Prompt engineering is the practice of designing inputs to an LLM to reliably get the output you actually need — not a trick or a "magic phrase" hunt, but a genuine, testable engineering discipline with real techniques and real failure modes.
A Before/After Example
Weak prompt:
"Summarize this"
Better prompt:
"Summarize the following support ticket in 2 sentences: the
customer's core issue, and the current resolution status.
Do not include greeting text or sign-offs.
Ticket: [ticket text]"The second version removes ambiguity about length, focus, and format — the model has far less room to guess wrong. This is the core of prompt engineering: reducing ambiguity so the model's output matches your actual intent.
The Techniques Covered in This Section
| Technique | What It Solves |
|---|---|
| Prompt Structure | Organizing a prompt's parts so the model can parse intent clearly |
| System Prompt | Setting persistent behavior/persona separate from the user's actual request |
| Zero-Shot, One-Shot, Few-Shot | Teaching the model your desired output format via examples (or none) |
| Chain-of-Thought, Reasoning Prompts | Improving accuracy on multi-step problems |
| Role Prompting | Framing the model's perspective/expertise for the task |
| Prompt Templates, Prompt Chaining | Making prompts reusable and composable in real applications |
| Optimization, Evaluation | Systematically improving and measuring prompt performance |
| Prompt Injection, Prompt Security | Defending against manipulated input |
Prompt Engineering Is Not Guesswork
A common misconception treats prompting as trial-and-error until something "feels" right. In production, it's treated as an actual engineering practice: a prompt is a piece of application logic, versioned, tested against real examples, and evaluated against clear criteria before and after changes — see Prompt Evaluation.
Common Mistakes
- Treating prompt wording as a one-time decision instead of something to test, measure, and revisit as the model or use case evolves
- Adding more and more instructions to "fix" bad output without first identifying which specific ambiguity is causing the problem
Interview Relevance
Q: "What is prompt engineering, really — is it just finding magic phrases?" A strong answer rejects the "magic phrase" framing and describes it as systematically reducing ambiguity and testing output against clear criteria.
Practice Question
Take the weak prompt "Write something about our product" and rewrite it with explicit structure, audience, length, and format.