Assessing the Impact of Conspiracy Theories Using Large Language Models