Prompt Engineering
Prompt engineering is the skill of structuring and refining instructions (prompts) so an AI produces accurate, relevant, and repeatable results. Despite the word 'engineering', it does not require a computer science degree — at its core, it is the craft of thinking clearly and communicating exact expectations to a model.
Example: Adding "Answer in a table" or "Give me 3 options" can change the result completely.
Why is it called 'engineering' if it is just plain writing?
When you talk to a coworker, you rely on shorthand because they already know your workplace, your tone, and what a good finished document looks like. When talking to an LLM or video model, a single vague phrase leaves the AI guessing.
It is called prompt engineering because experienced users treat a prompt like a reusable blueprint: they define the role, supply reference examples, lock the output format, and set clear guardrails. Once you test and refine that blueprint, you can reuse it every week and get dependable results in seconds.
4 practical prompt engineering techniques anyone can use
You do not need technical jargon to level up how you write prompts. Four straightforward habits solve most everyday frustrations:
- Persona (Role prompting): Start by assigning a specific lens, such as "You are a careful Bahasa Melayu editor reviewing a formal company letter."
- Few-shot prompting (Show 2–3 examples): Instead of describing your style in abstract words, paste two or three real examples of your past writing and tell the AI to match that pattern.
- Chain of Thought (Step-by-step reasoning): For logic, budgeting, or planning tasks, ask the model to lay out its reasoning steps before giving the final recommendation.
- Ask-me-first prompting: End your prompt with "Before you write the draft, ask me 3 clarifying questions about anything that is missing."
How to debug a prompt when the first reply misses the mark
Even seasoned users rarely nail a complex task on the very first try. The trick is knowing how to diagnose what went wrong and tighten the instruction:
- If the answer feels generic: Add concrete local context (who the reader is, what industry you are in, and what decision the text needs to support).
- If the answer rambles: Set hard structural boundaries, such as "Give me a 4-column table and keep each cell under 15 words."
- If the tone sounds stiff or robotic: Tell the model who is speaking and ask it to rewrite in natural Malaysian workplace English or Bahasa Melayu.
Prompt engineering for office text vs AI video directing
In text tools like ChatGPT and Claude, prompt engineering revolves around background rules, document structure, and tone examples. In generative video, prompt engineering becomes the craft of a film director.
Rather than throwing random adjectives at a video generator, a structured video prompt separates subject appearance, physical action, environment lighting, and camera movement so models like Seedance or Kling render a coherent shot.
Common questions
What is the difference between a prompt and prompt engineering?
A prompt is the single instruction you type into the AI, while prompt engineering is the method of designing, testing, and structuring that instruction — using roles, examples, and constraints — so it works reliably.
Are '10,000 secret prompts' bundles worth buying?
Almost never. Giant copy-paste lists are usually generic one-liners written without your context. Learning a handful of core prompt engineering principles lets you craft custom instructions for your own work in under a minute.
Does prompt engineering still matter as AI models get smarter?
Yes. Smarter models need less hand-holding on basic grammar, but they still cannot guess your internal company rules, your target audience, or the exact camera shot in your head unless you communicate those clearly.
Ready to practise prompt techniques you can use right away? Try the bite-sized workflows at /tips/ or read the guides at /belajar/.