Review summary

Summary

Last updated: 2 November 2026

Alongside the usual author-written abstract, each accepted paper is published with a review summary: a short, reader-oriented synthesis of what the review process learned. It neither perfunctorily summarizes the paper nor issues a verdict on it. Rather, it aims to be the summary a potential reader would most want to read before reading the full paper, conveying strengths, weaknesses, caveats, audience fit, and relationship to prior work — the information otherwise lost when the output of a confidential review is compressed to a single accept/reject bit.

The action editor asks one reviewer to write it, for additional compensation, drawing freely on the contributions of the editor and other reviewers. It runs roughly 150–500 words and avoids overall assessments, instead explaining the specifics that led the reviewer to their view. Drafting begins as soon as a paper is accepted. The authors need not endorse the review summary, but can request corrections and must find it acceptable for the paper to be published. Any of the reviewers may sign it, and we hope most will.

Contents

The review summary is intended neither to perfunctorily summarize the paper nor simply to issue a one-dimensional assessment of the paper’s importance. Rather, it aims to be the summary a potential reader would most want to read before reading the full paper. Its main goal is to help a potential reader decide — on the paper’s merits — whether the paper is worth reading.

In addition to summarizing the paper’s contents, a review summary should, as appropriate, convey the paper’s strengths and weaknesses, important caveats, its implications, its relationship to prior art, and its limitations and open questions. It should identify which readers are likely to find reading it worthwhile. This is precisely the information that is usually lost when the output of review is compressed to an accept/reject verdict, and it is also the information that makes results legible across subfield boundaries.

Where code, data, or other materials needed to reproduce or build on the work have not been shared, the review summary should say so, so the reader is alerted to the gap.

Even when an author abstract is well written and hype-free, the review summary provides substantial depth, contrast, and perspective for the reader.

Tone and framing

Review summaries are written in the third person (“the authors prove X”), not the first person (“we prove X”).

The governing principle is show, don’t tell. The point is not to convince the reader to read or not read the paper on the strength of the reviewer’s or the journal’s reputation, but rather to explain the aspects of the paper that led the reviewer to their assessment, so that the reader can make their own decision.

Accordingly, the tone should be professional and respectful, and the summary should avoid an overall verdict, positive or negative. If something is weak, describe the deficiency explicitly rather than asserting that “X is weak.”

Length and form

We suggest 150–500 words, using 1–3 paragraphs as appropriate, depending on reviewer effort and interest. Because a traditional author-written summary is always available alongside it, an optimal review summary may well be longer than a conventional abstract.

Written by a reviewer or action editor

For expediency’s sake, the action editor (who may be a senior editor acting as action editor) asks a single reviewer to write the review summary, rather than trying to coordinate joint authorship. If none of the reviewers are willing to write it, then the action editor should write it themselves as a fallback.

This is less burdensome than it appears. The writer is free to use, mix, and combine any of the text generated by reviewers and editors during the review process. The first paragraph of a reviewer’s report is often already close to an abstract, and many machine learning conferences already suggest reviewers summarize a paper before criticizing it. Reviewers are encouraged to treat the opening paragraph of their initial report as a draft of the review summary they think should appear with the paper if it is published, though this is not required.

Writing the review summary carries additional compensation under the reviewer compensation policy.

The action editor and any of the reviewers — not only the one who wrote it — may choose to publicly sign the review summary once it is finalized. This necessarily reveals their identity. Reviewers are strongly encouraged to sign, but may choose to remain anonymous provided they give the editor a reason; the editor will accept all reasons. Reviewers may sign with a reasonable pseudonym at the editor’s discretion.

An anonymous or pseudonymous signature still lends some credence to the summary, insofar as the journal has a reputation for picking good reviewers. We hope that most reviewers will sign most review summaries, and that these citable public artifacts will themselves incentivize thoughtful reviews.

Approved by authors

The authors do not have to endorse the review summary, since they do not sign it, but they must approve it for their manuscript to be published. If they object to a draft, they may request changes from the reviewer who wrote it. This essentially mirrors the normal abstract and paper, which are written by the authors and approved by the reviewers and editors, possibly after requested changes. If acceptable wording cannot be reached, the authors’ remedy is to withdraw the paper before publication, so that nothing is published.

Licensing

Because the review summary draws text from the reviewers’ and editors’ confidential review discussion, that text must be freely licensed in advance. Reviewers and action editors therefore release the text they contribute to the review discussion under the terms described in the review discussion license. Published papers and review summaries are released under the journal’s publication license (CC-BY).

Where it fits in the review process

The review summary process begins as soon as the paper is accepted for publication. The action editor is responsible for managing this process. The default target is that the summary be finalized about a week after provisional acceptance, and it is finalized along with reviewer identity and signing choices during post-acceptance production.

If a reviewer submits a summary that is not usable, the editor may return to review summary recruitment and invite a different reviewer.

The existence of the review summary also shapes the review itself. Reviewers are asked to distinguish issues that must be fixed to permit publication from issues that are optional but, if not addressed, would be highlighted as deficiencies in the review summary, and from suggestions that are strictly helpful. Because the review summary gives a large opportunity to make the contents and limitations of a paper clear to a potential reader, a submission need not be monumental to be publishable.

See the manuscript review process for the full sequencing.

Motivation

Review is a large investment of expert time, a precious resource, and that investment is substantially wasted when the publicly available output of a confidential review is compressed to a single bit: accept or reject.

Public (open) review avoids that waste, but introduces problems of its own. Without confidentiality, the conversation between authors and reviewers is less honest and more combative, since both sides are performing for an audience. Public review also produces an artifact that is poorly suited to a reader: the conversation meanders, raises disagreements that are only resolved much later, and leaves the reader to reconstruct the conclusion themselves.

The review summary is our experimental solution. It preserves the confidentiality of the discussion while publishing the part of the discussion that is actually useful to a future reader.

The review summary is slightly reminiscent of the “Paper Decision” paragraph on OpenReview (e.g., for NeurIPS), but it is much more extensive, and it is optimized for the potential reader rather than being merely a terse justification of the decision.

Citation

Review summaries are titled Review summary: <article title>.

BibTeX

In biblatex, we recommend that review summaries be given the entry type @review. In BibTeX, where @review is unavailable, @article should be used.

Crossref

Review summaries are registered with Crossref as journal-article, and the record includes an isReviewOf relation pointing to the paper it reviews.