Good point. Looking at my database it seems that most of the time I just choose fake answers, and sometimes use words that don't relate to the question (e.g., "Favorite food: Manchester").
As an earlier comment pointed out [1], random responses to security questions are a bad idea. I've personally tested this by entering a random security answer, calling the service, saying I forgot my password and entered gibberish as my security answer, and being let through. I presume technically-savvy people think this is more secure; if I can guess that, an attacker can too.
In social engineering, it is common to call back multiple times looking for a gullible customer service rep. Even using a recording of a crying baby in the background to Garner sympathy is something I've seen done.