brycewang-stanford AAAI Review Process
Use this to plan around AAAI review rather than treating it as a generic OpenReview rebuttal. Reopen the current review-process page and author FAQ before advising on timing or strategy.
Process model
- AAAI main technical track uses double-blind reviewing.
- AAAI-27 uses a two-phase review process. Phase 1 allocates three reviewers — two human reviews supplemented by one non-decisional AI-generated review. Papers with sufficiently negative reviews are rejected before author feedback (AAAI-27: notified 2026-09-24).
- Papers continuing to Phase 2 receive additional reviews, up to five in total, and one author feedback phase (AAAI-27: 2026-10-19 to 10-25, final decisions 2026-11-30). Phase 2 reviewers are not shown the Phase 1 reviews until they have submitted their own — so a Phase 2 review is an independent read, not a reaction to the earlier ones.
- Final decisions were made through reviewer discussion and senior program committee oversight, not by the AI review.
- Author feedback is short and constrained; it is mainly for correcting misunderstandings, not replacing the paper.
Author strategy
- Reduce Phase 1 reject risk before submission by making contribution, evidence, and checklist compliance obvious.
- When reviews arrive, distinguish human-review claims, AI-review errors, and AC/SPC decision questions.
- Use rebuttal to resolve the highest-impact factual issue under the character limit.
- Do not attack the AI review. Correct it when it contains consequential false statements.
- Avoid new experiments in rebuttal; use submitted evidence and camera-ready promises sparingly.
Stage-by-stage decision map
AAAI's pipeline differs from a single-round OpenReview venue, so plan actions per stage rather than treating every signal as a rebuttal opportunity.
| Stage | What is happening | Author leverage |
|---|---|---|
| Pre-submission | Phase-1 bar is set by clarity and checklist | maximal: fix the paper itself |
| Phase 1 | human reviews plus advisory AI review | none yet; summary reject possible |
| Phase 2 | additional reviews, one feedback round | one short response, no new results |
| Discussion | reviewers and SPC/AC weigh feedback | indirect: a clean correction can swing it |
| Decision | SPC/AC oversight, not the AI review | archive everything for appeal or journal |
Why papers die in Phase 1
Because the reviewer pool is large and submission volume is high, clearly-below-bar papers are cut early to protect later effort. Common triggers: an unreadable first page, a contribution a non-specialist cannot place, a checklist that contradicts the paper, or evidence too thin to trust. None of these can be repaired after the Phase-1 cut, so they must be eliminated before submission.
Worked vignette
An NLP paper with strong results buries its contribution under three pages of setup. A Phase-1 reviewer from a planning background cannot find the AI claim and scores it a reject; the paper never reaches feedback. The fix belongs entirely pre-submission: a first-page contribution statement and a checklist that matches the experiments, so the broad-AI reviewer can place and trust it fast.
Output format
[Stage] pre-submission / Phase 1 / Phase 2 / rebuttal / discussion / decision
[Decision risk] summary reject / borderline / likely accept / ethics-policy risk
[Best action] revise before submission / rebut / clarify evidence / escalate
[AI-review handling] ignore / correct / cite submitted evidence
[Rationale] <why this fits AAAI process>