Guide
Rejected in AAAI Phase 1: the reviews came, the reply did not
AAAI-27 sends its Phase 1 rejections on 24 September 2026. Two human reviews and one AI review are released with the notice, and AAAI's own wording is that the paper is rejected without any opportunity for author feedback. ICLR 2027 shut its abstracts six days earlier.
Published 20 September 2026
Rejected, so the paper is yours again and nobody downstream will know. Which is why the next committee finds the same weaknesses unless you find them first. The windows still open are on the deadline board
Read this before you resend the paper
- It is not a desk rejection. Two human reviewers and one AI reviewer read the paper. The screen that removes papers without a reviewer fires weeks earlier and silently.
- There is no rebuttal, and the reviews come anyway. AAAI's words: rejected without any opportunity for author feedback. The 19 to 25 October window is for Phase 2 papers only.
- One of your three reviews never had a vote. The AI review carries no rating and no recommendation, so two human ratings decided this.
- The next archival deadline is five days away. ICLR 2027 abstracts: 18 September, already gone. AISTATS 2027 abstracts: 29 September, papers 6 October.
Which rejection this was
Three ways an AAAI paper ends, and only one has reviewers behind it
The advice for a screening removal and the advice for a Phase 1 rejection are opposites, so find your row before you read anything else. The fourth row is what silence means.
| How it ends | When | Who read the paper | What you are handed |
|---|---|---|---|
| Compliance removal | Silently, weeks after the deadline | Nobody. A checklist: missing PDF, a name left in the manuscript, over length, an author over the cap of 10 | Nothing to read. AAAI-26 is the only edition to publish its funnel: almost 29,000 in, roughly 23,000 under review |
| Phase 1 rejection | 24 September 2026 | Two human reviewers plus one AI reviewer, all three allocated at the start | Every review, released with the decision, and no way to answer any of it |
| Phase 2 rejection | 30 November 2026 | Up to five reviewers, and you answered them in October | The reviews, your 2,500-character response on the record, and the discussion behind it |
| Nothing at all on 24 September | Silence is the outcome | The same two humans and one AI, who were not sufficiently negative | Nothing yet. The paper is in Phase 2 and the next date that concerns you is 19 October |
Read 20 September 2026: the AAAI-27 conference page, review process and main technical track call, whose Phase 1 is "two reviews supplemented by one AI-generated, non-decisional review". The funnel is from the AAAI-26 review process update, and AAAI-27 has published no equivalent account, so that row names the edition. The full funnel is on how AAAI review works.
The reviews arrive with the rejection, and that is documented
Instructions For AAAI-26 Reviewers put it plainly: if the paper is rejected in Phase 1, all reviews are made available to the authors then. The pilot write-up says the same of its author survey, which went out with the Phase 1 decisions. AAAI-27 has published no reviewer instructions page yet, so this is AAAI-26's rule under an unchanged two-phase process.
Nobody outside the committee will ever see this
AAAI's OpenReview venue for AAAI-27 sets public submissions, public withdrawn submissions and public desk-rejected submissions all to false. The rejection leaves no public trace, which is not true everywhere you might send the paper next: the ICLR 2027 call states that even rejected and withdrawn submissions stay public with author names attached.
What you were handed
Three reviews, two votes, and no scale to read them against
What arrived with the notice is everything AAAI will ever say about this paper. Its parts are worth very different amounts, and three of them are worth nothing at all.
| The part | What AAAI says about it | What it is worth to the next version |
|---|---|---|
| The two human reviews | Only human reviews carry ratings and recommendations, and a senior member decided on them. | This is the whole decision. Two people who read the paper were sufficiently negative, and they are who the revision has to convince. |
| The AI review | No scores, ratings or recommendations, by design, and clearly labelled as AI-generated. | Technical feedback with no vote behind it. Free, specific, and with failure modes AAAI itself published. |
| The rating numbers | AAAI publishes no review form and no rating scale for AAAI-25, -26 or -27. | Nothing you can calibrate. There is no published threshold, so a number in your packet cannot tell you how close it was. |
| How many papers went with yours | AAAI publishes no Phase 1 elimination rate for any edition. | Unknowable. The 98.98% you may have read counts Phase 2 papers with three or more human reviews, which is a different question. |
| How many papers AAAI-27 received | Nothing published. AAAI-26 is the last edition with a figure: almost 29,000 submitted, roughly 23,000 reviewed. | Nothing. Larger counts are circulating for AAAI-27 and none of them comes from AAAI, so this page prints none of them. |
Read 20 September 2026: the AAAI-26 reviewer instructions, the AAAI-26 rebuttal FAQ, which is also where the 98.98% and the rule that recommendations are not made by averaging scores come from, and the AAAI-27 call. On the missing scale, and why we will not reconstruct one from author reports, see the scale AAAI does not publish.
The AI review is the one part with published failure modes
AAAI ran it over all 22,977 full-review papers at AAAI-26 and reported on it afterwards: from 5,834 survey responses, authors and reviewers preferred AI reviews to human ones on technical accuracy, review focus and research suggestions. The same report names the weaknesses respondents raised most: premature judgement of novelty, significance and impact, nitpicking on minor issues, excessive verbosity, and factual errors and misreadings.
So sort it before you act on it
A checkable technical objection is worth fixing whoever raised it. A verdict-shaped sentence about how novel or significant the work is carries no rating in Phase 1 and is the category the report itself flags. The two human reviews are the ones that turned into a decision, and they are the ones to answer in the manuscript, since you will not be answering them anywhere else.
The calendar
What was open on 24 September, and what shut on the 18th
Every date below was read off the venue's own page on 20 September 2026. Where a venue has published nothing, the row says so rather than carrying last year's date plus a year.
| Route | Next date | What to know |
|---|---|---|
| ICLR 2027 | Abstracts 18 September | Closed six days before the notice. The 25 September paper deadline admits nothing without an abstract registered by the 18th, so this door needed a decision you could not make. |
| AAAI-27 Student Abstract and Poster Program | 28 September | Four days later, same conference, and explicitly open to work from a rejected AAAI-27 paper as long as it is not already published. Two pages including references, non-archival, student as primary investigator, notification 13 November. |
| AISTATS 2027 | Abstracts 29 September | Papers 6 October, reviews 16 November, decisions 20 January 2027. The first archival deadline after the notice, and the page carries its own hedge that dates may change. |
| AAAI-27 Undergraduate Consortium | 5 October | A research proposal and two statements, not a paper, and it wants an idea you have not already completed. |
| ACL Rolling Review, October cycle | 12 October | Eighteen days after, and the cleanest window for an NLP paper. It is the cycle that commits to NAACL 2027 and COLING 2027, with the commitment deadline on 23 December. |
| AAAI-27 Doctoral Consortium | 13 October | A two-page thesis summary inside a six-page application package. Not a route for this manuscript. |
| AAAI-27 Journal Track | 30 October | Not a route for this paper either. It takes work already published in one of its listed journals, with the final version online by 31 October 2026. |
| CVPR 2027 | Registration 10 November | Papers 16 November, and the call says that date will not be changed. Vision work only, and the dual submission bar runs to 22 February 2027. |
| AAAI-27 workshops | 20 November | AAAI's organiser timetable date for submissions to reach workshop organisers, notifications 2 December. Each workshop sets its own, and their calls were due to AAAI on 2 October. |
| IAAI-27, EAAI-27, AAAI-27 demonstrations | Closed 8 and 18 September | All three shut before the Phase 1 notice went out. IAAI and EAAI closed on 8 September, demonstrations on the 18th. |
| IJCAI 2027 | No call published | The conference dates are out, 7 to 13 August 2027 in Kyoto and 15 to 17 August in Hengqin, with no call for papers and no page for the edition. |
| ICML 2027, NeurIPS 2027 | No call published | Both 2027 conference pages return a 404. |
| UAI 2027 | No call published | The site lists nothing past UAI 2026. |
Tinted rows are reachable after the 24th. All read 20 September 2026, each from the venue's own page and none from an aggregator: AAAI-27 student abstracts, undergraduate consortium, doctoral consortium, journal track, workshops, IAAI-27, EAAI-27 and demonstrations; the ICLR 2027 call; aistats.org key dates, which is not the call for papers page and is the only one of the two carrying dates; ARR dates; the CVPR 2027 call; ijcai.org, icml.cc, neurips.cc and auai.org.
At AAAI itself, what is left is not the main track
Five AAAI-27 routes are still reachable on 24 September and two of them take a paper. The student abstract track closes on 28 September: two pages, non-archival, student as primary investigator, and explicitly open to work from a rejected AAAI-27 submission so long as it is not already published, judged by a separate committee whatever happened to the full paper. Workshops close on 20 November and each sets its own scope and its own rules. Either buys a room in Montreal in February rather than a main-track publication, and neither spends the paper.
What decides this is usually the calendar, and that is documented
NeurIPS 2026 notifies on 24 September as well, so two rejected cohorts reach the same five-day window. The one study that traced a large cohort of rejected papers, the 1,264 NeurIPS 2014 rejects, found they went to the outlets whose deadlines fell shortly after the decision rather than the ones that fitted the work, and its authors named that as the cause. The full account is on our NeurIPS page.
Every deadline we have verified, with the date each one was read
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.
Before you resend
Nothing downstream will ask what you changed
No rule stops you sending this text somewhere else next week. Five questions decide what you should do to it first, and each has a published answer.
| The question | What is actually published |
|---|---|
| Are you free to submit elsewhere | Yes. AAAI-27's bar is on simultaneous submission of overlapping work to another archival venue, and the paper is no longer under review. |
| Will the next venue see the reviews | No. AAAI keeps rejected submissions private, and no venue above asks for prior reviews or a change log. |
| Can you send it back to AAAI later | AAAI publishes nothing either way about resubmitting a paper it rejected, and silence is not permission. There is no AAAI-28 page yet to ask, either: it returned a 404 on 20 September 2026. |
| Can you have an LLM rewrite it | Not for AAAI. Its author policy prohibits text generated by a large language model unless the text is part of the paper's experimental analysis, and allows editing or polishing of author-written text. Check the venue you are aiming at now, because this rule is AAAI's. |
| Is sending it back nearly unchanged a reasonable bet | No, and the one experiment run at this scale put a number on it. That number and the study behind it are on our NeurIPS page. |
Read 20 September 2026: the AAAI-27 call for papers and author policies, plus the OpenReview venue configuration for AAAI-27. The rerun figure and the duplicate-review experiment behind it are on after a NeurIPS rejection.
The next committee starts from the paper, not from this
It cannot see the rejection and it will not ask what changed, so the only thing carried forward is whatever you fix. Specialists read the revised draft against the guidelines of the venue you are aiming at now and hand back the objections, with locations, before a second committee files them.
What the two human reviews were probably telling you: why ML conference papers get rejected, measured on 697 decisions. The screen that fires before any of this: desk rejection explained. If the next room is the other big AI conference, AAAI against IJCAI is the comparison.
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.