Research

Reviewing a thesis or a manuscript

Three-level review in Word tracked changes, verified bibliography, arbitration of co-author feedback, always on a new version.

WordPubMedSkill1 to 2 h instead of a full day

The problem

Manuscripts from students and co-authors arrive as .docx files, often in several crossed versions. The substantive review is drowned in formatting corrections, and the bibliography is rarely checked.

Life cycle of a manuscript

Figure. From draft to decision, without ever overwriting a version.

Step-by-step method

  1. Never overwrite: every review produces <name>_INITIALS-YYYYMMDD.docx next to the original.
  2. Three levels of review:
    • substance: question, methods, consistency between results and conclusion, key messages;
    • structure: outline, figures and tables, length;
    • form: language, abbreviations, journal guidelines.
  3. Every critical finding is proven: a word-for-word excerpt with section and page, and a status of observed, calculated or interpreted. An excerpt that cannot be found word for word means the finding is withdrawn. Every statement in the abstract is checked against the tables.
  4. Native Word tracked changes and comments, attributed to your name, so that the student sees and accepts each correction. Existing comments are preserved.
  5. Check the bibliography: each reference is checked in PubMed (existence, authors, year, DOI); references that cannot be found are flagged, never corrected at random. Reference callouts carried over from an old list are audited: does each number still point to the right reference?
  6. Arbitrate co-author feedback: Claude reads the body of the email as well as the comments, merges them, flags contradictions and proposes a coded decision for each piece of feedback (accept, partly, to check, decline), which you settle. The additional analyses requested are summarised separately: feasible or not, blocking or not.
  7. List of blocking points before submission: journal guidelines, declaration of interests, ethics committee, data availability.
  8. Proof of completion: the delivered file is reopened, the changes and comments are counted, their author is checked, and one page is rendered as an image for inspection.

Deliverable

A new reviewed version in tracked changes, and a summary note of the blocking points.

Safeguards

The unpublished manuscript remains confidential. The scientific writing remains the authors’ own: the AI reviews and suggests, it does not sign. Follow the journal’s policy on the use of AI.

Starter prompt

Here is the manuscript (V5) and the comments of two co-authors. Do a three-level review in tracked changes under my name, check each reference in PubMed, and list the contradictions between co-authors. Save under a new name.

← Back to the triptych