This is my exact issue. ChatGPT seems formulaic in part, because so much of the work it’s trained on is also formulaic or at least predictable.