If there's one thing that pollsters can tell you with great certainty, it's that respondants are to a great degree reticent to give the answer "I don't know" to anything, even if they really have no bleeping clue as to what's going on. It makes them feel stupid--they don't even care what the other person thinks of them; it's that they themselves don't like thinking that "they don't know" regarding an answer. There's a cottage industry devoted to determining how much "noise" results from surveys as a result of this false certainty. If you combine this finding with how automated polls generally work, you can figure out why their undecideds are so low.
Even though I've never been the subject of a SUSA automated poll, I can guess what they're like:
"Hello, this is the SUSA automated phone poll. If the primary election were today, who would you vote for? For Kerry press '1,' for Clark press '2', for Dean press '3'...if you are undecided press the '*' key."
Unlike live phone polls, where you can say, "uhh...I'm sort of leaning between Clark and Kerry--mostly Clark, but I'm still thinking about it," a SUSA poll would just tell you to push a button. And a lot of people who would in a live poll give the above answer, or even only name one candidate but add the adjective "leaning" to it and therefore get recorded by a human polltaker as 'undecided', would in a SUSA just push the button for the candidate for which they think they feel strongest instead of feeling silly about hitting the undecided key. That's why the undecideds are so low in SUSA polls.