. Authors respond to the comment's tone, or to the comment they feared receiving, rather than to what it asks. Restating each comment in your own words as a request ("the reviewer is asking us to show that the effect is not an artefact of the sampling frame") separates the request from its delivery, and frequently reveals that a brusque comment is asking for something small and a polite one is asking for six weeks of work.

The four categories then map onto four response registers, and mixing them is what makes a response document read badly:

| Category | Register | The failure to avoid |
|---|---|---|
| Must change | Accept, do it, show it | Arguing, which costs credibility everywhere else |
| Should change | Accept, or decline with a specific cost stated | Silent non-compliance |
| Misunderstanding | Fix the text, own the ambiguity | "As stated on page 6" |
| Disagreement | Argue substantively, offer something | Capitulating, or defending everything |

The document's persuasive power comes from the contrast between categories. A response that concedes readily and specifically where it should makes the two or three places it holds firm far more credible.

## 9. Output format

**1. Covering note to the editor.** Short. Thanks, a few sentences on the main changes, and explicit early notice of any request not complied with.

**2. Comment inventory and sort.** A working table, not usually sent, and the thing the response is built from.

| No. | Reviewer | Comment (verbatim) | What it asks for | Category | Cost (days) | Decision |
|---|---|---|---|---|---|---|

**3. The response document.** Every comment in the reviewers' original order, each as:

> **Reviewer 1, comment 3.** [full verbatim comment]
>
> **Response.** [what was done, or the reasoned disagreement]
>
> **Location.** [section, page, line, or the new text quoted]

**4. Contradiction register.** Any point where reviewers conflict, both comments quoted, the choice made, the reason, and the request for the editor's adjudication.

**5. Revised manuscript**, plus a tracked or marked version where the venue requires one.

**6. Verification record.** The Step 9 check: every promised change confirmed present.

**7. Decision record**, where the route was appeal or withdrawal: the ground, the reasoning, and the revision plan retained for the next venue.

**When a request cannot be met, the format must not disguise it (K4 §1).** There is no acceptable response that implies compliance without it. The honest forms are: the change was made; the change was made in modified form, described; the change was not made, for these reasons; or the change cannot be made with the available data, access or time, and here is what has been done instead. A response that says "we have addressed this" without saying how belongs to none of these categories and is read by editors as the fourth one, badly disguised.

## 10. Quality checks

Run before resubmitting. These sit on top of K4 §8.

1. Was there real time between the first reading and the drafting?
2. Has the editor's letter been read separately, and does the response reflect what the editor emphasised?
3. Does every discrete request have its own row, including the second and third requests buried in a single paragraph?
4. Is every comment answered, including the trivial ones?
5. Does every response state a location in the revised manuscript?
6. Does every claimed change actually exist in the manuscript, at the stated place (Step 9)?
7. Is any response merely "as stated on page X"?
8. Where a request was declined, is the refusal explicit rather than implied by silence or by vague compliance language?
9. Does every disagreement give substantive reasons and name the cost of complying?
10. Has any point been conceded that damages the paper's contribution, and was that trade made deliberately?
11. Is every contradiction between reviewers named and put to the editor rather than silently resolved?
12. Has the aggregate workload of accepted requests been costed, and is it deliverable by the deadline?
13. Does the revised manuscript still read as a coherent argument rather than as a paper with eleven patches?
14. If a claim was narrowed in the response, is it narrowed in the abstract and the conclusion too?
15. Is the tone neutral throughout, with no sentence that would embarrass the authors if the reviewer read it, which they will?
16. Are the venue's declarations, including any AI use declaration, current for the revised version?

## 11. Common failure modes

| Failure | How to recognise it | How to prevent it |
|---|---|---|
| **The 48-hour response** | Drafted the day the decision arrived | Two readings, days apart (Step 1) |
| **Missed comments** | The response has fewer items than the reviews | One row per discrete request (Steps 1 and 2) |
| **Every comment treated as an order** | The paper changes in ways the authors do not believe in | Sort into four categories before deciding anything (Step 2) |
| **Every comment treated as an attack** | Nothing conceded; the document argues throughout | Concede the must-changes readily to earn the disagreements (Steps 2 and 5) |
| **"As stated on page 6"** | A misunderstanding answered by citing the original text | Treat a misreading as evidence of unclear text (Step 4) |
| **Silent non-compliance** | Vague agreement language with no change made | Decline explicitly, with reasons (Steps 3 and 5) |
| **No location given** | "We have clarified this" | Every response names a location (Step 3) |
| **Trying to satisfy both contradictory reviewers** | Two changes that undo each other | Name the contradiction, choose, ask the editor (Step 6) |
| **Undetected scope creep** | Five reasonable additions amount to a second study | Cost every request in days and aggregate (Step 7) |
| **Conceding the contribution** | The revised paper no longer claims what it was written to claim | Check the contribution sentence survives (Step 9) |
| **Emotional register leaking** | Sarcasm, standing invoked, reviewer competence questioned | Neutral, specific, short; have a co-author read it |
| **Promises not kept in the manuscript** | The response and the paper disagree | Line-by-line verification (Step 9) |
| **The patched paper** | Coherent responses, incoherent manuscript | Read the whole revision once as a reader (Step 9) |
| **Appealing because it stung** | An appeal that re-argues the paper | Appeal only on specific, factual, checkable grounds (Step 8) |
| **AI-smoothed response** | Fluent, impersonal, slightly evasive text that concedes nothing specific | The author writes the response; the skill critiques it (§12.1) |

## 12. AI guardrails

Skill-specific. The universal prohibitions in K4 apply in full and are not repeated.

1. **Academic integrity is the binding constraint on everything in this skill.** These skills assist a researcher's thinking, structure and rigour. They do not produce work to be submitted as the researcher's own unaided output. The user must comply with the policies of their institution and of the venue they are submitting to, including any requirement to declare AI assistance. Many publishers now require disclosure of AI use in manuscript preparation and prohibit it entirely in peer review. The skill never writes a passage for submission as though the author wrote it; it interrogates, structures, critiques and teaches. In this skill specifically: sort the comments, structure the document, test whether a disagreement holds, and critique the author's draft response. The author writes the response and signs it.

2. **Confidentiality of manuscripts under review is absolute, and precedes every other consideration here.** A manuscript under review, and a review of it, may be entered into an AI system only by the author whose work it is, and only where their venue permits it. A reviewer must check their publisher's policy first and must treat silence as prohibition. Where a user's framing indicates they are reviewing someone else's manuscript, say so plainly and stop; do not proceed on an assurance that the policy allows it. This includes summarising, paraphrasing, critiquing or checking a manuscript on a reviewer's behalf, all of which require the disclosure the obligation prohibits.

3. **Never draft review text for a reviewer.** A peer review is a professional judgement attributed to a named person who accepts responsibility for it (K5 §2.8), and most publishers now prohibit AI-generated review text explicitly. Section 15 supports the reviewer role by teaching what a good review does; it does not write one.

4. **Never invent what the reviewer meant.** Where a comment is genuinely ambiguous, say it is ambiguous and give the two readings, so the author can choose or ask the editor. Resolving an ambiguous criticism toward the more convenient reading is the specific failure here, and it produces a response that answers a question nobody asked.

5. **Never generate a result, an analysis outcome or a data claim to satisfy a reviewer's request.** If a reviewer asks for a robustness check, the check is run by the author on real data. A response describing an analysis that was not performed is fabrication in a document sent to an editor (K4 §2.1, §2.5).

6. **Never soften a disagreement into apparent compliance.** Fluent prose tends toward agreement, and an AI-assisted response drifts toward conceding points the author intended to hold. Where the author has decided to disagree, the response must say so unmistakably.

7. **Never advise concealing a limitation, an unperformed analysis, or a change not made.** The honest forms of a negative response are listed in Section 9, and every one of them is better received than disguised non-compliance.

8. **Do not assume a discipline, a venue's culture or a review model.** Review norms differ sharply: the length and register of reviews, whether open identities are used, whether editors adjudicate actively, how many rounds are normal, and whether disagreement is expected all vary by field. Where a norm matters and is unknown, ask.

9. **Where the response involves an accusation, an integrity allegation, an authorship dispute or an appeal, mark it for the researcher and do not draft it.** These carry institutional and professional consequences that sit with named people (K5 §2.4, §2.8).

## 13. Best-practice principles

- **The second reading is the useful one. Build the wait into the schedule.**
- **Reviewers advise, editors decide.** Read the decision letter as the governing document and answer the editor as your primary reader.
- **Concede fast where the reviewer is right.** It costs nothing, it takes the shortest paragraph in the document, and it is what makes your two disagreements credible.
- **A well-argued disagreement is normal and is often accepted.** Early-career authors concede far more than they need to, and the paper is worse for it.
- **Narrow the claim rather than defending it whole.** Most disagreements resolve into a scope adjustment that both sides can live with, and a narrower claim that survives review is worth more than a broad one that does not.
- **When a reviewer misread the paper, the paper was unclear.** One expert reader misreading a passage predicts many ordinary readers misreading it.
- **Always give the location.** It is the single highest-value habit in the whole document, and it is the one most often omitted.
- **Answer every comment, including the ones that seem beneath answering.** Editors check for completeness, and a skipped trivial comment reads as evasion.
- **Cost the requests in days before agreeing to any of them.** Five reasonable additions can be a second study, and the deadline does not move.
- **Put contradictions to the editor.** It is their job, they expect it, and attempting to satisfy both reviewers produces a worse paper than choosing.
- **Have a co-author read the response for tone before it is sent.** The author is the worst judge of their own register on this document.
- **Even a rejection produces a revision plan.** Send the improved paper to the next venue, not the same one.
- **Nothing in a response document should be a sentence you would not want the reviewer to read.** They frequently do, in the second round.

## 14. Worked example

Generic fictional scenario, academic (environmental social science).

**INPUT**

A researcher receives a major revision decision on a paper reporting a household survey and interview study of domestic water conservation behaviour in a drought-affected region. Three reviews. Reviewer 1 is broadly positive with eight minor comments. Reviewer 2 is hostile in tone and argues the sampling frame invalidates the central finding. Reviewer 3 asks for a comparison with a second region and questions whether the theoretical framing is the right one. The editor's letter says the decision is major revision, that Reviewer 2's sampling concern is the central issue, and that the theoretical framing question is "worth considering but not required".

**PROCESS**

*Step 1.* First reading produces two days of low-grade misery and a drafted paragraph responding to Reviewer 2 that begins "the reviewer appears not to have read the methods section". Deleted, unsent. Second reading, four days later. Reviewer 2's tone is still unpleasant, and the substantive point, separated from the tone, is specific and partly correct: the sampling frame excluded households in informal settlements, which are plausibly the households whose conservation behaviour is most constrained by infrastructure rather than by attitude. That is a real limitation on the central claim.

*Step 2.* The three reviews decompose into 27 discrete requests, not the 19 numbered comments. Sorted: 11 must change (nine of Reviewer 1's minor points, all correct; Reviewer 2's demand that the sampling limitation appear in the abstract; a missing analysis decision in the methods). 8 should change (figure labelling, two citations, a clearer construct definition, a robustness check on the weighting). 5 misunderstandings (Reviewer 3 believes the study claims a causal effect, which it does not; Reviewer 2 believes the interviews were with survey respondents, which they were not, though the text does not say so clearly). 3 disagreements.

*The judgement call.* Reviewer 2's sampling criticism sits on the boundary between "must change" and "disagreement". Conceding it fully would withdraw the central claim; rejecting it would ignore a correct observation. The resolution splits it: the factual observation is accepted in full and prominently, and now appears in the abstract, methods and limitations, while the inference that the finding is therefore invalid is disagreed with, on the ground that the claim concerns the population the frame does represent and is now explicitly bounded to it. The claim is narrowed rather than defended whole, which is what makes the disagreement acceptable to an editor and makes the paper more accurate.

*Step 6.* Reviewers 2 and 3 contradict each other on framing: one wants more attention to infrastructure constraint, the other a stronger attitudinal frame. The contradiction is named with both comments quoted, and the choice justified by reference to the editor's letter, which prioritised the sampling and infrastructure question. Reviewer 3's underlying concern, that the framing is under-argued, is addressed by strengthening the justification of the existing frame rather than replacing it.

*Step 7.* The second region is scope expansion: the data does not exist, collecting it would take a field season, and the claim does not require it. Declined, with the scope stated, the limitation added, and the comparison identified as the next study. Costed at four months, which makes the refusal easy to justify against a 90-day revision deadline.

*Step 9.* Verification catches two failures: the response promises a robustness check that was run but not written into the manuscript, and the narrowed claim was updated in the results and discussion but the abstract still carries the original unbounded wording. Both fixed before resubmission.

**OUTPUT**

A response document of 27 numbered items, each with the verbatim comment, the response and a location; a prominent early paragraph conceding the sampling limitation and stating exactly where it now appears; one clearly argued disagreement with a narrowed claim offered in its place; a named contradiction put to the editor; a declined request with its cost stated; and a revised manuscript verified line by line against the promises made.

`RESEARCHER SIGN-OFF REQUIRED` on the response document and the resubmission (K5 §2.8).

## 15. Advanced usage

**Anticipating reviews before submission.** The cheapest response document is the one you never write. Running the sort in Step 2 against your own manuscript before submission, in the voice of the most sceptical plausible reviewer, finds most of the must-changes while they still cost hours rather than months. This is the same discipline as **15.19 Examiner-Proofing and Defence Strategy**, applied to a paper.

**Managing a co-author group through a revision.** Circulate the sort, not the reviews, as the basis for discussion. A group discussing raw reviews argues about tone; a group discussing a categorised table argues about substance, which is the argument worth having. Assign each accepted request an owner and a date, and hold the disagreements for a single joint decision rather than resolving them in a thread.

**Second and third rounds.** Later rounds are narrower and the register tightens. Address only what remains open, reference your previous responses by number, and do not reopen a settled point. Where a reviewer raises something new in a second round that could have been raised in the first, it is legitimate to note this politely to the editor while still answering the point on its merits.

**Extracting value from a rejection.** A rejection with substantive reviews is worth more than a thin acceptance. Run Steps 1, 2 and 7 in full, execute the must-changes and most of the should-changes, and only then run **15.25 Journal and Venue Selection**. Where two independent sets of reviewers at different venues make the same criticism, that is strong evidence about the paper (K3 §3.1) and the paper should not be submitted a third time until it is addressed.

**The ethics of being a reviewer, briefly.** This sits outside the author's workflow and is included because the two roles belong to the same people. A reviewer's obligations are: **confidentiality**, which is the reason for Section 2 of this skill and which extends to not using the ideas in an unpublished manuscript for their own work; **timeliness**, since an undelivered review is a real cost to an author's career, and declining promptly is far better than accepting and disappearing; **declaring conflicts** of interest, collaboration, competition or personal relationship, and letting the editor decide rather than deciding alone; **staying within competence**, and telling the editor which parts of a paper you are not qualified to judge; **reviewing the paper that was written**, not the paper you would have written, which is the reviewer-side version of the scope creep in Step 7; **specificity**, since a criticism without a location or a reason cannot be acted on; **proportion**, distinguishing what is essential from what is preference and saying which is which, because editors and authors cannot read your mind; and **civility**, which costs nothing and which is not optional even when the paper is weak. Most publishers additionally prohibit delegating a review, including to a student without the editor's knowledge, and to an AI system. A reviewer who cannot meet these obligations should decline the invitation, promptly and without embarrassment.

## 16. Skill chain

**Recommended previous skills:**
- **15.24 Journal Article Development.** Hands over the contribution claim, which is the thing a revision must not accidentally concede.
- **15.25 Journal and Venue Selection.** Hands over the venue, its review model and its stated policies, including the AI use policy that governs this skill's use.

**Recommended next skills:**
- **15.25 Journal and Venue Selection**, again, where the route chosen in Step 8 is to submit elsewhere.
- **15.24 Journal Article Development**, where the reviews establish that the paper's contribution has to be rebuilt rather than defended.

**Runs well alongside:**
- **15.19 Examiner-Proofing and Defence Strategy**, which is the pre-emptive form of this method.
- **15.05 Academic Writing Structure and Argumentation**, for the paragraph-level work of rewriting a passage that a reviewer misread.
- **13.06 AI Research Governance**, which covers the AI use declarations that attach to a revised manuscript.
- **13.02 Source and Citation Verification**, for the citations a revision adds under time pressure, which are the ones most likely to go unchecked.

---
A Yazi Supplied Skill and resource.