The first time I asked ChatGPT for help, I typed: "Write me a product description for noise-canceling headphones."
It gave me garbage.
I blamed the AI. But the problem was me.
"Be Clear" Doesn't Mean "Be Long"
For months, I wrote these rambling prompts explaining every little detail.
Results were still hit-or-miss.
Then I learned something simple:
DELIMITERS
They're the lines you draw in your prompt to say, "This part is my instruction. This part is the content I want you to work with. Don't confuse the two".
Think of it like writing on a whiteboard during a meeting. You don't just scribble everything together in one giant blob. You draw boxes. You underline headers. You separate the problem from the solution.
Delimiters do the same thing, they keep your instructions from bleeding into your content, so the AI knows exactly what's what.
Eg: Triple quotes """, XML tags , or even simple dashes ---— pick whatever feels natural.
Suddenly, the AI knew exactly what I wanted.
Role:
You are an AI writing assistant for students.
Task:
Improve clarity; keep it short; avoid complex words.
Rules:
- Do not change meaning
- Output only the improved text
- No extra explanations
Input | Metadata:
topic: AI prompting | level: beginner | tone: casual
---
Original text:
"""
AI works good but sometimes it gives wrong answers because the question is not clear.
"""
---
Output format:
improved_text: <one improved sentence>
Stop Rushing the AI (And Yourself)
I used to treat AI like a vending machine.
Input -> enter -> expect perfection.
Then someone told me: give the model time to think.
Not literal time. Structure.
Instead of "Solve this," I learned to say:
"First, analyze the problem" - Understand what you're dealing with
A train travels 60 km in 1.5 hours.
The speed of the train is constant.
We are asked to find the time required to travel 100 km at the same speed.
So this is a speed–distance–time problem.
"Then, identify approaches" - Explore different ways forward
We can solve this by:
Calculating the speed of the train, then
Using the same speed to find the time for 100 km
Formulae involved:
Speed = Distance ÷ Time
Time = Distance ÷ Speed
"Finally, provide your solution" - Deliver the answer that fits best
Calculate speed:
Speed = 60 km ÷ 1.5 hours
Speed = 40 km/h
Now calculate time for 100 km:
Time = 100 km ÷ 40 km/h
Time = 2.5 hours
The difference was massive.
Four Ways to Ask (That Actually Work)
Zero-shot: Just the instruction. No examples.
Works when the task is straightforward.
Prompt:
Roast my coding skills.
Output: Your code works, but even it looks surprised by that.
Few-shot: Multiple examples showing a clear pattern.
My go-to for consistency.
Prompt:
Roast my coding skills.
Examples:
"Your variable names are so creative, even you forget what they mean."
"Your code has comments because future-you will be very confused."
Now roast me.
Output: Your code is like a mystery novel - long, confusing, and the ending barely makes sense
Context-based: Background + constraints + question.
Changed everything.
Prompt:
Context:
I’m a student who codes regularly.
This roast can be harsh.
Tone should be playful and relatable.
Task:
Roast my coding skills in one line.
Output: Your code works, but only because the bugs are too tired to fight back
Same question. Completely different answer.
What Finally Made It Click
I stopped thinking of prompting as "giving commands."
I started thinking of it as talking to someone who wants to help but needs to know what you actually need.
You wouldn't tell a colleague "Do the thing."
You'd give context, constraints, examples.
That's all a good prompt is.
Start Here
- Use delimiters.
- Separate your instructions from your content.
- That alone will transform your results.
- The rest will come naturally once you realize AI isn't a tool you command, it's a collaborator you communicate with.
- And the quality of what you build together depends on how clearly you communicate what you're trying to build.
Top comments (3)
That sounds good.
Good work__
Woww!! Nice experience.