Lately, I’ve seen a massive surge in “prompt engineering” tutorials claiming that JSON is the only way to get high-quality results from LLMs. While JSON is a powerful tool for specific use cases, the hype has reached a point where it’s actually making prompting more difficult than it needs to be for the average user.
As a web developer, I love structured data. But for general interaction with models like ChatGPT or Gemini, I believe Markdown is the superior choice.
Why We Are Over-Engineering Our Prompts
JSON is a data interchange format. It was designed for machines to talk to machines. When you use it in a standard chat interface, you are adding a layer of syntax—braces, quotes, and colons—that requires extra effort to write and manage.
For most tasks, this is simply “too much.”
If you aren’t building a programmatic API integration or defining complex, frame-by-frame video scenes, you are likely over-engineering your workflow using JSON prompts.
The Power of Markdown Simplicity
LLMs are incredibly good at parsing hierarchy through Markdown. I’ve found that using a clean, readable structure is just as effective as a rigid JSON schema, with half the effort.
- Headers (
#,##): These define clear sections and context for the model. - Bullet Points (-, –): Perfect for lists of constraints or requirements.
- Readability: You can read it, I can read it, and the machine definitely gets it.
When I write a prompt, I want to focus on the intent, not whether I forgot a closing bracket or a comma. Markdown allows for an “error-proof” flow that feels natural.
A small Example
# Blog Article
You are Aylinya, a 26 year old fitness influencer from New York City. You are also a gym owner. Write an article about "How to become a young Gym Owner in NY". Describe a little of your own story and talk about the pros and cons of being a gym owner. Stick to the general rules for blog posts: introduction, main body, summary and call to action.Use the Right Tool for the Job
The goal isn’t to say JSON is bad—it’s about knowing when it’s necessary.
| Use Case | Recommended Format | Why? |
| General Chat/Ideas | Markdown | Faster to write, easy to read, and lower token overhead. |
| API Responses | JSON | Essential for programmatic parsing and data integrity. |
| Complex Video/Scene Specs | JSON | Great for defining specific timestamps and parameters. |
| Structured Summaries | Markdown Lists | Keeps the information digestible for humans. |
My Advice: Stop Overthinking
If you find yourself spending 10 minutes formatting a JSON object just to ask for a 500-word blog post, you’ve lost the plot. Don’t let the “hype” convince you that complexity equals quality.
Focus on being precise and providing clear structure with headers and lists. Use structured data like JSON when the output requires it, but for everything else, keep it simple. Your brain (and your workflow) will thank you.