Don't just ask.
Engineer the answer.

1 What is a token and how does the system count it?

A token is the smallest unit of text processed by the model (word fragment, punctuation, whitespace).

The model tokenizes all incoming text and every outgoing response:

• prompt tokens (what you write)
• system + assistant + developer messages
• response tokens
• full context (history)

2 How many tokens does a prompt use?

It depends on:

3 How to measure actual token usage?

"usage": {
  "prompt_tokens": 123,
  "completion_tokens": 98,
  "total_tokens": 221
}

Tools: tiktoken (OpenAI official), online token counters

4 How to optimize token usage?

Task:
Style:
Length:
Output format:

5 How much can you save with optimization?

1 The 4 Building Blocks of a Good Prompt

You are a senior JavaScript developer.
Task: explain how fetch API works.
Context: I'm a beginner.
Output: 5 steps, with short code.

2 Explanation + Example Technique

Always ask for an example in the learning process, otherwise you'll only get theory.

Explain briefly what a closure is in JavaScript.
Also give me a very simple example.

3 Fix My Code – Fix + Explain + Example

Always ask for a fixed version, explanation, and working example.

You are an expert developer.
Fix the bug and give me a correct example.

4 Learning Prompt (ELI5)

Simplify complex topics. Explain as if I were 10 years old.

What is an API? Give me a very simple example.

5 Code Generation – Output Format

Always specify the expected format: HTML / CSS / JS blocks.

Create a navbar in separate HTML-CSS-JS blocks.

6 Iterative Refinement

Use "Refine" command to fine-tune the response.

Refine: make it shorter, in Hungarian, give example.

7 JSON or Table Output

For structured responses, always specify the format.

{
  "fogalom": "",
  "magyarázat": "",
  "példa": ""
}

8 Prompt Optimization

Have the AI optimize your own prompt.

Act as a prompt engineer. Optimize this: [PROMPT]