Common sense. Case by case basis. Sometimes telling someone you're going to kill them online is a threat, sometimes it's bullshit (although still possibly prosecutable as harassment, just not assault).
How do you differentiate the bullshit from the legitimate threat?
That's the judge or jury's job. I hope they will err toward very stringently disallowing people from saying hateful and stupid things on the internet (especially social media, which is more personal than a message board), but then I don't really give a damn about the insane obsession with free speech. EDIT: incidentally, there are currently guidelines in place for differentiating threats and non-threats, because various legal questions already turn on that issue. (Presidential threats and so on.) I'm not sure if those rules are any good, but the procedural precedent is there.