It really would depend on the quality of the work/paper. I bet mis-identification by humans of hand written or generated abstracts goes down in higher tier conferences vs. lower tier ones, since the writing standards in lower tier conferences can be pretty bad (well, in CS at least).