Manuscript Preparation & Peer Review

Advanced · Communication & Dissemination track · ~35 min hands-on + readings and quiz

← Back to the Communication & Dissemination track

Use AI to critique and strengthen your own manuscript — within strict authorship, confidentiality, and disclosure rules.

What you’ll be able to do

  • Brainstorm and critique your own manuscript sections the way a peer reviewer would — not have AI draft them
  • Apply reporting checklists (STROBE/CONSORT) with AI as the critic, and use AI for structural tasks like organizing reviewer responses
  • Know what ICMJE and COPE require you to disclose, and respect journal-specific AI policies in peer review

Overview

Use AI to brainstorm framings, search and organize literature, and critique your own draft the way a peer reviewer would — not write the first draft of your science. AI should not draft your methods, results, or interpretation from scratch. That’s not just a compliance issue: writing those sections yourself is how you keep authorship, originality, and accountability over the work. The workflow that pays off is the one in Activity 1 — you write the section, then AI stress-tests it against a reporting standard and tells you what’s missing.

Structural and administrative tasks are a different matter. Having AI organize a point-by-point response-to-reviewers letter from decisions you’ve already made, or tighten language you’ve already written, is a reasonable use — the substance is yours; AI is just formatting and polishing it.

Peer review has a hard rule: a manuscript you’re reviewing is confidential. Never paste it into any AI tool — and note that many journals prohibit any AI use during review, not just uploading the manuscript. Check the journal’s reviewer guidelines before touching an AI tool in a review.

ImportantRead before you draft: authorship, disclosure, and journal policy

AI cannot be an author. Both ICMJE and COPE are explicit: AI tools cannot be listed as authors, because they cannot take responsibility for the accuracy, integrity, and originality of the work. The human authors are fully responsible for all content, including any AI-assisted portions — you must review AI output (it can be incorrect, incomplete, or biased), verify it isn’t plagiarized, and ensure citations are real and accurate.

Disclose your AI use. ICMJE says disclosure belongs in both the cover letter and the appropriate section of the manuscript itself; failing to disclose may require corrective action and can be construed as misconduct. COPE requires documenting transparently which tools were used, typically in the methods or acknowledgments section.

Journal policies vary and change often. Some journals are stricter than the ICMJE/COPE baseline — some prohibit any AI use in peer review outright, and some restrict which sections or data may touch AI tools at all. Check your target journal’s specific author guidelines and reviewer guidelines every time you submit or review.

Peer review confidentiality is absolute. Never paste a manuscript you’re reviewing into any AI tool; many journals prohibit any AI use in review, not just uploading the text.

The recommended drafting workflow is Activity 1 below: you write the section, AI critiques it against a standard — not the other way around.

Practice activities

Activity 1 · Advanced — Tighten a section against a reporting checklist

Time ~18 min · Tools ChatGPT Edu

Goal. Practice the sanctioned pattern — you write it, AI critiques it against a standard. Use a reporting guideline to strengthen a methods section you have already drafted yourself.

Setup. Your own draft methods section (non-sensitive) and the relevant checklist (e.g., STROBE for observational studies).

Steps.

  1. Find the gaps:

    Here is my methods section and the STROBE checklist. Which checklist items are missing, unclear, or under-reported? Be specific. [methods] [checklist]

  2. Revise to address them yourself; confirm each change is true to what you actually did.

  3. Improve reproducibility:

    Suggest where I should add detail to make this reproducible, without padding.

Expected result. A checklist-aligned methods section — still written by you, sharpened by an AI critique.

Check your work. Every added detail must match your real methods — don’t let AI invent procedures you didn’t use.

Common pitfalls. AI “improves” by adding things you didn’t do. You own accuracy. Use an approved tool for unpublished work.

Stretch (optional). Generate a STROBE/CONSORT compliance table to submit with the paper.

Activity 2 · Advanced — Structure a response to reviewers

Time ~15 min · Tools ChatGPT Edu

Goal. Organize a clear, point-by-point reviewer response — while respecting confidentiality. This is a structural/administrative use of AI, not substantive drafting: every scientific decision in the letter is one you already made; AI only formats and organizes it.

Setup. Use this fictional methods excerpt and reviewer comments (or substitute your own). Never paste a manuscript you are reviewing for a journal.

Fictional methods paragraph:

We conducted a cross-sectional analysis of pedestrian crashes in Philadelphia from 2018–2023 using PennDOT crash records linked to road-segment characteristics from OpenStreetMap. Fatal and serious injuries (KSI) were modeled using logistic regression. Covariates included lane count, posted speed limit, crosswalk presence, and lighting. Census-tract income quartile was included as a fixed effect. We excluded crashes on highways and those involving intoxicated pedestrians. Missing speed-limit data (18% of segments) were imputed using the median of adjacent segments.

Fictional reviewer comments:

Reviewer 1:

  • Major: The exclusion of intoxicated pedestrians is unexplained and potentially introduces selection bias. Please justify or conduct a sensitivity analysis including these cases.
  • Major: Imputing speed limits using the median of adjacent segments is ad hoc. How sensitive are results to this choice? Consider a sensitivity analysis with complete cases only.
  • Minor: Table 2 header says “OR” but the footnote says “adjusted OR” — which is it?

Reviewer 2:

  • Major: The cross-sectional design limits causal claims, but the discussion reads as if these are causal findings (“reducing lanes prevents KSI”). Reframe.
  • Minor: Why OpenStreetMap rather than PennDOT’s road inventory? Address data-quality concerns.
  • Minor: Line 142 — “significant” used without a p-value or CI.

Steps.

  1. Organize the response:

    Organize these reviewer comments and my planned responses into a professional, point-by-point response letter grouped by reviewer, with polite framing. For each comment, acknowledge the concern, describe the revision, and note the page/line where the change was made. Reviewer comments: [paste above]. My planned responses: Reviewer 1 Major 1 — we will add a sensitivity analysis including intoxicated pedestrians. R1 Major 2 — we will add a complete-case sensitivity analysis. R1 Minor — it’s adjusted OR; we’ll fix the header. R2 Major — we’ll reframe to associational language. R2 Minor 1 — we chose OSM for coverage; we’ll add a data-quality note. R2 Minor 2 — we’ll add the CI.

  2. Verify every “we have revised…” statement: do you actually plan to make each change? Don’t promise a revision you won’t do.

  3. Soften where needed:

    Flag any response that sounds defensive and suggest a more constructive phrasing.

Expected result. A clean, professional response-to-reviewers letter organized by reviewer, with polite framing that acknowledges each concern before describing the revision.

Check your work. Confirm each claim of a change matches what you’ll actually revise. The most common error is promising a sensitivity analysis you don’t actually run.

Common pitfalls. Hard rule: never put a manuscript you’re peer-reviewing into an AI tool — it’s confidential and many journals prohibit any AI use in review. Separately, as an author, any AI use in preparing your own manuscript (including help organizing this letter) should be disclosed per the target journal’s policy and ICMJE/COPE guidance — that’s a different rule from review confidentiality, and both apply.

Stretch (optional). Draft a short cover letter to the editor summarizing the three major revisions.

Activity 3 · Advanced — Cross-check your manuscript package with Claude Cowork

Time ~20 min · Tools Claude Cowork (personal/PI-purchased — not Drexel-supported; Low Risk Data only)

Goal. Extend the “AI as reviewer” pattern across files instead of one document. Before submission, your methods section, supplementary materials, and reporting checklist must agree with each other — journals and reviewers absolutely catch internal inconsistencies. Cowork connects to a folder and reads all three at once, so it can flag where they disagree.

Boundary this activity does NOT cross. This is your OWN manuscript package only. The hard rule from the top of this page applies to Cowork with zero exceptions: never put a manuscript you are peer-reviewing into any AI tool — not a chat window, not a connected folder. Confidential peer-review material never touches Cowork. And because Cowork is a consumer tool approved for Low Risk Data only, practice with the clearly fictional files below, not real unpublished work, unless your use has been separately approved.

Setup. Create a folder (e.g., cowork-practice/) containing three small files, then connect Cowork to that folder. Use these fabricated excerpts — one inconsistency is planted deliberately:

methods-excerpt.txt:

We conducted a cross-sectional analysis of pedestrian crashes in Philadelphia, 2019–2023, using PennDOT crash records linked to OpenStreetMap road segments. Posted speed limit was analyzed as a three-level categorical variable (<25 mph, 25–35 mph, >35 mph). Models adjusted for segment length, land use, and time of day. Segments with missing speed limit data (n = 214) were excluded.

supplementary-table-S1.txt:

Table S1. Variable definitions. Posted speed limit: continuous, mph, from OSM maxspeed tag. Segment length: continuous, meters. Land use: 4 categories (residential, commercial, industrial, mixed). Time of day: 4 categories. Sidewalk presence: binary, from OSM sidewalk tag.

strobe-notes.txt:

Item 7 (Variables): all covariates defined in Methods para 2 and Table S1. Item 12 (Statistical methods): missing data handled by complete-case analysis. Item 6 (Participants/setting): eligibility = all mapped road segments with a crash record, 2019–2023.

Steps.

  1. Connect Cowork to the folder, then ask for a consistency audit:

    These three files are the methods excerpt, supplementary variable table, and STROBE notes for my own manuscript. Compare them against each other and list every place they disagree — variables defined differently across files, variables in one file but not another, checklist claims the other files don’t support. Cite file and line for each.

  2. Review the findings. It should catch the plant: speed limit is categorical in methods but continuous in Table S1 — and may also flag that sidewalk presence appears in Table S1 but never in methods.

  3. Decide the fix yourself (which file reflects what you actually did?), edit the files yourself, then re-run:

    I’ve revised the files. Re-check the folder and confirm whether any inconsistencies remain.

Expected result. A file-and-line list of internal contradictions in your submission package — the same critique a careful reviewer would make, caught before submission instead of after.

Check your work. Verify every flagged inconsistency by opening the files yourself — and resolve each one toward what you actually did in the study, not toward whichever wording the AI prefers. You decide the fix; Cowork only points.

Common pitfalls. Connecting Cowork to a folder that also contains confidential material (a manuscript under review, student records, identifiable data) — Cowork can read the whole folder, so make a clean practice folder. Letting the AI “fix” the files for you and silently rewrite your methods. Forgetting that Cowork is Low Risk Data only.

Stretch (optional). Add a fictional abstract as a fourth file and ask Cowork to check that every number in the abstract appears identically somewhere in the other files — abstract/text mismatches are among the most common reviewer catches.

Check your readiness

Answer these, then check — your score suggests whether to dive in or skim the readings first.

Useful resources