Should decoder only models like GPT used for classification?

Hi I am wondering if I should use GPT like decoder only models for classification? I understand that these models are most suited for things auto-complete the sentences predicting the most probable next word, due to the representation computed for a given token in this architecture depending only on the left context. However, since it is technically possible can/should such a model be used for classification?
Are there any use cases for using it at a classification task?