30 Claude Prompts for Peer Review Comments You Can Sign

<context> I need a major comment on whether this manuscript answers the question it set. Argue from the paste only. Do not invent a better question from a paper I did not paste. </context> <inputs> - Manuscript: [PASTE MANUSCRIPT] - Question or aim as I read it: [RESEARCH QUESTION OR AIM] - Venue: [VENUE] - Field: [FIELD] </inputs> <task> Write one major comment on question fit: quote the aim, quote what the results actually speak to, name the mismatch if any, and say what would close it without inventing new data. If the paste answers the aim, say so and do not invent a problem. </task> <constraints> - Do not invent papers, findings, or a different research question the authors "should have" asked unless the paste already states that aim. - Do not demand a design they did not run. - If the paste is too thin to judge fit, write NEED NOTES, not a guess. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> One numbered major comment (quote / mismatch / request), then UNVERIFIED rows. </format>

<context> Reviewers reject papers when the gap is louder than the cited set. Stress-test the gap from papers the authors named. Do not invent a missing classic. </context> <inputs> - Manuscript: [PASTE MANUSCRIPT] - Gap sentence I spotted, if any: [GAP STATEMENT OR FIND IT] - Reference list if separate: [PASTE REFERENCE LIST OR IN MANUSCRIPT] </inputs> <task> Audit the gap claim against the cited set in the paste. Tag STRONG / STRETCH / NOT IN PASTE. If it is louder than the notes, rewrite a weaker gap the authors can defend. List searches they still owe as types, not as invented titles. </task> <constraints> - Do not invent papers, authors, years, DOIs, or quotations. - Do not claim "nobody has studied X" unless the paste supports "not in this cited set." - Do not fill a hole with a recalled famous paper. Write SEARCH FOR that topic instead. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Verdict, quoted gap, weaker rewrite the authors could use, search-to-run list, UNVERIFIED rows. </format>

<context> I need major comments on overclaimed results. The discussion may say more than the tables. Do not invent the number that would make a stretch true. </context> <inputs> - Manuscript: [PASTE MANUSCRIPT] - Results tables or excerpts: [PASTE RESULTS OR IN MANUSCRIPT] - Discussion claims I distrust: [WATCH LIST OR NONE] </inputs> <task> Pair each material discussion claim with the result line that is supposed to carry it. Tag MATCH, STRETCH, or NOT IN PASTE. Write a major comment only where the paste shows a stretch. Offer a weaker sentence the authors can keep. No new findings. </task> <constraints> - Do not invent n, p-values, effect sizes, or a table cell. - Do not invent a comparison study. - If a count is missing, write NOT IN PASTE, not a rounded guess. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Table: discussion quote | result quote or none | tag | major-comment draft or NONE. Then UNVERIFIED rows. </format>

<context> I need to know whether this design can support the claims, or whether the problem is fatal. Do not invent a larger n or a second coder that would save it. </context> <inputs> - Manuscript: [PASTE MANUSCRIPT] - Design as I read it: [STUDY DESIGN] - Venue bar: [VENUE] - What I will not demand: [OFF LIMITS] </inputs> <task> Decide if a design problem in the paste is FATAL, FIXABLE, or NOT SHOWN. Quote the methods line. Explain what the design can and cannot support. If FATAL, write the comment a decision letter can reuse. If NOT SHOWN, ask for the missing cell instead of guessing. </task> <constraints> - Do not invent n, an instrument, a second rater, or a finding. - Do not invent a methods citation that would bless or kill the design. - Honor [OFF LIMITS]. Do not demand a trial they did not run unless the claims require it and the paste shows the mismatch. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Verdict, quoted line, can-support / cannot-support, comment draft or NEED CELL, UNVERIFIED rows. </format>

<context> I have a first read and some notes. I need a ranked major-comment list I can sign. Drop any comment that needs a paper, finding, or rule I did not paste. </context> <inputs> - Manuscript: [PASTE MANUSCRIPT] - My notes: [PASTE NOTES] - Venue: [VENUE] - How many majors I will keep: [3 TO 6] </inputs> <task> Write [3 TO 6] major comments, ranked by severity for [VENUE]. Each comment: a short heading, a quote from the paste, why it matters, and a concrete request. Merge duplicates. Move copy-edits to a MINOR parking list. Delete any comment that needs an invented source. </task> <constraints> - Do not invent papers, findings, or journal rules. - Do not write more majors than I asked for. - If two notes are the same problem, merge them. - If a comment cannot be tied to a quote, drop it or mark NEED QUOTE. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Ranked major comments, then a MINOR parking list, then dropped-as-unfounded rows, then UNVERIFIED rows. </format>

<context> I need minor comments on clarity. Fix wording and structure the authors already wrote. Do not add a new argument or a new citation. </context> <inputs> - Manuscript or marked pages: [PASTE MANUSCRIPT] - Pages or sections to stay in: [SECTION SCOPE] - Venue style notes if any: [VENUE RULES OR NONE] </inputs> <task> Write line-level minor comments: quote the line, name the problem (ambiguity, overloaded sentence, undefined term, missing cross-reference), and offer a rewrite that does not change the claim. Skip argument fights. Those belong in majors. </task> <constraints> - Do not invent citations, findings, or a better theory name. - Do not turn a minor into a major by demanding new data. - Do not invent a journal style rule I did not paste. - If a rewrite would change the claim, stop and tag MOVE TO MAJORS. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Numbered minors: location, quote, problem, rewrite. Then MOVE TO MAJORS, then UNVERIFIED rows. </format>

<context> I need minor (or major, if the mismatch is fatal) comments on figures and tables versus the results prose. Use only what I pasted. Do not invent a cell. </context> <inputs> - Manuscript, including figure and table text: [PASTE MANUSCRIPT] - Extra captions or notes: [PASTE CAPTIONS OR NONE] - Venue: [VENUE] </inputs> <task> Check each figure and table mentioned in the paste against the results sentences that cite it. Flag: missing call-out, caption that claims more than the table, number that does not match the prose, and axes or units that the text never defines. Do not invent the missing number. </task> <constraints> - Do not invent n, a data point, or a figure that is not in the paste. - If a figure is named but not shown, write FIGURE NOT IN PASTE and ask for it. - Do not add a citation to "standardize" a caption. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Table: item | text quote | figure or table quote or NONE | mismatch | comment draft. Then UNVERIFIED rows. </format>

<context> I pasted the manuscript and the reference list the authors used. Check completeness. Do not complete records from memory and do not add readings. </context> <inputs> - Manuscript: [PASTE MANUSCRIPT] - Reference list: [PASTE REFERENCE LIST] - Citation style: [CITATION STYLE] </inputs> <task> Cross-check in-text cites against [PASTE REFERENCE LIST]: cites with no reference, references with no cite, duplicate entries, and missing fields. Mark suspect (tidy but implausible) entries as SUSPECT, VERIFY OR DELETE. Output only their items. </task> <constraints> - Do not add a reference they did not paste. - Do not fill a missing DOI, page range, or author from memory. - Do not invent the missing in-text match. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Mismatch lists, missing-field list, suspects, then minor-comment drafts I can paste. UNVERIFIED rows last. </format>

<context> I need minor comments when key terms change meaning mid-paper. Use only definitions the draft already supports. Do not cite a theorist I did not paste. </context> <inputs> - Manuscript: [PASTE MANUSCRIPT] - Terms to lock: [TERM LIST] - Definitions the authors already gave, if any: [DEFINITIONS OR NONE] </inputs> <task> Find every use of each term in [TERM LIST] and of close variants. Propose one locked definition per term from wording already in the paste (or from [DEFINITIONS OR NONE]). List drifted sentences as minor comments. If one word is used two ways, show both and ask the authors to pick. </task> <constraints> - Do not import a textbook definition from memory if they did not paste one. - Do not add sources. - Do not treat a methods synonym as drift if the paste defines both as the same measure. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Term, locked definition, variant list, numbered minor comments, UNVERIFIED rows. </format>

<context> I need a copy-edit pass that does not change the science. No new claims, no new citations, no invented house rule. </context> <inputs> - Manuscript or excerpt: [PASTE MANUSCRIPT] - Citation style: [CITATION STYLE] - Venue rules if I have them: [VENUE RULES OR NONE] - Words the journal bans, if any: [BANNED WORDS OR NONE] </inputs> <task> Write minor copy-edit comments: grammar, agreement, figure call-outs, [CITATION STYLE] consistency on records that already exist, and any rule in [VENUE RULES OR NONE]. Keep every citation unchanged except for style on fields they already supplied. </task> <constraints> - Do not add sources or findings. - Do not invent a journal rule I did not paste. - Do not replace a hedge with a stronger verb for elegance. - If [VENUE RULES OR NONE] is NONE, skip house-style guesses. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Numbered copy-edits with location and a short fix. Then a do-not-touch list (claims you left alone). UNVERIFIED rows last. </format>

<context> I will paste the methods I have. Describe and critique only that sample or corpus. Do not complete a missing n from a typical paper in this field. </context> <inputs> - Manuscript or methods: [PASTE MANUSCRIPT] - Design as I read it: [STUDY DESIGN] - Headcount if I can see it: [SAMPLE N OR UNKNOWN] - Venue: [VENUE] </inputs> <task> Audit the sample or corpus: who or what is in the set, how they enter, who is missing, and the n they supplied. If [SAMPLE N OR UNKNOWN] is UNKNOWN, keep every sentence n-free and list the cells the authors must add. Write comments a reviewer can sign. </task> <constraints> - Do not invent n, a response rate, a demographic split, or an archive they did not name. - If a cell is missing, write NOT IN PASTE. - Do not upgrade a convenience sample into a representative one. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Audit prose, table of known vs NOT IN PASTE fields, numbered comments, UNVERIFIED rows. </format>

<context> I need a methods critique of the analysis they described, not a new analysis I wish they had run. Do not invent a test, a model, or an n they did not name. </context> <inputs> - Methods or analysis section: [PASTE METHODS] - Research question or aim: [RESEARCH QUESTION] - Software they named, if any: [SOFTWARE OR NONE] - Sample n if known: [SAMPLE N OR UNKNOWN] </inputs> <task> Stress-test the analysis against [RESEARCH QUESTION]: units, preparation, primary analysis, secondary or exploratory work, missingness if they mentioned it, and what the analysis cannot support. Write comments that ask for clarification or a weaker claim, not a new dataset. </task> <constraints> - Do not invent n, p-values, effect sizes, or a finding they did not report. - Do not add a test they cannot have run unless you label it OPTIONAL, THEY DID NOT RUN THIS. - If [SAMPLE N OR UNKNOWN] is UNKNOWN, refuse a power claim. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Numbered analysis comments, will-not-support list, OPTIONAL tests clearly marked as not in the paste, UNVERIFIED rows. </format>

<context> I need ethics comments that match what the authors stated. Do not invent an approval number, a waived review, or a population they cannot have reached. </context> <inputs> - Manuscript or ethics paragraph: [PASTE MANUSCRIPT] - Ethics status as I can see it: [ETHICS STATUS] - Data type: [HUMAN / ARCHIVAL / PUBLIC / OTHER] - Known risks I spotted: [KNOWN RISKS OR NONE] </inputs> <task> Check ethics and consent wording: approval status only as they stated it, consent or data-use notes from the paste, risks they named, and limits they already accept. Write comments that ask for a missing statement. Separate ethics facts from study limits. </task> <constraints> - Do not invent an IRB, REB, or ethics ID. - If [ETHICS STATUS] is NONE or silent, write that the review must not claim they have approval. - Do not invent a risk-closing procedure they did not name. - Include a note if they used an AI tool and the venue asked for disclosure, but only if I pasted that rule. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Ethics comments, consent comments, missing-statement list, UNVERIFIED rows. </format>

<context> I need a hostile language audit: does the manuscript claim more cause than the design can support? Do not repair an overclaim with extra data they did not collect. </context> <inputs> - Manuscript: [PASTE MANUSCRIPT] - Design: [STUDY DESIGN] - Data they had: [DATA ACCESS OR IN MANUSCRIPT] - Sample n if known: [SAMPLE N OR UNKNOWN] </inputs> <task> Audit [PASTE MANUSCRIPT] for causal overclaim: prove, show, effect of, representative, dual coding they did not name, or a completed result written stronger than the table. For each: quote the line, name the overclaim, and rewrite a weaker line they can keep. </task> <constraints> - Do not invent n, a second coder, an instrument, or a finding. - Do not invent a methods citation to bless the design. - If [SAMPLE N OR UNKNOWN] is UNKNOWN, refuse a power or representativeness claim. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Table: quoted line | overclaim | weaker rewrite | major or minor. Then UNVERIFIED rows. </format>

<context> I need the methods section of my reviewer report, not the whole letter. Stay inside the protocol they pasted. Do not invent a second protocol. </context> <inputs> - Methods text: [PASTE METHODS] - Venue: [VENUE] - Design as I read it: [STUDY DESIGN] - Question or aim: [RESEARCH QUESTION] - My methods notes: [PASTE NOTES OR NONE] </inputs> <task> Write a methods-only review block: design fit, sample or corpus, collection, analysis, validity or trustworthiness, ethics as stated, and limits. Use future or past tense only as the authors did. Leave counts as UNKNOWN unless they pasted them. Number comments as major or minor. </task> <constraints> - Do not invent n, instruments, software, or a second reviewer. - If they worked alone, do not upgrade that to dual independent coding. - Do not write literature or contribution comments in this block. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Methods review block with numbered comments tagged MAJOR or MINOR, UNKNOWN cells, UNVERIFIED rows. </format>

<context> I have decided this manuscript can come back. Draft a revise-and-resubmit letter I can edit. Use only comments I can defend from the paste. </context> <inputs> - Manuscript: [PASTE MANUSCRIPT] - Venue: [VENUE] - Ranked comments I already accept: [PASTE COMMENTS] - Tone: [CIVIL / FIRM] - What would make me reject next round: [MUST FIX] </inputs> <task> Draft an R&R letter: a one-paragraph decision, the must-fix majors, the minors, and what a successful resubmission must show. Every comment must point at the paste. Do not add a paper they must cite unless that paper is already in my pasted comments. </task> <constraints> - Do not invent citations, findings, or a journal rule. - Do not soften [MUST FIX] into optional polish. - Do not promise accept after revision unless I said so. - If a comment in [PASTE COMMENTS] cannot be tied to the manuscript, drop it. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Decision paragraph, numbered majors, numbered minors, resubmission test, UNVERIFIED rows. </format>

<context> I am recommending reject. The letter must be specific and fair. Reasons must appear in the paste. Do not invent the paper that would have saved them. </context> <inputs> - Manuscript: [PASTE MANUSCRIPT] - Venue: [VENUE] - Fatal problems I can quote: [PASTE COMMENTS] - What still has value: [STRENGTHS I CAN QUOTE] - Tone: [CIVIL / FIRM] </inputs> <task> Draft a reject letter: the decision in the first paragraph, 2 to 4 reasons tied to quotes, one sentence on what the work still does, and what kind of study (not a title) might belong at this venue later. No invented citations. </task> <constraints> - Do not invent papers, findings, or a comparison article they failed to cite. - Do not pile minors as if they were the reject reason. - Do not insult the authors. - If a reason is not in [PASTE COMMENTS] or the manuscript, drop it. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Decision paragraph, numbered reasons with quotes, one strength sentence, next-study type (not a title), UNVERIFIED rows. </format>

<context> I am recommending accept with minor fixes. The fix list must be finishable without new data. Do not sneak a major into the minors. </context> <inputs> - Manuscript: [PASTE MANUSCRIPT] - Venue: [VENUE] - Minor fixes I can point at: [PASTE COMMENTS] - Any remaining doubt: [DOUBT OR NONE] </inputs> <task> Draft an accept-with-minors letter: the decision, why the contribution in the paste is enough for [VENUE], and a short fix list. If [DOUBT OR NONE] is a real methods doubt, stop and say this should be R&R instead of dressing it as a typo. </task> <constraints> - Do not invent citations or findings. - Do not add a new analysis as a minor. - Do not invent a journal style rule I did not paste. - If a listed fix cannot be done from the current paste, retag it MAJOR and abort accept. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Decision paragraph, contribution sentence from the paste, numbered minors, abort-to-R&R note if needed, UNVERIFIED rows. </format>

<context> I am writing the editor note that sits above the reviewer reports. Reconcile conflicts from the reports I pasted. Do not add a finding a reviewer did not write. </context> <inputs> - Manuscript: [PASTE MANUSCRIPT] - Reviewer reports: [PASTE REVIEWER REPORTS] - My decision: [R&R / REJECT / ACCEPT MINOR] - Venue: [VENUE] - Conflicts I already see: [CONFLICTS OR FIND THEM] </inputs> <task> Write editor notes to the authors: the decision, which comments are required, which are optional, and how to handle conflicts in [PASTE REVIEWER REPORTS]. Quote the reviewers. Do not invent a third report. </task> <constraints> - Do not invent a reviewer comment that is not in the paste. - Do not invent a citation to break a tie. - If two reviewers contradict and I did not pick a side, say so and pick from the manuscript, not from a paper I recall. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Editor note, required vs optional table, conflict rulings, UNVERIFIED rows. </format>

<context> I need to match my decision to the venue's own bar, not a generic prestige bar. Use only rules I paste. Do not invent a journal criterion. </context> <inputs> - Manuscript: [PASTE MANUSCRIPT] - Venue criteria or reviewer form: [VENUE CRITERIA] - My draft comments: [PASTE COMMENTS] - Draft decision: [R&R / REJECT / ACCEPT MINOR / UNDECIDED] </inputs> <task> Score [PASTE MANUSCRIPT] against each line in [VENUE CRITERIA]. For each line: met / not met / not in paste. Then say whether [R&R / REJECT / ACCEPT MINOR / UNDECIDED] still holds. If the draft decision is harsher or softer than the criteria, say so and propose the closer call. </task> <constraints> - Do not invent a venue rule I did not paste. - Do not invent a comparison paper from this journal. - If a criterion is not in the paste, write NEED CRITERION, do not guess. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Criteria table, verdict on the draft decision, closer-call paragraph, UNVERIFIED rows. </format>

<context> I have reviewer or editor comments and the manuscript they point at. I need a revision plan, not a finished reply. Do not invent sources to look responsive. </context> <inputs> - Comments: [PASTE COMMENTS] - Manuscript: [PASTE MANUSCRIPT] - Changes I already made or will make: [PASTE REVISION PLAN OR NONE] - Venue: [VENUE] </inputs> <task> Map each comment to: agree and change, ask for clarification, or decline with evidence already in [PASTE MANUSCRIPT]. Describe the change from [PASTE REVISION PLAN OR NONE] or write NOT YET DONE. If a reply needs a new source, write I MUST FIND A SOURCE, do not invent one. </task> <constraints> - Do not add papers the comments "reminded you of." - Do not claim a change that is not in the plan. - Do not treat every comment as law if it contradicts the pasted evidence. Say so. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Comment table (comment, decision, change or NOT YET DONE, FIND A SOURCE). Then UNVERIFIED rows. </format>

<context> I need a calm, point-by-point author response that reports changes I actually made or planned, from [PASTE REVISION PLAN]. Do not invent a citation to satisfy add literature. </context> <inputs> - Comments: [PASTE COMMENTS] - Revision plan or changed passages: [PASTE REVISION PLAN] - Venue or course: [VENUE] - Tone: [CIVIL / FIRM] </inputs> <task> For each comment: quote the comment, state agree / clarify / decline, describe the change in [PASTE REVISION PLAN] or say NOT YET DONE, and draft the response sentence. If a reply needs a new source, write I MUST FIND A SOURCE. </task> <constraints> - Do not claim a change that is not in the plan. - Do not invent citations to satisfy "add literature." - Do not insult the reviewer. - Decline only when the pasted evidence cannot support the requested claim, and say so plainly. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Numbered responses. Each: comment, decision, change, reply text. Then UNVERIFIED rows. </format>

<context> One reviewer comment is wrong or cannot be done given the pasted evidence. I need a polite decline, not a fight. Do not invent a paper that would win the argument. </context> <inputs> - The comment I may decline: [PASTE ONE COMMENT] - Manuscript evidence I already have: [PASTE MANUSCRIPT] - Why I think I can decline: [MY REASON] - Venue: [VENUE] </inputs> <task> Draft a decline: quote the comment, show the pasted evidence that already answers it or shows why the request cannot be met, and offer a smaller change if one exists. If my reason is stronger than the paste, weaken the decline or switch to NOT YET DONE, FIND A SOURCE. </task> <constraints> - Do not invent citations, findings, or a methods precedent. - Do not mock the reviewer. - If the paste does not support a decline, say so and draft an agree-and-change instead. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Quoted comment, evidence lines, decline draft, fallback agree draft, UNVERIFIED rows. </format>

<context> I need a short cover letter after revision. No sales fiction and no invented papers in the journal. Claim only changes that appear in my plan. </context> <inputs> - Revised manuscript or summary: [PASTE MANUSCRIPT] - Revision plan: [PASTE REVISION PLAN] - Venue: [VENUE] - Editor's last note if I have it: [PASTE EDITOR NOTE OR NONE] - Required statements: [ETHICS / CONFLICTS / NONE] </inputs> <task> Draft a cover letter: what the paper is, what we changed (only from [PASTE REVISION PLAN]), how we handled the editor note if pasted, and any required statements I listed. Do not claim fit from papers I did not name. </task> <constraints> - Do not invent similar papers in the journal. - Do not invent an editor's name. - Do not claim a change that is not in the plan. - Do not add citations unless I pasted them for the letter. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Cover letter, then a list of claims in the letter that I must still verify, then UNVERIFIED rows. </format>

<context> I need an honesty pass on an author response I already drafted. Does each reply match a change I can show? Do not invent the missing edit. </context> <inputs> - Author response draft: [PASTE AUTHOR RESPONSE] - Original comments: [PASTE COMMENTS] - Revised passages or plan: [PASTE REVISION] </inputs> <task> For each comment: tag REPLY MATCHES CHANGE, REPLY OVERCLAIMS, or COMMENT UNANSWERED. Quote the change in [PASTE REVISION] for MATCH. For OVERCLAIMS, rewrite the reply so it only claims what is in the revision. For UNANSWERED, draft a honest stub or NOT YET DONE. </task> <constraints> - Do not invent a change that is not in [PASTE REVISION]. - Do not invent a citation to close an unanswered comment. - Do not treat a planned change as done. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Table: comment | tag | quoted change or none | rewritten reply. Then UNVERIFIED rows. </format>