The ICLR 2027 rebuttal and discussion period

From the official dates page, which gives time zones only for the submission deadlines. Treat Nov 18 as end of day Anywhere on Earth, and don't leave anything to the last hour.

• It's public. Official reviews are anonymous and publicly visible on OpenReview, and so are most comments. You can restrict a comment to reviewers, area chairs or program chairs.
• It's a conversation. Each comment has a word limit, but there is no limit on the number of comments. Reviewers can reply, and you can follow up.
• Scores can move. ICLR's reviewer guidelines say reviewers can change their reviews after the author response.
• You can revise the paper until the end of the discussion on Nov 18, and the main-text limit rises from 9 to 10 pages to make room for new results. Reviewers are not required to read every revision, so tell them what changed.
• Your own reviewing counts. Under ICLR 2027's reciprocal-reviewing policy, authors who don't deliver complete, high-quality reviews by the rebuttal stage may have their own submissions desk-rejected. If you're reviewing too, finish those first.

In 2026 ICLR reviewers rated papers on an even-numbered scale: 0 strong reject, 2 reject, 4 marginally below the acceptance threshold, 6 marginally above, 8 accept, 10 strong accept. ICLR doesn't publish these labels on its own site (they are quoted from the 2026 form by secondary sources), so check your own review form.

ICLR 2026 accepted 26.95% of 19,814 submissions. According to Paper Copilot's public OpenReview data, accepted papers averaged 5.39 (posters 5.35, orals 6.39) and rejected papers 3.96. Use those numbers as rough reference points, not as a cutoff:

1. Read all reviews once. Don't reply to anything today.
2. Put every distinct concern in one table: reviewer, quote, their rating and confidence.
3. Merge concerns raised by two or more reviewers. Those go in a general comment to everyone.
4. Tag each one: factual misreading, missing experiment, clarity, or scope/significance.
5. Find the persuadable reviewers: ratings of 4 or 6 with concrete questions. A 4 with confidence 2 is easier to move than a 4 with confidence 5.
6. Start the experiments now. You have about two weeks, and reviewers need time to read and respond before Nov 18.

• Post one general comment summarising all changes, then one reply per reviewer.
• Answer the key weakness first. ICLR asks reviewers to lead with the one or two reasons behind their rating, so start your reply there.
• Each answer: restate the concern in one line → direct answer → evidence (number, table, section) → what changed in the PDF.
• Reply early. A reply posted on Nov 6 can get a follow-up question; one posted on Nov 17 usually can't.
• If a reviewer goes quiet, one short, polite nudge that lists what changed is fine. Repeated pings aren't.
• Everything is public. Write for the area chair and future readers, not only the reviewer.

For the general method (triage table, answering “reviewer 2”), see how to write a rebuttal.

• Highlight changed text (a colour is fine) and list the changes in your general comment, since reviewers don't have to re-read every version.
• Keep the main text within the discussion-phase limit of 10 pages. References and appendices don't count.
• Don't turn it into a different paper. The guidelines say a revised title or abstract shouldn't read like a new submission, and reviewers are told new experiments shouldn't significantly change the content.
• If LLMs did substantive work in new results or analysis, update the AI-use section required by ICLR 2027's author AI policy.

• Replying on the last day, when no one can respond.
• Long replies to minor points and one vague line on the main weakness.
• Promising results "in the final version" when you could run them now.
• Revising the PDF without saying where, so the changes go unnoticed.
• Arguing about the rating itself. Answer the reasons; the rating follows.
• Neglecting your own assigned reviews, which can now put your submission at risk.