I think a big problem is Sybil attacks. Let's suppose I make a bot that posts thousands of reports of my computer crashing after using product X. To determine if such a post is bad behavior or not, you have to know if the claim is genuine. That can't be determined by the content of the post itself.