This time, I would like to share a few ideas that I think can be helpful when writing rebuttals for AI conferences.

1️⃣ Not all reviewer comments are equally important

Reviewers usually raise several comments or concerns, but not all of them have the same weight. In many cases, only a few issues are actually critical to the reviewer’s final decision, while the rest may simply be smaller weaknesses they noticed while reading the manuscript. One of the most important things to do when writing a rebuttal is therefore to identify which concerns are truly driving the reviewer’s decision. Those are the issues you should prioritize and emphasize in your response, rather than giving every comment the same amount of space. This becomes especially important when the rebuttal has a strict word limit.

2️⃣ Your rebuttal is also for the AC

Everyone has their own perspective, so we cannot expect a rebuttal to convince every reviewer. In fact, the rebuttal process is more like a debate. The reviewers and the authors each present arguments and evidence from their own perspectives about whether the paper makes a sufficiently strong contribution. The AC then plays a role similar to a judge. When the two sides disagree, the AC evaluates the arguments and evidence presented by both sides and decides which position is more convincing.

Because of this, getting a reviewer to raise their score is not the only goal of a rebuttal.

Even if you cannot fully convince a particular reviewer, the rebuttal can still be successful if you clearly demonstrate to the AC that the reviewer’s concerns have been reasonably addressed, or that certain criticisms do not actually undermine the paper’s core contribution.

3️⃣ Authors and reviewers are equals in the discussion

Based on the perspective above, It becomes easier to see the rebuttal as an academic discussion between authors and reviewers as equals. You should not feel obligated to accept or satisfy every reviewer request simply because you are worried about upsetting them. Reviewer feedback is important, but that does not mean every request is necessarily reasonable. Authors also have a responsibility to evaluate whether a suggestion genuinely helps clarify an issue with the paper or affects its central conclusions. A common example is when a reviewer asks for an additional experiment. If you believe that the experiment is only loosely related to the paper’s main argument, or that its outcome would not change the current conclusions, the best response may not be to immediately run the experiment. Instead, it may be better to first explain why the experiment is not necessary and try to understand whether the request comes from a misunderstanding of some part of the paper.

4️⃣ Reiterate the paper’s contributions

When responding to reviewer comments, it can be useful to periodically remind reviewers of the paper’s core contributions or point out that certain ideas have already been addressed in the paper. For example, if you add a new experiment, explain how the new result provides further support for the paper’s original argument. If a reviewer becomes overly focused on a relatively minor detail, you can also redirect attention to the problem the paper is actually trying to solve and what its main contribution is. The goal is not simply to repeat the paper. Instead, you are helping the AC evaluate whether the reviewer’s concern actually affects the paper’s core contribution.

5️⃣ A rebuttal does not need to be extremely long

Some authors write very long rebuttals with many new experiments and extensive details, but more is not always better.

First, reviewers may not have enough time to digest a large amount of new information. Each reviewer is usually handling multiple papers at the same time, and some time may have passed since they last read your paper. As a result, it is easy for some important arguments to be overlooked or misunderstood. Instead of presenting a large amount of new material all at once, it is often better to first summarize the paper’s core contribution in one or two sentences, then explain in another sentence or two what conclusion the new result supports, and only after that provide the necessary details.

Second, from the AC’s perspective, if a paper requires a large number of new experiments and extensive explanations during the rebuttal in order to address the reviewers’ concerns, it may raise questions about whether the original manuscript is sufficiently complete. This is especially true if the new results are substantial enough to change how the paper should be positioned or framed.

So the goal of a rebuttal is not to provide as much information as possible. It is to communicate the most important information as clearly and efficiently as possible, so that both the reviewers and the AC can quickly understand what really matters.