The AISTATS 2026 review form, explained

How AISTATS reviewers read your paper

AISTATS 2026 reviewers summarize the paper, then choose Soundness, Significance and Novelty options each with a written justification (incremental or known-results verdicts must cite the prior work), note non-conventional contributions, assess clarity, relation to prior work and reproducibility, and give a 1–7 rating and 1–5 confidence. Reviewers should be thoughtful, fair, specific and timely, avoid dismissive criticism of novel or cross-disciplinary work, and respond to every rebuttal.

• Thoughtful, fair, useful and specific; avoid dismissive criticism, especially of novel or cross-disciplinary work.
• Cite the specific prior publications whenever you call a result incremental or known.
• Justify every Soundness / Significance / Novelty choice.

Where the bar sits. The AISTATS 2026 call says acceptance tends to be around 25% (2025: 31%, 2024: 28%). No score statistics are public; on the 1–7 scale a mean of 5 (Borderline Accept) or more is a rough proxy for acceptance, 4–5 borderline.

The reviewer form, field by field

In the order reviewers see it, with every score and its official wording.

1. 01

   Summary And Contributions

   Free text

   Briefly summarize the paper and its contributions.

2. 02

   Soundness

   Single choice

   Choose one; justify it in the strengths/weaknesses.

   • Correct / minor errors (e.g., typo or errors that do not affect the main results)
   • Major errors (e.g., an incorrect theorem or derivation)
   • Not applicable
3. 03

   Significance

   Single choice

   Choose one; justify it in the strengths/weaknesses.

   • Significant contributions
   • Somewhat significant
   • Not significant
   • Not applicable
4. 04

   Novelty

   Single choice

   Choose one. "Incremental" or "Known results" requires citing the prior publications.

   • Ground-breaking
   • New results
   • Incremental compared to existing results
   • Known results, or a trivial extension of known results
5. 05

   Justification Of Soundness, Significance And Novelty

   List of points

   The form has a justification box under each of Soundness, Significance and Novelty. Give the specific evidence behind each choice: errors found (with location), why the contribution matters or does not, and the prior work it builds on (cited).

6. 06

   Non-conventional Contributions

   Free text

   Contributions outside the usual mould, e.g. bridging disciplines or posing new problems. "None" if not applicable.

7. 07

   Clarity

   Free text

   Is the paper clearly written? Consider whether it (1) clearly states its contributions, notation and results, (2) supports the claims made in the title, abstract and introduction with results, (3) explains the meaning of its theoretical assumptions, and/or (4) motivates the proposed methodology well, e.g. with examples.

8. 08

   Relation To Prior Work

   Single choice

   Choose one. Name any missing work under Additional Comments.

   • All related works are clearly discussed
   • All related works are cited but the discussion is less clear
   • Some minor related works are missing or described incorrectly
   • Some important related works are missing or described incorrectly
9. 09

   Additional Comments

   List of points

   Feedback and questions for the authors, including any missing related work.

10. 10

    Reproducibility

    Single choice

    Choose one.

    • Sufficient amount of details available for reproducing the main results
    • Some amount of details available
    • Insufficient amount of details available
    • Not applicable (the paper does not have experiments)
11. 11

    Rating

    Score · 1–7

    Overall rating. Note the official order: 4 ("Weak Accept / Weak Reject") sits between Borderline Reject and Borderline Accept; use it sparingly.

    Score	Meaning
    7	Strong Accept (Technically excellent paper with clear novelty and broad potential impact beyond a single sub-area of AI and stats. The work is very well executed with convincing evaluation and good attention to reproducibility and ethical considerations.)
    6	Accept (Technically solid paper, with high impact on at least one sub-area. The contribution is convincing and the evaluation is adequate, though there may be some limitations or open questions.)
    5	Borderline Accept (Technically solid paper where reasons to accept outweigh reasons to reject, e.g., limited evaluation.)
    4	Weak Accept / Weak Reject (Paper has some merits but also notable weaknesses. Reasons to accept and reject are roughly balanced, and the paper may require further evaluation or discussion. Please use sparingly.)
    3	Borderline Reject (Technically solid paper where reasons to reject, e.g., limited evaluation, outweigh reasons to accept, e.g., good evaluation.)
    2	Reject (For instance, a paper with technical flaws, weak evaluation, inadequate reproducibility and incompletely addressed ethical considerations.)
    1	Strong Reject (For instance, a paper with well-known results or unaddressed ethical considerations.)

12. 12

    Confidence

    Score · 1–5

    How certain you are about your assessment.

    Score	Meaning
    5	Absolutely certain (Use this sparingly; please ensure you are very familiar with the related work, you have carefully checked and understood the proofs and/or experimental details if applicable)
    4	Confident, but not absolutely certain. (It is unlikely, but not impossible, that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work.)
    3	Fairly confident (It is possible that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work. Math/other details were not carefully checked.)
    2	Somewhat confident (You are willing to defend your assessment, but it is quite likely that you did not understand central parts of the submission or that you are unfamiliar with some pieces of related work. Math/other details were not carefully checked.)
    1	Educated guess (The submission is not in your area or the submission was difficult to understand. Math/other details were not carefully checked.)

13. 13

    Ethical Concerns

    Single choice

    Choose one (visible to PC/SAC/AC only).

    • No concern
    • Minor concerns
    • Major concerns

Area chair meta-review

AISTATS ACs recommend accept/reject and nominate orals. The meta-review form itself is not public.

• Accept (oral)
• Accept (poster)
• Reject

AISTATS reproducibility checklist (in the paper, after the references)

Reviewers check these. Gaps are easy points to lose.

1. For all models and algorithms: mathematical setting, assumptions and algorithm described; properties and complexity (time, space, sample size) analysed; anonymized source code (optional)
2. For theoretical results: all assumptions stated; complete proofs; clear explanations of any assumptions
3. For experiments: code, data and instructions to reproduce; training details (splits, hyperparameters, how chosen); error bars or variance over runs; compute resources
4. For existing assets used: creators cited; license stated; new assets documented and released
5. For crowdsourcing or human-subjects research: full instructions; potential risks and IRB approval; compensation

Policy on LLMs in reviewing