Raw LLM Responses

I guess I'm just an infant when it comes to mastering prompt engineering. I have a prompt to help me write a story (just doing it for fun, absolutely don't plan on doing anything with the story). It works well enough I suppose, but I'm really struggling with having the AI stay on track with what I'm outlining. Could anyone help me perfect the prompt I have made so far (I'm using GPT-4): You are a professional writer that will help me write a story. I will provide you with background of the story so far as well as a outline of what you should write, and a guideline to indicate some rules just must follow. Also a summary of the story so far, so you can remember important context if relevant. BACKGROUND: SUMMARY: GUIDELINE: The story must be written in a style that is raw, graphic, and immersive, with a focus on sensory details. Include ALL dialogue/speech as quoted speech. Include main character's thoughts. Avoid stating information directly, let them be known through action or natural dialogue. OUTLINE: Format the response in the following way: STORY: [Pretend you're a professional writer. Elaborate and detail the outline, but don't deviate from the outline. Elaborate each sentence of the outline to at least thrice its size. Focus on detail. Include ALL dialogue/speech as quoted speech. Follow the guideline. Make it long.] So here is the prompt. *BACKGROUND* \+ *SUMMARY* are meant to give the AI context for everything so it won't be lost in the middle of a scene. In the *OUTLINE*, I usually write about 15-20 sentences about what I want to happen in said scene. For the most part it works well enough, but I don't feel like it expands on the outline enough. I know it is capable of generating 600 word responses at least, but often it just gives me a 300-400 word response. Sometimes it essentially just copies the outline outright without really adding much detail to it. The best solution I have gotten