ICLR 2027 rebuttal: how to reply to reviewers

The ICLR 2027 rebuttal is the author-reviewer discussion period on OpenReview, where you answer your reviewers and can revise your paper. ICLR 2027 released reviews on 5 November 2026, and the discussion closes on 18 November 2026.

If you got your first ICLR reviews this month, you will find what the period is, what the rules say, and how to write replies that reviewers actually read.

What are the ICLR 2027 rebuttal dates?

The ICLR 2027 rebuttal dates, from the official call for papers, are:

• Reviews released: 5 November 2026
• Author-reviewer talk: 5 to 18 November 2026
• Reviewer-AC talk: 19 November to 16 December 2026
• Final decisions: 16 December 2026

So as of 12 November 2026, you have about six days left. After 18 November, authors stop replying, and reviewers and the area chair (AC) discuss your paper among themselves. The reviewer guidelines also say ACs may hold a virtual meeting on borderline papers between 26 November and 2 December.

How does the ICLR rebuttal work on OpenReview?

The ICLR rebuttal on OpenReview is an open, written discussion, not a single letter. Each reviewer posts a review under your paper. You reply in comments under each review, and reviewers can reply back. The talk is asynchronous, so there is no live call.

Three rules from the ICLR 2027 guidelines matter here:

• You can revise: authors may upload a new version of the paper to address concerns.
• One more page: the main text limit goes from 9 pages to 10 pages during the rebuttal, for new results. The limit is strictly enforced.
• Scores can change: reviewers are told to stay open to changing their score, up or down, and to update their review after the discussion.

ICLR is also public. Every submission, even a rejected or withdrawn one, stays on OpenReview with the authors' names. Your replies will be read by people long after December, so write them as if they are part of the paper.

How to write an ICLR rebuttal: step by step

1. 1Wait one day before replying. Read every review twice. Angry replies read badly in public.
2. 2Make a table of every concern: reviewer, the point, how serious it is, and your answer.
3. 3Find the shared concerns. If two or three reviewers raise the same point, answer it once in a general comment to all reviewers.
4. 4Sort the rest. Some points need a new experiment, some need a clearer explanation, and some come from a misreading.
5. 5Run the most important new experiments first. You have days, not weeks, so pick the ones that could change a score.
6. 6Update the PDF. Mark changed text in a different colour so reviewers can find it fast, and stay within 10 pages.
7. 7Reply to each reviewer. Quote their point, answer it directly, and say where the change is in the paper.
8. 8Post early, then follow up. Reviewers need time to read and reply before 18 November. A short, polite reminder a few days later is fine.

An ICLR rebuttal reply template

Here is a simple structure for one reply. Change it to fit your paper.

• Thank and summarise: "Thank you for the careful review. We address each point below and have updated the PDF (changes in blue)."
• Quote the concern: "W1: The method is only tested on small datasets."
• Answer in one line: "We added results on a larger dataset."
• Give the evidence: one sentence or a small table with the new numbers, and the section or table where they now appear.
• Admit limits: if you could not run something, say why, and say what you will add in the final version.
• Close with a question: "Do these changes address your concerns? We are happy to answer any further questions."

Keep each answer short. The ICLR 2027 reviewer guidelines ask reviewers to raise only points that could change their decision, because long reviews add work for everyone. The same is true of replies. A reviewer with six papers will skim a wall of text.

Common ICLR rebuttal mistakes to avoid

• Arguing: telling a reviewer they are wrong without evidence. Show it with a number, a citation or a line from the paper.
• Ignoring points: skipping a hard question. Reviewers notice, and so does the AC.
• Only promising: "we will add this" is weaker than "we added this in Section 4".
• Replying late: a reply on 18 November may never get a response.
• Breaking anonymity: ICLR 2027 is double blind. Do not name yourself or your lab in replies or the new PDF.
• Fake fixes: new citations or numbers you have not checked. See our post on fake citations in AI research papers.

Can I use AI to write my ICLR 2027 rebuttal?

The ICLR 2027 AI policy for authors asks you to state how you used large language models (LLMs), both in the paper and in the submission form. The paper must have a required AI use statement, which does not count toward the page limit.

The policy is clear about blame. The authors are responsible for everything in the submission. A serious falsehood, plagiarism or misrepresentation produced by an LLM counts as a Code of Ethics violation by the authors, and can lead to desk rejection. The Code of Ethics also covers paper discussion, which includes your rebuttal.

Reviewers have their own rule. The ICLR 2027 AI policy for reviewers allows limited use of AI tools to help write reviews, but use must be disclosed. Reviewers who use an LLM must also report their original, self-written assessment. LLM-made falsehoods or low-quality reviews can lead to discipline.

Our view: use an AI tool to fix grammar if English is your second language, but write the arguments yourself and check every number. If you think a review was written by an LLM and contains false claims, say so politely to the AC, with the exact lines.

What if a reviewer does not reply?

Many reviewers never answer, and that is normal. Your reply still counts, because the AC reads the whole discussion before writing the meta-review. If a reviewer clearly misread the paper and stays silent, you can post a short, calm comment to the AC that points to the key evidence.

Peer review is a skill you can learn. Our free Publishing and Peer Review course covers reviews, rebuttals and what happens after a decision.

At APJ-RAISE, students read reviews and draft replies together. Joining is free.