Social biases in AI-generated creative texts: A mixed-methods approach, in the Spanish context