The implementors of these sites have come up with a set of heuristics that attempt to detect spamming. I don't know exactly what they are, and if I did, I don't think they would be happy with me for typing them here. However, from my own observation, I've concluded that the site shows a Spam Haiku when the heuristics go off. Human beings scratch their heads, retry, and succeed. Actual robot spam devices are stymied. I imagine that if you thought a bit about what sorts of time and content patterns might be associated with automated spam, you could figure out why you got to read one.