This is “chain of thought” (and a few others based on “chain of thought”), and yes it gives much better results. Very common thing to train into a model. Chatgpt will do this a lot, surprised it didn’t do that here. Only so much you can do I suppose.