If our spammers are, in fact, bots—sometimes I wonder about some of them—it might be easy enough to fool them by suggesting the answer in the question (e.g. "Which of the following is related to Cao Cao: Zhou Yu, Zhuge Liang, or Cao Teng?"). A human can probably deduce, easily enough, that the answer is Cao Teng. If they are familiar at all with the Three Kingdoms, that's even easier due to knowing the 'first name is family name' association. It should defeat bots because there is no way to construct an effective query to get the answer without human intelligence. At least without a pretty sophisticated algorithm, which could also be defeated by picking an answer that the question does not reflect in Google results as a proper noun.
Good point about Wolfram Alpha. It would be relatively easy to detect that a question is math and run it through their search for an answer to plug into the quiz.