For peer reviewers

Built for peer reviewers

Your editor sent you a 40-page manuscript on a Friday. Your deadline is Tuesday. Run the AI pre-pass, then spend your human time on what actually matters.

8
Specialist agents
15 min
Review in <
69K
Training reviews
5+
Reporting checklists

How experienced reviewers use the tool

Peer review is triage plus depth. Experienced reviewers allocate the first 60 minutes to triage (what's the claim? is it defensible? is the methodology sound in principle?) and the next 3–5 hours to depth (do the statistics hold? are the figures honest? is the literature complete?).

The AI pre-pass replaces most of the triage. You get back a structured report on 8 dimensions before you crack the PDF. Your 4 hours of deep reading become 4 hours of deep reading, not 90 minutes of triage plus 2.5 hours of deep reading.

What the 8 agents check for you

The agents mirror the standard rubrics of 19+ peer-review platforms we trained on:

  • Methodology — CONSORT/STROBE/PRISMA/ARRIVE/MIAME compliance, power calculations, randomisation reporting.
  • Formulas & Equations — dimensional analysis, algebraic consistency, derivation shortcuts.
  • Originality — overlap with prior work across 6 external databases and a 4.4M-paper institutional library.
  • Literature Coverage — citation completeness against OpenAlex's 250M-work corpus; surfaces missed classics and recent adjacent work.
  • Reproducibility — code availability, dataset accessibility, methods-section sufficiency.
  • Clarity & Language — readability, structural flow, non-native-friendly (we deliberately avoid penalising idiom).
  • Figures & Tables — caption completeness, colourblind safety, appropriateness of visual encodings.
  • Prior Publication — parallel fan-out to CrossRef, Unpaywall, arXiv, medRxiv, bioRxiv, and our library.

The honest use pattern

We do not suggest you copy-paste AI output into a review report. Editors don't want that, authors don't want that, and frankly the AI is wrong enough of the time that you shouldn't either.

Instead:

  • Read the agent summaries before the manuscript. Form a prior.
  • Read the manuscript. Confirm or refute each agent's flagged issue.
  • Write your own review, anchored in the 3–5 issues that actually matter after reading.
  • Use the AI output as a 'did I miss anything' second-pair-of-eyes pass before submitting to the editor.

Editor-view and platform integrations

If you're reviewing for a journal that uses our review engine end-to-end, the editor dashboard gives you the agent reports alongside author-declared reporting-checklist statements. For legacy journals (eLife, PLOS ONE, Nature Communications, OpenReview), you can run the AI pre-pass on the PDF in parallel with your own workflow — the API is stable and the JSON output is easy to parse.

Frequently asked questions

Most journals have now published explicit policies. The consensus (Elsevier, Wiley, Springer Nature, eLife) is: using AI as a reading aid is fine; passing AI-generated text as your own review is not; uploading the manuscript to a public tool where it could train a model is forbidden. Our tool is single-tenant and manuscripts never enter the training corpus. Follow your specific journal's letter-of-policy.
Apply to reviewRead the engineering blog

Command palette

Jump anywhere, run any action.