Usually, a conference enforcing a rebuttal process will also define what kind of rebuttal an author can write. For instance, the conference ACM CCS (considered as one of the best conferences in security) enforces a rebuttal phase. Let me quote what they wrote during last year edition:
You now have the opportunity to view the preliminary reviews of your
submission and, if you wish, enter a response limited to 500 words.
This response is completely optional and there is no requirement to
respond. If you do, the deadline is [3 days after receiving this message].
Your response must focus on the following:
We stress that your paper is being evaluated as submitted. You may
NOT use your response to provide new research results or reformulate
Another conference, POST, uses a similar process with similar requirements, with an extra suggestion that I think it's worth mentioning:
- Your response will be seen by all PC members who have
access to the discussion of your paper. Please be polite
So, the bottom line is: keep it factual, polite and constructive. If a reviewer doesn't like your paper, then it's unlikely you can change his mind during the rebuttal phase. However, it's just a good opportunity to address some very specific point. For instance, if a reviewer asks: "Isn't your approach undecidable?", then you can answer "yes/no, and we can include the proof in the final version of the paper" (and ideally, link to a research report where the proof is already written). Or if a reviewer wrote "this problem has been already solved 20 years ago by X", then you can answer "We released one of X's assumption, that we believed was too strong for this particular context".
Basically, the rebuttal might be unlikely to change a particular review (unless there was an obvious mistake), but can encourage the PC Chair to ask for another review of your paper. And Jeff's remark is very good, don't write the rebuttal the same day than you receive it :)