The web site says that Cleverbot learns from the conversations it has with people.

I assume that means he keeps all the things that people say to him in a big database, and tries to repeat them back at appropriate times.

The difficult part, from a programming point of view, is making his responses seem somehow related to what you have said to him. He is sometimes successful at this and sometimes not.

Because of this, I try to always use correct grammar, correct spelling, and appropriate punctuation when I talk to Cleverbot. I also try to use simple sentences, without a lot of subordinate clauses, and so on. I figure that will help Cleverbot look smarter and make more appropriate responses in the future.

It's like talking to a child: you want the child to learn from you and not pick up bad habits.

Obviously, not everyone cares about this, which is why you see so many misspelled words, bad grammar, and non sequiturs in Cleverbot's responses.

Probably some people think it's more fun to confuse Cleverbot, and trick him into saying things that sound totally stupid.

I have seen this kind of behavior in kids at a science museum. Some kids think it's more fun to manipulate the controls of the robot (or whatever) to make it do something totally chaotic than something sensible and potentially useful.