We've all done it. We open ChatGPT or Claude and type: "Write a function to validate emails."

And the AI gives us a decent function. But it's in Python, and you wanted JavaScript. It uses a regex that doesn't validate real domains. It doesn't handle errors.

This is the "garbage in, garbage out" principle of AI. Prompt Engineering isn't about tricking the AI; it's about providing the context you already have in your head but forgot to type.

1. The Persona (Who)

LLMs are trained on the entire internet—Reddit threads, academic papers, and YouTube comments. If you don't steer them, they aim for the average.

Assigning a persona narrows the search space to high-quality data.

Bad: "Explain how to center a div."

Good: "Act as a Senior CSS Developer. Explain how to center a div using modern Grid and Flexbox techniques, avoiding legacy floats."

2. Context & Constraints (What & How)

Tell the model what you don't want. Constraints are often more powerful than instructions.

  • "Use TypeScript, not vanilla JS."
  • "No external libraries."
  • "Keep the response under 100 words."
  • "Format the output as a JSON object."

3. Chain of Thought (The Reasoning)

For complex logic, LLMs often hallucinate because they try to predict the answer immediately. Asking them to "think step by step" forces them to generate intermediate reasoning tokens, which increases accuracy.

Prompt: "Critique this design. Think step by step. First, validate the usability, then check color contrast, and finally give a score."

4. Few-Shot Prompting (examples)

The best way to get a specific format is to show, not tell. Give the AI 1 or 2 examples of what you want.

Task: Convert movie titles 
Example 1:
Input: "Star Wars"
Output: "star-wars"

Example 2:
Input: "The Matrix"
Output: "the-matrix"

Input: "Lord of the Rings"
Output:

The AI will almost certainly output `"lord-of-the-rings"`.

Automating the Perfect Prompt

Remembering all these rules (Persona, Context, Constraints, Format) is tedious every time you want to write an email or code snippet.

That's why we built the Prompt Optimizer. It's a local tool that takes your basic request and expands it into a structured, engineer-grade prompt using these exact principles—no account required, no data sent to a server.