# Get an AI review of your ICML paper before you submit > Upload your draft PDF to reviewpaper.ai and choose ICML. In a few minutes, three simulated reviewers fill in the ICML main-track review form: summary, strengths, weaknesses, Soundness, Presentation, Significance and Originality (1–4), key questions for the authors, limitations, and the 1–6 Overall Recommendation with a confidence level. An area chair writes the meta-review, and you get a predicted score range, a ranked fix list and a check against ICML's submission requirements. Every point links to the passage in your PDF. The first three reviews are free. - Venue: International Conference on Machine Learning - Review form: official, field by field at https://reviewpaper.ai/venues/icml.md - Next ICML deadline: ICML 2027 abstract, ~Sat, Jan 23, 2027 · 23:59 AoE (estimated from last year): https://reviewpaper.ai/deadlines/icml-2027.md - Web page: https://reviewpaper.ai/ai-review/icml ## What the ICML review contains Each of three simulated reviewers fills in the ICML 2026 form, in this order: | Field | Type | |---|---| | Summary | free text | | Strengths | list of points | | Weaknesses | list of points | | Soundness | score, 1–4 | | Presentation | score, 1–4 | | Significance | score, 1–4 | | Originality | score, 1–4 | | Key Questions For Authors | list of points | | Limitations | free text | | Overall Recommendation | score, 1–6 | | Confidence | score, 1–5 | | Ethical Concerns | free text | Overall Recommendation scale: 6 = Strong Accept, 5 = Accept, 4 = Weak accept, 3 = Weak reject, 2 = Reject, 1 = Strong Reject. On top of the three reviews: - Area-chair meta-review with a recommendation (Accept (oral), Accept (spotlight poster), Accept (poster), Reject) - Score outlook: an expected score range and a likely accept, borderline or likely reject call - Fix list: what to change before submitting, ranked by impact - ICML submission requirements: 5 items checked against the paper - Every point linked to the passage in the PDF ## When it helps at ICML - **Before the deadline.** ICML reviewers list the key questions whose answers could change their recommendation. The simulated reviewers do the same, so you can answer them in the paper instead of in a character-limited reply. - **Which axis is weak.** Soundness, Presentation, Significance and Originality are scored separately (1–4), so you see whether the problem is the method, the writing or the pitch. - **After a rejection.** Rework the paper and run it against the next venue's form: [NeurIPS](https://reviewpaper.ai/ai-review/neurips), [ICLR](https://reviewpaper.ai/ai-review/iclr) or [AAAI](https://reviewpaper.ai/ai-review/aaai). - **When the real reviews land.** Import them in the Rebuttal tab and get a draft reply to every concern, quoting your paper. ## Sample report https://reviewpaper.ai/sample: a full report on a published paper (ICLR form), no sign-up. A ICML report has the same parts in the ICML form. ## FAQ ### Is ICML's scale the same as NeurIPS's? Both use a 1–6 overall score with similar labels (ICML calls 3 and 4 Weak reject and Weak accept, NeurIPS calls them Borderline). Our [score scales page](https://reviewpaper.ai/venues/score-scales) compares every venue's scale side by side. ### What scores does a paper need at ICML? ICML 2026 accepted about 26.6% of submissions. No public score statistics; inferred from NeurIPS's identical scale: mean ≥ 4 likely accept, 3.5–4 borderline. The outlook in every report uses this bar. ### How accurate is the predicted ICML score? Treat it as a stress test, not a forecast. Real reviewers disagree a lot, and simulated ones can miss what a domain expert would catch or raise issues that aren't real. The value is in the specific criticisms, each linked to a passage, that you can fix before the deadline or answer in the rebuttal. ### Is the ICML review form you use official? It follows ICML's published 2026 reviewer form and guidelines, field by field, with the official score labels. The [ICML form page](https://reviewpaper.ai/venues/icml) lists every field and links the official source. ### Am I allowed to run my own paper through an AI reviewer before submitting to ICML? Venue LLM rules are written for reviewers handling someone else's submission. ICML's rule for reviewers: ICML 2026 assigns each reviewer Policy A (no LLMs) or Policy B (LLMs may help understand the paper and polish wording, but may not identify strengths/weaknesses or write the review). Violations can get the reviewer's own papers desk-rejected. Checking your own unpublished paper is your decision as the author, and nothing is shared with the venue. If you are an assigned reviewer, don't upload a paper you are reviewing. Our [LLM policies guide](https://reviewpaper.ai/guides/llm-policies-by-venue) lists every venue's rules with official links. ### Who can see my paper? Does it break double-blind review? Only you. The PDF is stored privately, served only to your signed-in account and never used to train models, so uploading a paper under double-blind review reveals nothing to reviewers, the venue or other users. Delete the paper and its file, reviews and chat are gone. Details in the [privacy policy](https://reviewpaper.ai/privacy). ### How is this different from asking ChatGPT to review my paper? A general chatbot writes one review in whatever format you prompt for. reviewpaper.ai reads the whole PDF, appendices included, has three reviewer personas fill in ICML's own form and score scales, adds an area-chair meta-review, a score outlook and a ranked fix list, and links every point to the passage it is about so you can check it in seconds. ### My paper is already submitted. Is it still useful? Yes. Reviews take weeks, so seeing the likely objections now gives you time to run the missing experiment before the author response. When the real reviews arrive, import them for a draft rebuttal. If the paper is rejected, run the revision against the next venue's form. ### How much does it cost? Your first three reviews are free, with no card. After that, Pro is $10 a month for 12 credits, or buy 5 credits for $5. A Standard review (Claude Sonnet 5.5) costs 1 credit; Deep (Claude Opus 5.5) costs 2 and Max (Claude Fable 5.1) 6. See [pricing](https://reviewpaper.ai/pricing). --- Simulate reviews of your own paper in this venue's form: https://reviewpaper.ai/new?venue=icml (3 free reviews). Sample report: https://reviewpaper.ai/sample. reviewpaper.ai is for authors checking their own paper at any stage: a draft, a submission under review, or a rejected paper being reworked. If you are an assigned reviewer, follow your venue's LLM policy: most venues forbid putting a submission you are reviewing into any external AI tool. Not affiliated with any conference.