Reviewer Guidelines

This page provides guidelines for reviewers for AISTATS. It references previous AISTATS reviewer guidelines and the reviewer guidelines for ICML 2025.

New This Year

This year, the review process includes three new elements.

1. Flagging problematic papers. A button Flag button on the OpenReview page of each paper lets you flag a problematic paper. A flag is only visible to you, the PCs, and the SACs and ACs assigned to the paper. Please use it if you suspect a serious problem that goes beyond a low rating, and select all that apply:
   • Research integrity: plagiarism, dual submission, fabricated or manipulated results.
   • Ethical concerns: harmful applications, discriminatory bias, privacy violations.
   • Incorrect or invalid results: trivial results, fatal errors, or unaddressed fundamental issues.
   • Low-quality or LLM-generated content: unfinished or incoherent writing, missing sections, undefined notation, hallucinated references, undisclosed LLM-generated text.
   Please flag problematic papers as early as possible, so that the chairs can examine them and, where warranted, desk reject them before further reviewing effort is spent. Please justify your flag briefly, pointing to specific sections, results, or passages where possible. These tools are meant to be used thoughtfully and with discernment: a flag is not a substitute for a low score, and should not be used to express disagreement with a paper’s approach, or because a paper is merely weak or not to your taste. Reserve it for cases where you have concrete, well-founded concerns.
2. AI-generated review. Every submission receives an AI-generated review from an LLM-based system operated by the conference. It targets factual correctness only, including incorrect or unsupported claims, errors or gaps in proofs, inconsistencies among text, tables, and figures, mismatches between the described and reported experimental setup, and misattributed related work. It does not assess significance or novelty, does not produce a score, and makes no recommendation.
   Human reviewers do not see the AI review while writing their initial reviews. It is first made available to ACs for preparing their preliminary meta-reviews. It is then released to authors with the human reviews at the start of rebuttal, and becomes visible to reviewers during the discussion phase, together with the authors’ response. All recommendations and decisions remain the responsibility of the ACs, SACs, and PCs, and must be grounded primarily in human assessment and discussion.
3. Confidential suggestions to authors. The review form includes a new optional field, Confidential Suggestions to Authors, for constructive feedback that you would like to give the authors but that should not affect the evaluation. Think of it as the feedback you would give on a colleague’s paper: ideas or changes you would be pleased to see explored, with no obligation for the authors to act on them or to respond. It is visible only to you, the authors, and the PCs; it is not visible to other reviewers or to the AC. Anything relevant to your assessment of the paper must go in the evaluation sections (strengths, weaknesses, questions), not here.

Responsibilities of a Reviewer

1. Report problematic papers to the Area Chair (e.g., unacceptable formatting issues, dual submissions, plagiarism, etc.); see this FAQ for a list of possible issues.
2. Submit high-quality reviews.
3. During the rebuttal period, discuss with the other reviewers and the AC to help arrive at a good decision.
4. Recommend papers for oral talks, awards, etc.

Important Dates

Dates may be subject to change. All dates are Anywhere on Earth (AoE; UTC-12), end of day, unless specified otherwise. 

• Abstract submission deadline: September 29, 2026
• Full paper submission deadline (including all supplementary material): October 6, 2026
• There is no separate supplementary material deadline this year.
• Bidding phase: October 7-13, 2026
• Review period: October 15, 2026–November 5, 2026
• Checking and soliciting emergency reviews: November 6–13, 2026
• AI review released to ACs for the initial meta-review: November 8, 2026
• Initial AC meta-review period: November 8–13, 2026
• Reviews, AI review and initial meta-review released to authors; author rebuttal period: November 16–22, 2026
• Author–reviewer discussion period: November 23–December 3, 2026
• Reviewer–AC discussion period: December 4–approximately December 14, 2026
• Final AC meta-review deadline: December 15, 2026
• AC–SAC discussion period: December 16–approximately December 22, 2026
• SAC initial decisions: approximately January 5, 2027
• Paper final decision notifications: January 20, 2027
• Journal-to-conference track submission due: to be announced
• Conference dates: May 3–6, 2027

Reviewer FAQ

• I accepted the reviewer invitation, but I have not been assigned any paper to review. What happened? 
  • Not all reviewers have been assigned a paper to review. You may still be contacted for a replacement/emergency review during the review period, so please be on the lookout. 
  • To be eligible for reviewing, you should ensure that your publications are imported correctly to your OpenReview profile. When logged onto your profile page, your publications should show on the right-hand side. (Merely adding links to your Google Scholar or DBLP is not enough!)
• When submitting my paper, I nominated myself (or my coauthor) as a reviewer, but I (or the coauthor) was not assigned any paper to review. Does this violate the reciprocal reviewer requirement? 
  • At this point, any submission that did not nominate an author has been contacted.

How to Review?

The purpose of the review process is twofold: to identify papers which offer significant contributions to the fields of artificial intelligence and statistics, and to provide constructive feedback to authors to help improve their work.

When reviewing a paper, consider its potential long-term impact, particularly out-of-the-box ideas, novel problems, and contributions that bridge fields. Such works may be easy to criticize but are often essential for progress, so avoid dismissive criticism.

Best Practices

• Be thoughtful. Remember some authors may be submitting for the first time.
• Be fair. Avoid letting personal feelings affect your review.
• Be useful. A good review is useful to authors, reviewers, and ACs.
• Be specific. Avoid vague comments.
• Be flexible. Update your review when new information is presented.
• Be timely. Meet deadlines and respond promptly during discussions.

If someone pressures you to bias your review, notify the Program Chairs at the official contact form. Report unethical or suspicious behavior to your Area Chair immediately.

Review Form

1. Code of Conduct (Visible only to PCs/SACs/ACs)
   •  Agree: I confirm that this review was written by me (and not generated by an LLM) and that I have no conflicts of interest with the authors.
2. Confidentiality Agreement (Visible only to PCs/SACs/ACs)
   •  Agree: I agree to keep the manuscript, supplementary material, code, and reviews confidential. I will not upload or share the paper with external parties or unauthorized LLM services.
3. Flag the Paper (Visible only to PCs/SACs/ACs): Flag this paper for the chairs? Check all that apply if you suspect a serious problem that goes beyond a low rating.
   •  Research integrity: plagiarism, dual submission, fabricated or manipulated results
   •  Ethical concerns: harmful applications, discriminatory bias, privacy violations
   •  Incorrect or invalid results: trivial results, fatal errors, or unaddressed fundamental issues
   •  Low-quality or LLM-generated content: unfinished or incoherent writing, missing sections, undefined notation, hallucinated references, undisclosed LLM-generated text
   •  No major concerns
   • Justification (visible only to PCs/SACs/ACs): If you flagged the paper, please provide a brief justification. Indicate N/A if not necessary.
4. Summary: Briefly summarize the paper’s key goals, methodology, and theoretical/empirical contributions. (1-2 paragraphs)
5. Strengths: What are the main strengths of the paper? (e.g., theoretical rigor, novelty, clarity of presentation, empirical thoroughness, relevance to the AISTATS community). Please number the strengths to simplify discussions (e.g., S1, S2, ...).
6. Weaknesses: What are the primary weaknesses or limitations? Please be specific and technical (e.g., overly restrictive assumptions, gaps in proofs, missing baselines/ablations, overclaimed results). Please number the weaknesses to simplify discussions (e.g., W1, W2, ...). Indicate N/A if not necessary.
7. Questions: Specific questions and actionable requests for the authors’ rebuttal. What clarifications or evidence could realistically lead you to improve your score? Please number the questions to simplify discussions (e.g., Q1, Q2, ...).
8. Clarity – Presentation: Is the paper well written and well organized? Consider whether the contributions, notation, assumptions, and results are clearly stated, and whether the methodology is well motivated.
   •  4: Excellent (Very clear and well organized; contributions, notation, and assumptions are precisely stated and well motivated)
   •  3: Good (Generally clear; minor issues with organization, notation, or exposition that do not hinder understanding)
   •  2: Fair (Understanding is hindered by unclear writing, inconsistent or undefined notation, or poor organization; substantial revision needed)
   •  1: Poor (Hard to follow or unfinished; key parts cannot be understood or evaluated as written)
9. Soundness – Correctness: Are the claims and contributions supported by sound theoretical proofs and/or empirical methodology?
   •  4: Sound and rigorous (Proofs are complete and correct; empirical protocol is solid)
   •  3: Minor gaps (Generally sound; minor gaps or unstated benign assumptions)
   •  2: Questionable (Significant unaddressed gaps, overly strong unverified assumptions, or weak empirical support)
   •  1: Major errors (Fatal mathematical flaws or invalid empirical methodology)
10. Contribution – Significance: How significant are the theoretical and/or empirical contributions, and how novel are they relative to prior work?
    •  4: High (Novel and significant; likely to influence future work in AI & Stats beyond a single sub-area)
    •  3: Good (Solid contribution with clear value to at least one sub-area; some novelty, with limited scope or impact)
    •  2: Limited (Incremental over existing work, or empirical gains are small / lack key baselines or statistical support)
    •  1: Minimal (Known or trivial results, or no clear contribution beyond prior work)
11. Rating: Overall evaluation of the paper.
    •  5: Strong Accept (Groundbreaking or top-tier contribution; technically excellent, broad interest)
    •  4: Accept (Technically solid, clear contribution to AI & Stats; convincing evaluation)
    •  3: Borderline Accept (Technically solid; merits outweigh minor limitations)
    •  2: Borderline Reject (Interesting, but limitations/unclear claims outweigh contributions)
    •  1: Reject (Major technical flaws, weak evaluation, or unconvincing claims)
12. Confidence: Confidence in your assessment.
    •  5: Absolutely certain (Familiar with literature; carefully verified proofs/experiments)
    •  4: Confident (Understand the paper well; minor chance of missing subtle details)
    •  3: Fairly confident (Understood the core contribution; math/details not checked thoroughly)
    •  2: Somewhat confident (Willing to defend rating, but paper is outside my core expertise)
    •  1: Educated guess (Not my area or difficult to evaluate)
13. Confidential Suggestions to Authors (optional; visible only to you, the authors, and, for safety reaons, the PCs). This field provides a space for constructive suggestions to help the authors improve their paper, beyond the points raised in the sections above. Its purpose is to support improvement, not to inform the paper’s evaluation or selection. Any points relevant to your assessment of the paper should therefore be included in the appropriate evaluation sections above, rather than here. This field will thus not be visible to other reviewers or the paper’s AC. Only you and the authors, (and for safety reasons PCs) will be able to access it. Authors don't have to respond to or implement these suggestions. This feedback is meant to be separate from the evaluation, so that authors do not need to defend the paper against the points raised here or to justify choosing not to act on them. You may think of this as feedback you would give on a colleague’s paper: ideas or changes you would be pleased to see explored or implemented, while fully accepting that the authors may choose not to pursue them. You can therefore use this field to raise points on which you would not want the authors to be evaluated, but which you nevertheless believe could provide useful feedback for improving the paper.
14. Confidential Comments (optional; visible only to PCs/SACs/ACs): Confidential comments for the Area Chairs, Senior Area Chairs, and Program Chairs.

Author Rebuttal and Discussion Period

Authors will be given the opportunity to respond to your reviews before decisions are made. This is to enable them to address misunderstandings, point out parts of their submissions that were overlooked, or disagree with the reviewers’ assessments.

In previous years, some authors felt that their responses were ignored. As a reviewer, you are expected to read and (if appropriate) respond to each author’s response. Keep an open mind during the discussion period. (Have you overlooked something?) It is not fair to ignore any author response, even for submissions that you think should be rejected. Even when an author’s response did not change your assessment of a submission, you must convey to the authors that you have carefully considered their comments. 

Ethical Conduct for Peer Review (Use of Generative AI)

AISTATS reviewers are expected to follow standard ethical conduct for the peer review process. In particular, AISTATS prohibits:

• use of privileged information (e.g., information and discussions about submissions) for any purpose other than reviewing;
• use of Generative AI tools in reviewing (as described below);
• all forms of collusion, whether explicit or tacit (e.g., an arrangement between authors and reviewers, ACs, or SACs to obtain favorable reviews).

It is strictly prohibited to use Generative AI tools, such as large language models (LLMs), for reviewing AISTATS submissions. In particular, reviewers cannot use Generative AI tools to write their reviews, and reviewers cannot input any content from any submission or review into a Generative AI tool.

If you believe someone may be engaging in unethical conduct, please notify AISTATS via the confidential reporting form https://forms.gle/VugVjWBBFUi8T4kn9 .

All suspected unethical conduct will be investigated by the AISTATS organizing committee. Individuals found violating the rules may face sanctions, have their own submissions rejected, etc.

AI Review of Submissions

New in 2027: every submission receives an AI-generated review from an LLM-based system operated by the conference. It targets factual correctness only, including incorrect or unsupported claims, errors or gaps in proofs, inconsistencies among text, tables, and figures, mismatches between the described and reported experimental setup, and misattributed related work. It does not assess significance or novelty, does not produce a score, and makes no recommendation.

Human reviewers do not see the AI review while writing their initial reviews. It is first made available to ACs for preparing their preliminary meta-reviews. It is then released to authors with the human reviews at the start of rebuttal and becomes visible to reviewers during the discussion phase, together with the authors’ response. All recommendations and decisions remain the responsibility of the ACs, SACs, and PCs and must be grounded primarily in human assessment and discussion.

Other Remarks

1. Minor formatting issues: Do not desk-reject; flag them instead.
2. Supplement: Supplementary material may be appended; conflicts between appended and separately uploaded material can be resolved at the reviewer’s discretion.
3. Links in submission: Exercise caution when accessing external links.
4. Code in GitHub: Reviewing submitted code is optional; if accessing, ensure anonymity and be aware of possible security risks.