Peer Review Checklist for Authors: Score Your Own Research Paper the Way Reviewers Do

Peer review is slow, unpaid and stretched. In 2020, reviewers worldwide spent more than 100 million hours on manuscripts, and US reviewers’ time alone was worth over US$1.5 billion [1]. The load is uneven, too: across 2,329 Elsevier journals, women submitted proportionally fewer papers than men in the first COVID-19 wave while taking on more review work [2].

A busy reviewer reads quickly and relies on visible signals: a clear question, a defensible design and transparent reporting. A structured manuscript self-assessment helps you avoid:

• a desk rejection for scope, format or missing reporting items;
• a first review round spent on presentation problems you could have fixed;
• comments about missing ethics, registration, funding or data statements;
• abstract claims the results do not support.

Reviewer bias is measurable. Huber and colleagues sent the same finance paper to reviewers under three conditions. With a Nobel laureate named as author, about 23% recommended rejection; with a junior researcher named, more than 65% did; with no author shown, almost 50% did [3].

You cannot change your reputation before submission, but you can remove every other reason to reject. State the contribution early, label each figure so it stands alone, and make the methods easy to find.

Table 1 lists the questions reviewers commonly ask. Score each item 0 (missing or weak), 1 (present but incomplete) or 2 (clear and complete). Be strict: if you would need to explain an answer aloud, it is not yet on the page.

Item 6 needs particular care. A review of author instructions at 559 biomedical journals from 11 publishers found that endorsement of reporting guidelines was inconsistent, with no standard way to show adherence [4]. Identify the guideline yourself: CONSORT for randomised trials, PRISMA for systematic reviews and STROBE for observational studies.

Treat the total as a guide, not a verdict. A paper can score 18 and still fail on one fatal flaw, such as a design that cannot answer the question. Any 0 on design, statistics or ethics should stop submission until it is fixed. Re-score after revision to check that the edits worked.

Large language models now produce reviewer-style feedback, and the evidence shows both value and limits. For 3,096 Nature-family papers and 1,709 ICLR papers, GPT-4’s comments overlapped with human reviewers’ points by 30.85% and 39.23%, similar to the overlap between two humans (28.58% and 35.25%). In a study of 308 researchers, 57.4% found the feedback helpful [5].

LLM peer review is already happening: an estimated 6.5% to 16.9% of review text at four AI conferences may have been substantially modified by language models [6]. AI scores are less reliable than AI comments. When ChatGPT-4 scored 51 articles against REF 2021 criteria, a single run correlated with expert assessment at r = 0.281, and the average of 15 runs at r = 0.509 [7].

• Use AI to flag unclear passages and missing reporting items, and to list the questions a sceptical reviewer will ask.
• Do not use AI to set your final score, or upload unpublished data without co-author agreement and the journal’s permission.

Check the target journal’s policy on AI use before you start.

1. Read only the abstract and conclusions and confirm they make the same claim (5 minutes).
2. Score the ten items in Table 1 (10 minutes).
3. Open the relevant reporting checklist and record a page number for each item (10 minutes).
4. Run one AI second read for unclear passages and verify every suggestion yourself (5 minutes).
5. Fix the weak items, re-score and submit.

Keep the scored checklist with your submission files. If reviewers raise a point you already scored, the record speeds up revision and helps you write a precise response to reviewer comments.