AI peer review of your manuscript, before a journal reviewer reads it
Fukurō runs an automated paper review on your full manuscript. Specialist passes read the argument, methods, statistics, references, and writing the way a skeptical reviewer would, and every finding quotes the passage that triggered it. Use it as a first pass before expert review, while the draft can still change.
Some of the comments that send a paper back concern problems the authors could have found alone: a percentage that does not match its own table, a Discussion with no limitations, a reference cited in the text and missing from the list. Authors miss them because they read their own draft as its writer, already knowing what each sentence was meant to say. An AI peer review reads the manuscript with no such memory.
Fukurō is built for that outside read. It will not rewrite your paper or predict what an editor will decide. You get a prioritized list of problems, each tied to a quotation from your manuscript, so you can check the reasoning yourself and choose what to fix.
If your concern is readiness for one particular journal (reporting checklists, required statements, desk-rejection risk), the pre-submission manuscript review page walks through the same report from that angle.
What the review checks
The review runs as a set of passes over the whole manuscript. Which passes run depends on your field, which is detected from the text, and on your plan.
Argument and claims
Whether your conclusions follow from the evidence you present, where claims run ahead of the data, and where a reader loses the thread between sections.
Methods and reproducibility
Design choices a reviewer would question, method steps described too loosely to repeat, and missing baselines or controls. Reproducibility gets its own rating.
Statistics that can be recomputed
Percentages, sums, and p-values are extracted and recomputed from the numbers and test statistics you report. Reporting is also checked against SAMPL guidance, for example an effect size given without its confidence interval. Fields without quantitative results skip this pass.
Consistency across sections
Contradictions a single-section read cannot see: a sample size in the abstract that differs from the one in Results, a term defined one way and used another, a number that changes between a table and the text.
What the manuscript leaves out
A dedicated pass looks for absences, which a line-by-line read tends to miss: no limitations, no data or code availability statement, no discussion of threats to validity.
References
In a numbered reference list, entries the text never cites and citations that point to entries that do not exist.
Writing quality
Typos, acronyms that drift, and terminology or numbers that change between sections, limited to problems that cost the paper precision. Line-by-line language editing is a separate tool, the language check.
Passes for your field
Ten field profiles adjust what counts as a defect. Mathematics gets a proof-checking pass. Essays in the humanities and law get an argument-structure pass that looks for unsupported claims and ignored counter-arguments. Qualitative social science gets a rigor pass on sampling, saturation, coding transparency, and reflexivity, and a paper with pseudocode gets its listings audited against the prose.
Figures, on the Complete plan
Extracted figure images are read for legibility, missing axis labels, and captions that do not match what the figure shows.
Reporting guidelines, on the Complete plan
Life-sciences manuscripts are audited item by item against CONSORT, ARRIVE, STROBE, or PRISMA, chosen from the study design, and the report includes a compliance table.
How an automated paper review works
- 01
Upload the manuscript
Add a PDF, DOCX, or TXT file to a project, with the reference list included so the reference checks have something to work with. You see an estimated credit cost before the review starts.
- 02
Specialist passes read it, then a verifier challenges them
The main review and the specialist passes run in parallel, and every finding has to carry a verbatim quotation from your manuscript. A verification pass then does two things. It confirms that each quotation actually appears in your text, and it asks a separate model to refute the finding against the surrounding context. Refuted findings are removed. A finding whose quotation cannot be located stays in the report with a warning, so you can judge it yourself.
- 03
Read the report and act on it
Duplicate findings are merged, and what remains is ranked by severity and turned into an action plan. Read the report in the app, download it as a PDF for coauthors, or open a chat on the review to ask which issues to fix first. When you upload a revised version to the same project, open findings from the previous review are marked resolved or persisting.
What the report looks like
A report opens with an executive summary and a severity rubric, then gives each concern its own card. Severity runs from 5, a problem that invalidates a core claim, down to 1, which is polish. You can read three complete reports on open-access papers, with the original PDFs, on the sample reviews page before you upload anything.
Prioritized issues
One card per concern, grouped by severity. Each card shows the quoted passage and where it sits, why it weakens the paper, the standard it violates when one applies, a recommended revision, and whether the finding was verified.
Section-by-section guidance
Where the concerns cluster, so you can plan revision one section at a time.
Ratings
Scores out of 5 with a justification, on up to seven dimensions: technical soundness, novelty, organization and clarity, writing quality, statistics, reproducibility, and ethics and privacy. Labels adapt to your field, and a dimension that does not apply is left out rather than filled with a placeholder.
Action plan and self-audit checklist
The concerns restated as prioritized work items, each with a condition that tells you when it is done.
Guideline compliance, on the Complete plan
When a reporting-guideline audit ran, a table of every core checklist item with the evidence found for it or a note that it is missing.
Limitations and what to preserve
What the review could not assess, how many findings were independently verified against your text, and the strengths a revision should keep.
Who uses an AI paper reviewer
PhD students run it before a thesis chapter or a first paper goes to a supervisor. The supervisor's reading time then goes to the science and the framing, not to a missing limitations section or a table that disagrees with the text.
Researchers and postdocs run it between coauthor rounds and again on the version they plan to submit. It is especially useful when you write in English as a second language and want the structural problems settled before anyone polishes the prose. On the Complete plan, rebuttal analysis checks whether your response letter and revised manuscript address each reviewer comment, and manuscript comparison shows exactly what changed between two versions.
Professors and group leaders use it to give students a structured first read before their own, and as a consistent check on manuscripts before they leave the lab.
AI peer review compared with human peer review
A reviewer who works in your subfield can judge whether the contribution is new, whether the question matters, and whether you ran the right experiment. An AI peer review cannot make those calls reliably, and Fukurō does not claim to. The report does rate novelty, but deciding what counts as original against the literature of a subfield still takes an expert.
The automated review is strongest on the checkable part of reviewing: recomputing a p-value from its test statistic, finding the number that changed between the abstract and Table 2, noticing that the paper never states where its data can be found, reading a CONSORT checklist item by item. Human reviewers catch these too, but only once the paper is under review, and a round of comments about them costs a revision cycle.
So the order that works is automated review first, then colleagues, then the journal. Fix what the first pass finds and send the cleaner draft to coauthors and expert readers, who can spend their attention on the science. The report is a first pass before expert review and cannot guarantee acceptance.
How Fukurō compares with other AI review tools
Several tools now offer AI review of manuscripts, and they differ in what they read and what they return. Among the tools in our current comparison, Fukurō is the only one we identified with both rebuttal analysis and manuscript version comparison built in.
The comparison overview puts Fukurō and 9 other tools side by side, feature by feature. For single matchups, read Fukurō vs Reviewer3, Fukurō vs Refine.ink, or Fukurō vs PeerReviewAI. Each page draws on the other tool's own product pages and names the kind of author it suits best.
Frequently asked questions
What is automated paper review?
An automated paper review uses AI to read a full manuscript and return reviewer-style findings before submission. In Fukurō, a main review and several specialist passes cover argument, methods, statistics, consistency, references, and writing, and a verification pass checks that each finding is anchored to a quotation from your text.
Is an AI peer review as good as a human reviewer?
No, and it is not meant to be. It is a first pass before expert review. It does best on problems that can be checked against the text, such as recomputable statistics, internal contradictions, and missing statements. Judging novelty and significance in your subfield still needs human experts.
Which file formats can I upload?
PDF, DOCX, and TXT. Upload the version you plan to submit, with its reference list, so the reference checks have a list to compare against the citations in the text.
Does it rewrite my paper?
No. Fukurō assesses what you wrote and points out problems. Recommended revisions describe what to change, and some cards include a short example of weaker and stronger phrasing, but it never rewrites your manuscript. How you resolve each finding is your work.
How do you deal with invented or wrong findings?
Every finding must quote the manuscript. The quotation is matched against your text, and a separate model tries to refute the finding in context. Refuted findings are dropped, and a finding whose quotation cannot be located is labelled so you can double-check it before acting. The limitations section of each report states how many findings were independently verified.
Which features need the Complete plan?
Figure evaluation, reporting-guideline audits, rebuttal analysis, and manuscript version comparison are on the Complete plan. The Standard plan includes the full text review with the statistics, consistency, reference, and writing checks, and the separate language check.
Will my manuscript be used to train AI models?
Not by default. Your manuscript is sent to our AI providers only to produce your review, and they do not train on it. Extracted content is automatically deleted from our systems within three days. A few optional chat models come from providers that do train on submitted content; we flag those and ask for your explicit consent before you can use one.
Will a journal know I used it?
Fukurō does not share your use with journals, coauthors, or your institution. Some journals ask authors to disclose AI tools used while preparing a manuscript, so check your target journal's policy and disclose where it asks you to.
Run the first pass before the journal does
Upload a manuscript, read the findings, and decide what to change. Plans and their monthly credit allowances are on the pricing page.