Prompt engineering — how to write effective prompts for LLMs

What is prompt engineering?

Prompt engineering is the art of phrasing queries to language models so that you get predictable, accurate and useful results.

A good prompt = clear context + a specific task + a response format.

Bad vs good prompts

❌ Bad — too general:

Write some PHP code.

✅ Good — specific:

Write a PHP 8.3 function that:
- takes an array of strings
- filters out elements longer than 5 characters
- returns a new array with the results
- uses array_filter and fn()
Add a usage example with var_dump().

Techniques

1. System prompt — assign a role

You are a senior PHP developer with 10 years of experience in Symfony.
Answer concisely. Write code following PSR-12.
If a question is unclear, ask before answering.

2. Few-shot — show examples

Convert titles into slugs:
"Hello World"       → "hello-world"
"How are you?"      → "how-are-you"
"PHP 8.3 Tutorial"  → "php-8-3-tutorial"
"Zażółć gęślą jaźń" → ?

3. Chain of thought — think step by step

Solve this performance problem step by step:
1. Describe what this code does
2. Identify the bottleneck
3. Propose an optimization
4. Show the fixed code

Code:
[paste code]

4. Output format — specify the response format

Respond in JSON format:
{
  "problem": "...",
  "solution": "...",
  "code": "..."
}

Tools

A prompt for code review

Analyze the PHP code below as a senior developer.

Look for:
- security bugs (SQL injection, XSS, CSRF)
- performance problems
- SOLID violations
- missing error handling

Format:
## Critical
- [problem description] → [suggested fix]

## Important
- ...

## Suggestions
- ...

Code:
```php
[your code]

---

## LLM limitations

- **hallucinations** — the model may make up facts; always verify
- **knowledge cutoff** — the model doesn't know the latest libraries
- **context** — long conversations may "lose" earlier decisions
- **determinism** — the same prompt may give different answers (the `temperature` parameter)