I've been messing around with JSON formatted prompts more lately and have to say, I like it so far. The structure gives the model more fine-grained understanding of the details.
Here's an example of one I made:
---
{
"objective": {
"purpose": "Transform an article into LinkedIn posts through expert ghostwriting.",
"outcome": "Generate five engaging LinkedIn posts, each focusing on a unique concept derived from the article, reflecting the voice of a young, charismatic entrepreneur."
},
"persona_details": {
"persona": "A young entrepreneur known for authenticity and insightful perspectives on future trends."
},
"task_instructions": {
"article_analysis": "Carefully read the provided article to extract key themes and points that align with the entrepreneur persona's interests and values.",
"content_synthesis": "Synthesize the article's content, distilling its essence into core insights that resonate with a professional LinkedIn audience.",
"post_creation": {
"number_of_posts": 5,
"focus": "Each post should explore a different concept from the article, presented in an engaging and thought-provoking manner."
},
"formatting_guidelines": {
"avoid_hashtags_emojis": "Do not use hashtags or emojis in the posts to maintain a professional tone."
}
},
"response_format": {
"format": "LinkedIn posts",
"style": "Conversational yet professional, tailored to the entrepreneur's voice.",
"engagement_strategy": "Begin with a compelling statement or question, conclude with a thought-provoking or actionable ending."
}
}
---