Frankly, I'd expect the percentage to be significantly less than 50%. Determining good from bad bots without a manual check is an interesting problem, though, and it's not cool on Twitter's part if they ban legit bots that people find useful/entertaining on top of the bad actors.