Evaluating policy position stances using large language models