How Korean Street Food Taught Me to Write Better AI Prompts
Topic: Busan, South Korea
Ever tried ordering at a Busan pojangmacha at midnight with zero Korean and maximum hunger? I have. And somewhere between my garbled pointing at tteokbokki and the auntie's patient "What EXACTLY do you want?" — I had an AI epiphany.
Your prompts are basically food orders. And most people are ordering like confused tourists.
The Pojangmacha Problem
Here's what bad ordering looks like: "Give me... food?"
Here's what bad prompting looks like: "Write me a blog post about AI."
Both will technically get you something. But the auntie is going to guess, the AI is going to hallucinate, and you're going to end up with a bowl of things you didn't want.
The magic of a good Busan street food stall is specificity. You don't just say "tteokbokki." You say: mild or spicy? With fish cakes? Extra sauce? To eat here or take away? The more you specify, the closer you get to the exact taste you had in your head.
AI prompts work the exact same way.
5 Street Food Rules That Will 10x Your Prompts
1. Specify the Spice Level — Set the Tone
In Korea, "not too spicy" means very different things to different aunties. In AI, "write something good" means very different things to every model.
Instead of: "Write a fun blog post."
Try: "Write in a casual, friendly tone — like you're texting a knowledgeable friend, not writing a textbook. Use short sentences. Add one moment of humour per section."
Spice level: specified. 🌶️
2. Name the Dish — Give a Clear Task
You wouldn't walk into a Korean BBQ restaurant and say "do the meat thing." You say galbi, samgyeopsal, bulgogi. Specific names get specific results.
Instead of: "Help me with social media."
Try: "Write 5 Instagram caption hooks for a post about AI productivity tools. Each hook under 15 words. Punchy, not corporate."
Dish: named. 🥩
3. Tell Them Where You're Eating — Provide Context
The same bibimbap hits different at a Gwangjang Market stall versus a Gangnam restaurant. Context changes everything.
Instead of: "Explain machine learning."
Try: "Explain machine learning to a 45-year-old restaurant owner in Busan who has never written a line of code but wants to use AI to manage inventory. No jargon. Use a cooking analogy."
Context: delivered. 🍚
4. Point at the Example — Show What You Want
When language fails, Koreans point. It's efficient. It works. You show exactly what you want.
In AI: include an example output. Even a partial one does the job.
Instead of: "Write a product description."
Try: "Write a product description for [product]. Match this tone and length: [paste an example]. Avoid corporate buzzwords and vague superlatives."
Example: pointed at. 👉
5. Check Before You Pay — Test and Refine
A good Busan food stall owner always does a final taste before serving. No shortcuts on quality.
In prompting: always read, then refine. If the first output is 80% there, prompt again: "Make section 2 funnier. Section 3 is too long — cut to 3 sentences."
Output: quality-checked. ✅
The Real Prompt Engineering Secret
Here's what Busan street food taught me about AI: the auntie wants to help you. She is not trying to get your order wrong. She just needs information.
ChatGPT, Claude, Gemini — they all want to give you exactly what you are after. They are standing there, ladle in hand, ready to serve. Your only job is to order clearly.
Vague order = vague food. Specific prompt = specific, useful output.
The best prompt engineers are not the ones who know the most jargon. They are the ones who communicate the most clearly. Just like the best food orders at 1am in Busan are not written in perfect Korean — they are delivered with a confident point and a clear expression of exactly what you need.
Key Takeaways
-
Every AI prompt is a food order — the more specific your request, the closer the output to what you actually wanted.
-
Setting tone, task, context, and an example covers 90% of what makes a prompt go from mediocre to excellent.
-
Prompting is a conversation, not a one-shot command. Refine after the first output just like you would adjust your order mid-meal.
-
The best prompts come from people who think clearly about what they want — not people who know the most technical terms.
-
Busan tteokbokki is better than anywhere else in Korea. That one is not a metaphor. It is just a fact.
Frequently Asked Questions
Q: How long should my prompts be? A: Long enough to be specific, short enough to stay clear. A good prompt is like a good food order — every word earns its place. Two to five sentences is usually the sweet spot for most tasks.
Q: What if my prompt still gives bad results after trying these tips? A: Add more context, show an example output, or tell the AI what you do NOT want. Breaking a complex task into steps — one prompt per stage, like ordering courses — also helps dramatically.
Q: Does this approach work for all AI tools? A: Yes. ChatGPT, Claude, Gemini, and local models all respond to clearer prompts with better results. The specificity principle is universal across every model and platform.
Busan's street food is some of the best in the world — not because the ingredients are magical, but because generations of vendors have refined the art of knowing exactly what the customer wants and delivering it fast, hot, and with pride.
Your prompts can reach that same standard. Start ordering clearly.
JindoPrompt helps you write sharper prompts and get real results from AI tools — for free. Explore the prompt pack at jindoprompt.com.