Launch bonus Yearly plans 20% 40% off

Peer Review a Paper

A referee report rests on two kinds of reading: what the paper argues, and whether its evidence carries the conclusions. This guide gets you a first draft of both in a few minutes. A ready-made prompt in the reader writes a structured AI peer review of one paper, with every concern pinned to the page it comes from, and an optional second pass verifies the paper's claims against the rest of your library.

The screenshots come from a real run on an open-access experiment, Walsh et al. (2024), "Does putting down your smartphone make you happier?", 25 pages, using the default model.

Before you start

  • Check the journal's policy first. Many journals do not allow reviewers to upload a confidential manuscript to an AI tool. For a manuscript under review, ask the editor or follow the journal's guidance. Published papers and preprints are fine.
  • Add the paper to a library and wait until it is indexed. See the Quickstart.
  • Optional: add the papers it cites, if you have them, for step 3.

Step 1: Ask for a referee report

Open the paper from your library to bring up the reader, then open the Chat tab on the right. Pick the ready-made prompt Peer review.

The prompt asks the agent to write as a journal referee in the paper's field, in five parts:

  1. Summary of the paper and its contribution
  2. Major concerns that affect the validity of the conclusions, each with what the authors should do
  3. Minor concerns about clarity, presentation, and missing detail
  4. Requested revisions
  5. Recommendation: accept, minor revision, major revision, or reject, with a justification

Send it as it is, or add a line about what your journal cares about, such as "Pay special attention to the statistical analysis."

Step 2: Read the report against the pages

The agent reads the paper page by page and searches it for detail before it writes. On Walsh et al., the report took under three minutes and cost 1.1 credits.

The major concerns were substantive. The report pointed out that 33 participants were excluded after randomization because their use went up, and asked for an intention-to-treat analysis. It flagged the number of outcomes and contrasts and asked how false positives were controlled. It noted that the title says "happier" while the happiness measures did not change, and asked the authors to say which outcomes actually improved.

Every point cites the pages it rests on. Click each citation and read the passage before you keep a concern. A concern you cannot find on the page does not go into your report.

Treat the report as a draft. Your expertise decides which concerns matter, which are wrong, and what the agent missed.

Step 3 (optional): Verify the paper against your library

The referee report reads the paper on its own terms. To test whether its claims agree with the wider literature you hold, map its claims and check them:

  1. In a project whose libraries hold the paper and related work, click Mindmap and Review a Paper and click the paper.
  2. Keep Check support and Check the reasoning ticked, and click Run. Every claim gets a verdict against your library, and the root gets a verdict on whether the paper's conclusions follow from its claims. See Mindmap and Review a Paper.

A Contradicted or Contested verdict on a claim about prior work is often the most useful line in a review: it shows where the paper's account of the literature does not match the papers it cites. This step needs the Starter plan or higher.

Follow-up questions

Keep asking in the same chat. Each answer cites its pages.

Which of the paper's conclusions rely on a single analysis or a single outcome measure? Cite the page for each.

List every exclusion criterion and when it was applied, before or after randomization.

Does the discussion claim anything the results section does not report? Quote both pages.

Write three questions to the authors that would most improve the paper.

The ready-made prompt Methods critique is a shorter alternative that focuses on whether the design can support the conclusions.

Tips

  • Ask for page numbers you can check, and check them. The report is only as useful as the passages behind it.
  • Use a stronger model for a final pass. Pick a different model in the chat box when the review matters.
  • Keep your report in your own words. Use the agent's concerns as a checklist, then write the review yourself.

What's next