The design of homework tasks is oriented towards objectives of varying cognitive complexity. Determining the cognitive demand of these tasks can be a challenging activity especially for novice teachers. In this paper we use ChatGPT to determine the level of cognitive demand implied by didactic tasks considering different models or taxonomies. As a case study, we consider tasks present in textbooks, in the area of Mathematics and History. In an a priori evaluation, there is a lack of consensus among the experts’ assessment when assigning the level of cognitive demand in the tasks of both subjects, even when using a frequently used taxonomy such as Bloom’s; this shows how complex to classify tasks can be, even for experts with extensive experience and specialization. Regarding the classification and argumentation made by ChatGPT, there is a better valuation of the tool within the experts of the area of History than in the Mathematics, both in the classification and in the arguments delivered. All the experts positively value the potential of the chat for the purposes discussed, but they mention a series of risks that need to be considered.