Are generative AI text annotations systematically biased?