Guide
AAAI's review process: the two rejections you cannot answer
A compliance screen removed roughly 6,000 of AAAI-26's almost 29,000 submissions before review. Phase 1 then rejects with the reviews attached and, in AAAI's words, without any opportunity for author feedback.
Published 14 August 2026
Already in the author response window? Then the process is behind you and the 2,500 characters are the whole job. How to write a rebuttal
The short version
- Most of your outcome is settled where you cannot speak. Two gates close before any reply is possible.
- The compliance screen is mechanical. A missing PDF, a name left in the manuscript, an extra page, one author over the cap.
- Phase 1 rejects on two human reviews plus a non-scoring AI review. You are handed all of them at the moment of rejection, with nowhere to reply.
- Survive both and you get 2,500 characters. No URLs and no new experiments, for up to five reviews.
Gate one
Roughly one submission in five never reached a reviewer
AAAI-26 published its own funnel: almost 29,000 in, roughly 23,000 under review. That gap is a checklist.
| What ends it early | The rule | Stated in |
|---|---|---|
| Missing PDF | A record with no manuscript. | AAAI-26 removals |
| Names left in the paper | Non-anonymised manuscripts are removed, not returned. | AAAI-26 removals |
| Over length | 7 pages of main content, 9 in total, and past page 7 references only. | AAAI-26 removals, AAAI-27 call |
| Too many submissions | 10 per individual author, across every technical track paper you appear on. | AAAI-26 removals, AAAI-27 instructions |
| The abstract changed | Chairs may reject without review if it changes substantially between the two deadlines. | AAAI-27 instructions |
| Placeholder title or abstract | Deleted. | AAAI-27 instructions |
| No reproducibility checklist | From the author kit, submitted with the paper, not afterwards. | AAAI-27 instructions |
| Under review elsewhere | Nothing under review, accepted or published at another archival venue. | AAAI-27 call |
| Near-duplicate siblings | LLM-rewritten variants filed as multiple lottery tickets are an ethics violation. | AAAI-27 ethics policy |
| Posting a preprint | Not a trigger. arXiv and non-archival workshops are permitted. | AAAI-27 call |
The funnel is from AAAI's AAAI-26 review process update, whose examples of papers not fully compliant are missing PDFs, non-anonymized manuscripts, over-length papers and authors over the cap. The 6,000 and the fifth are ours. Other rules from the AAAI-27 submission instructions and call for papers, read August 2026.
None of that screen is a judgement about your work
Every line is checkable in an afternoon and none of it needs a reviewer, because none is involved. AAAI-26 lost about a fifth of its submissions there while recruiting 28,000 people to read the rest.
Gate two
Eight weeks after the deadline, some authors get the reviews and nothing else
The AAAI-27 calendar, with what is still yours at each stop.
| Date, AoE | Stage | What you can still do | What has closed |
|---|---|---|---|
| 21 July 2026 | Abstracts due | Register the real title, and count co-authorships against the cap | Placeholders, and later abstract changes |
| 28 July 2026 | Full papers due | The last edit anyone will read | The paper. Every review is written on this version |
| 31 July 2026 | Supplementary and code | Add code, data, appendices | Supplementary, for good |
| 24 September 2026 | Phase 1 rejections | Nothing. The reviews arrive with the rejection | The run, with no author feedback |
| 19 to 25 October 2026 | Author feedback, Phase 2 only | One response of 2,500 characters | New results, URLs, supplementary |
| After 25 October | Committee discussion | Nothing. They read what you sent | Everything |
| 30 November 2026 | Final notification | Nothing | The decision |
| 14 December 2026 | Camera-ready | Fix what the reviews found | Anything that changes the decision |
| 16 to 23 February 2027 | AAAI-27, Montreal | Register and present | AAAI-26 required an author in person |
Dates from the AAAI-27 conference page, all AoE, read August 2026. Tracks differ: the Special Track on AI Alignment closed on 21 August 2026, three weeks after the main track, and gives feedback 27 October to 2 November. Read your own track's page: two official AAAI pages carry different AAAI-26 dates. Track dates re-read 10 September 2026.
A Phase 1 rejection is all of the criticism and none of the conversation
The notice lands on 24 September 2026, about eight weeks after the deadline by our count, and the reviews come with it. A paper receiving sufficiently negative reviews is rejected without any opportunity for author feedback, so you read reviews you cannot answer and the only thing left is to fix the paper for somewhere else. Silence on 25 September is the good outcome: it means Phase 2.
Three editions
The rules moved twice in two years, and the wording narrowed
Advice about AAAI ages badly. Check the edition on anything you read.
| AAAI-25 | AAAI-26 | AAAI-27 | |
|---|---|---|---|
| Phase 1 reviewers | Two, both human | Two human reviews plus one AI review | Three allocated: two human plus one non-decisional AI review |
| Phase 1 rejection | Two sufficiently negative reviews, no author feedback | Same rule, decisions made solely by humans | Sufficiently negative reviews, no author feedback |
| Phase 2 reviews | Not stated on the pages read | 5 human and 1 AI aimed for, 3 human enough to decide | Up to five, new reviewers kept from Phase 1 reviews until they file |
| What author feedback is for | Comment on all reviews | Comment on all reviews | Address questions or factual mistakes in reviews |
| Who assigns a rating | Humans, the only kind of reviewer | Only human reviews carry ratings and recommendations | Humans alone decide, and no reviewer is replaced by AI |
| Generative AI in your paper | Not stated on the pages read | Prohibited unless the text is part of the experimental analysis | Permitted judiciously, the responsibility yours |
From the review-process pages for AAAI-25, AAAI-26 and AAAI-27, each edition's call, and the AAAI-26 reviewer instructions. Not stated on the pages read means exactly that. No author-policies page exists for AAAI-27, so its call is all there is on the LLM question.
Your third Phase 1 reviewer is a machine that cannot score you
AAAI-26 ran all 22,977 of its full-review papers through it in under a day, and the review carries no rating and no recommendation by design. You may rebut it as you would a human review, and AAAI says it will not be updated in response, though the humans read your answer. What you cannot do is try it first: authors cannot submit to the AI system beforehand.
Mechanics from AAAI's FAQ for the AI-assisted peer review pilot; volume from the pilot write-up, which also reports verbosity and factual errors.
The only place you speak
Around 400 words, once, for every review you received
AAAI's rebuttal FAQ is the clearest thing it publishes for authors. These are its rules.
| The constraint | What AAAI-26 says |
|---|---|
| Length | A single rebuttal of 2,500 characters maximum, answering all reviews. |
| How many reviews it answers | AAAI aims for 5 human and 1 AI, and can decide from 3. 98.98% of papers had 3 or more when responses opened. |
| URLs | None. Responses detected to contain URLs may be withheld. |
| New results | Do not provide them, even if the reviewers asked: rushed experiments cannot be validated, and reviews address the paper as submitted. |
| Arguing an experiment is unnecessary | Acceptable, and the version of that reply AAAI reads. |
| Supplementary material | Cannot be updated at this stage under any circumstances. |
| The AI review | Rebut it like any other. It will not be updated, and the humans can read your response. |
| A wrong or empty review | Factually wrong, AI-generated or lacking reviews belong in the rebuttal, not in a report to the Ethics Committee. |
| Tone and fairness | Not your subject. Authors are told not to focus on either. |
| How it is weighed | Not by averaging scores. Senior members decide after considering all available information. |
From the AAAI-26 FAQ for authors on the rebuttal, read August 2026. AAAI-27 has published no equivalent, and narrows the phase to questions or factual mistakes.
There is no number to move, so stop writing as if there were
Recommendations are not made by averaging scores: a senior member decides after reading everything. That kills the standard play of picking the harshest reviewer and negotiating a point out of them. What you would argue in October is worth more as a paragraph written in July.
What AAAI does not publish
There is a rating, and AAAI does not tell you what it means
We could not find the scale, so we are not printing one
AAAI publishes no review form and no rating scale for AAAI-25, -26 or -27. The fields exist: its OpenReview configuration carries a review rating, a confidence and a meta-review recommendation, but the invitation holding their values answers a guest with a permission error. What circulates is crowdsourced reports of what authors say they got, which is neither a scale nor a threshold. This site once published a NeurIPS scale built from that kind of consensus and it was wrong. If a venue has not published its scale, do not plan around it.
What AAAI documents is more actionable: ratings are attached by humans only, never averaged, and the questions behind them are published in full.
| AAAI's own grouping | What the reviewer is told to ask |
|---|---|
| Story of the paper | The problem, the limitations in the state of the art it addresses, whether results really support the claims, the key contribution. |
| Presentation | Sound and clearly described, no unstated assumptions, enough detail to reproduce, limitations stated. |
| Related work | Placed in context. A reviewer demanding more must give three citations from non-overlapping authors to count as substantive. |
| Evaluations | Baselines, established metrics, suitable benchmarks, an analysis of the errors made, replicable evaluations. |
| The overall assessment | Clarity, technical and experimental soundness, intellectual novelty, relevance to the field. |
Condensed from Instructions For AAAI 2026 Reviewers, linked from the AAAI-26 reviewer page, the closest thing to a form AAAI makes public. It publishes no Phase 1 elimination rate and no acceptance rate reconcilable with its own counts, so neither is here.
Free tool, no account
Where should this paper go?
Paste the title and the abstract. It ranks the venues that publish this kind of work, with the reach and the article charge OpenAlex has on record.
At AAAI, the argument is the paper
One screen you pass with a checklist, one rejection you cannot answer, then 2,500 characters. Before any of it comes the question of whether AAAI is the room for this paper at all.
Matching needs no account, and a paid review does not predict a decision: what it cannot do is written down. Gate one is desk rejection explained, and the 2,500 characters are how to write a rebuttal. What ends the papers that do get read is in why papers get rejected.
phdflow reads a finished draft against the guidelines of the venue you are aiming at and returns an accept or reject call with the objections behind it. Scored against 697 real decisions: AUC 0.91 and 96% accuracy on 297 ICLR 2025 papers, one false accept in 149 rejects. The full measurement. Read a finished review before you pay for one. Your first paper gets a free preview; the full report is 60 credits, €9. No account needed. Pricing.