Research paper review: how to check your own paper before submission

This guide is for graduate students and early-career researchers about to submit a paper, thesis chapter or proposal. It sits in our series on reviewing and publishing, next to how to respond to reviewer comments, which covers what to do once the real reviews arrive.

What is a research paper review?

A research paper review is a structured reading of a manuscript against the criteria journals use to decide whether it is sound, clear and worth publishing. Peer reviewers do it for the journal. You can do the same for yourself before submission, and catch the problems that lead to desk rejection or a long revision.

A self-review is a research paper review you run on your own draft, in planned passes, before anyone else reads it. The point of passes is focus: when you read for argument you ignore commas, and when you proofread you ignore the argument.

What do reviewers look for?

Journal reviewers ask the same core questions across fields:

Reporting guidelines turn the "complete reporting" row into a checklist. The EQUATOR Network lists them by study type: CONSORT for randomized trials, STROBE for observational studies, PRISMA for systematic reviews, STARD for diagnostic accuracy, SRQR and COREQ for qualitative research, CARE for case reports and ARRIVE for animal research [1]. Find yours before you start.

How to review your own paper: the five passes

Run five passes in this order, from the biggest problems to the smallest, and do not fix commas until the structure is settled.

1. Rest the draft. Leave it for at least two days. You need distance to read what you wrote rather than what you meant.
2. Structure pass. Check the shape of the paper.
3. Argument pass. Check that every claim follows from evidence.
4. Sources pass. Check every citation and reference.
5. Language pass. Edit and proofread.
6. Fit pass. Check the paper against the journal's instructions.

Then get a second reader.

Pass 1: structure check

Read only the title, abstract, headings, first sentence of each paragraph, figure titles and the conclusion. Together they should tell the whole story.

• Does the last paragraph of the introduction state the question or aim, and the gap it fills?
• Does each heading match the journal's expected order (often IMRaD: Introduction, Methods, Results and Discussion)?
• Does every result in the results section answer something set up in the introduction or methods?
• Does the discussion open with the main finding, not a repeat of the background?
• Does the abstract give the same numbers as the results?

Pass 2: argument check

Underline every claim in the introduction and discussion. For each, ask: what supports this, a citation, my data or nothing?

• Overreach: "X causes Y" from a cross-sectional survey. Change to "X was associated with Y".
• Missing limits: a limitation a reviewer will spot that you don't name. Name it first.
• Unanswered alternatives: another explanation for your result that you don't discuss.
• Method mismatch: a design that can't answer the question as written. Fix the question or the claim.

Pass 3: citation and source check

This is the pass people rush, and the one that produces embarrassing reviews.

1. Every claim that needs a source has one. Background statistics, definitions and "previous studies show" sentences all need citations.
2. Every citation says what you say it says. Open the cited paper for each key claim. Citing a review for a finding that came from a primary study, or citing the abstract's conclusion when the results are weaker, are common errors.
3. Every reference exists and is correct. Check authors, year, title, journal and DOI. Resolve each DOI at doi.org to confirm it leads to the right paper.
4. No retracted papers. Crossref makes the Retraction Watch database of retractions, corrections and expressions of concern publicly available [2]. Check every reference, not just the ones you doubt.
5. Recent and balanced. Look for studies from the last two or three years and for studies that disagree with you.
6. In-text citations match the list. Every citation is in the reference list and every reference is cited, in the journal's style.

Pass 4: language and tone

• Cut sentences over about 35 words into two.
• Define each abbreviation once, at first use, and use it consistently.
• Use one name for each variable, group and measure throughout.
• Check hedging: strong enough to say something, careful enough to be true ("suggests" for one study, "shows" for a body of evidence).
• Check tense: past for what you did and found, present for established knowledge and what your figures show.

How to proofread a research paper: proofread last. Change the font or print the paper so it looks unfamiliar, read it aloud, and check one thing per read: spelling and grammar, then every number in the text against its table, then abbreviations, then reference formatting.

Pass 5: journal fit

Open the journal's instructions for authors next to your manuscript.

• Word limit, abstract format (structured or not) and maximum number of tables, figures and references.
• Required statements: ethics approval, consent, funding, conflicts of interest, data availability, author contributions.
• AI use statement. The ICMJE recommends that journals ask authors to disclose AI-assisted technology use at submission, and says authors stay responsible for everything submitted [3]. Follow your journal's exact wording.
• Is the journal trustworthy? The Think. Check. Submit. checklist covers peer review, fees, indexing and publisher identity [4].

A self-review checklist you can copy

Copy this into your notes and tick it off for each submission.

Worked example: planning a self-review (invented)

Example (invented, for illustration). Amira has a 7,400-word observational study for a public health journal, due in a week.

• Day 1: rests the draft. Downloads the STROBE checklist and the journal's instructions.
• Day 3, morning: structure and argument passes (2 hours). She finds the discussion claims an "effect" from cross-sectional data and rewrites 4 sentences.
• Day 3, afternoon: sources pass (2.5 hours). Of 46 references, 2 have wrong years, 1 DOI leads to a different paper and 1 cited study has an expression of concern. She replaces it.
• Day 4: language pass and journal fit (1.5 hours). She cuts 650 words to meet the 7,000-word limit (main text, excluding references) and adds the data availability statement.
• Day 5: a colleague reads it; she fixes two unclear paragraphs and submits.

Can an AI reviewer review my research paper?

An AI reviewer is a tool that reads a manuscript and comments on it against review criteria, the way a peer reviewer would. It is useful as a second reader when no colleague is free, or as a check before a colleague reads it, so their time goes on the questions only a person in your field can answer.

Use it well:

1. Run it after your own passes, not instead of them.
2. Treat each comment as a question. Accept a fix only when you agree with the reason.
3. Ignore it on novelty. An AI reviewer can't tell you whether your contribution is new to your field's last six months.
4. Check your journal's and institution's AI policies, and disclose use if they require it [3]. Unpublished work is confidential: read a tool's data policy before uploading a manuscript.

Review your paper in AI for Research

AI for Research has a Reviewer workspace and free reference checks. Costs as of 12 November 2026:

• Reviewer reads a thesis, proposal or manuscript at the level you choose (undergraduate, honours, master's or PhD). It scores five criteria out of 10: structure and argument, method and statistics, use of sources, clarity and style, and formatting. Each comment quotes the passage, names the section and suggests a fix; a quote that isn't in your text is dropped. Standard review costs 10 credits per 10,000 words, High quality 30 (Pro). A preview of the first 3,000 words, with the full scorecard and top 3 comments, costs 3 credits. See AI peer review.
• Citation check confirms each reference exists (Crossref, OpenAlex) and flags retractions, corrections, expressions of concern, preprints and wrong years or authors. 0 credits; the optional AI check that citing sentences match the abstract costs 2. See the citation checker.
• Claim check labels up to 10 claims from a paragraph as supported, partly supported, not supported or no evidence, with the papers (2 credits).
• Projects let a co-author or supervisor edit or comment on the draft.

Using the example above: Amira's 7,400-word manuscript costs 10 credits for a Standard review, and the citation check costs nothing. The free plan has 45 credits a month. See pricing.

Limits: Reviewer doesn't know your journal's scope or your field's newest work, and it doesn't check the reporting guideline item by item. The judgment on every comment stays with you and your co-authors.

FAQ

How do I review my own research paper?

Leave the draft for at least two days, then read it in five passes, each with one job: structure, argument, sources, language and journal fit. Use a checklist so you don't drift into line editing on the first pass. Fix big problems before small ones, because rewriting a section makes earlier proofreading wasted. Finish with a second reader, a colleague or an AI reviewer.

How do I proofread a research paper?

Proofread last, after structure and content are settled. Print the paper or change its font so it looks unfamiliar, read it aloud slowly, and check one thing at a time: spelling and grammar, then numbers against tables, then terms and abbreviations, then reference formatting. Search for your common errors, such as doubled words or inconsistent spelling of key terms.

What do reviewers look for in a research paper?

Reviewers look for a clear question that matters to the journal's readers, a method that can answer it and is described well enough to repeat, results that support the conclusions, honest limitations, and a fair use of the literature. Many also check that the paper follows a reporting guideline such as CONSORT or STROBE and that the statistics are reported fully.

Can an AI reviewer check my research paper?

Yes, as a second reader. An AI reviewer can score a manuscript against criteria such as structure, method and clarity, and quote weak passages with a suggested fix. It can't judge whether your contribution is new in your field or replace your co-authors and supervisor. Check your journal's policy on AI use and disclose any use the policy requires.

How long should a self-review take?

For a 6,000 to 8,000-word manuscript, plan about one working day spread over two or three sittings: an hour each for structure and argument, two hours for sources and references, an hour for language, and half an hour for journal fit. Add time to fix what you find. Rushing the reference check is the most common false saving.

Next steps

• How to respond to reviewer comments: what to do when the reviews come back.
• Thesis to journal article: turning a chapter into a paper before you review it.
• How to check a claim against the research: the method behind pass 3.
• How to write a research paper: structure and steps, if the structure pass finds big gaps.

Sources

1. EQUATOR Network. Enhancing the QUAlity and Transparency Of health Research: reporting guidelines for main study types. https://www.equator-network.org/ (checked November 2026)
2. Crossref. Retraction Watch data. https://www.crossref.org/documentation/retrieve-metadata/retraction-watch/ (checked November 2026)
3. International Committee of Medical Journal Editors. Artificial intelligence: AI use by authors. https://www.icmje.org/recommendations/browse/artificial-intelligence/ai-use-by-authors.html (checked November 2026)
4. Think. Check. Submit. Journals checklist. https://thinkchecksubmit.org/journals/ (checked November 2026)