Most LLM users wants to do less work, not more. Shaping LLM output to be human like is a lot of work, and that assumes you know what the human version looks like.