Short answer: Prompt engineering is the practice of writing clear instructions for an AI model so that it produces useful, checkable output. For most professionals the basics are: state the task, give context, specify the format, provide an example, set limits, and review the answer. Good prompts reduce rework, but they do not remove the need to check facts, because language models can produce plausible but incorrect statements (the provider guides listed below describe the limitations of the models).
What is a prompt, and why does wording matter?
A prompt is the text you give a model. Models respond to what you wrote, not what you meant. Vague instructions produce generic output, and specific instructions usually produce more relevant output. Providers publish prompting guides, and their advice changes as models change. Provider guides, such as Anthropic's prompt engineering documentation and OpenAI's prompt engineering guide listed below, are the best place to check current advice for the model you use.
What are the building blocks?
- Role and audience. "You are helping a finance manager write for a non-technical board."
- Task. One clear verb: summarise, compare, draft, classify, check.
- Context. Paste the relevant background. Leave out confidential or personal data unless the tool is approved for it.
- Format. Ask for a table, bullet list, or word limit.
- Examples. One or two samples of the output you want often help.
- Constraints. What to avoid, what sources to use, what to do when unsure.
What does a before-and-after look like?
Before: "Write about our product."
After: "Draft a 120-word email to existing small-business customers announcing a new report feature. Tone: plain and friendly. Include one sentence on what it does and one clear next step. Do not claim performance numbers."
How do I get better answers when the first one is poor?
Iterate. Say what was wrong: too long, wrong tone, missing a point. Ask the model to ask you clarifying questions first. Break a big task into steps. Ask it to list assumptions it made.
How do I handle accuracy?
Treat output as a draft from a fast but fallible assistant. Verify numbers, names, citations, legal statements and anything you will publish. If you ask for sources, check that they exist and say what the model claims. Never submit unchecked AI output as expert advice.
What should I never put in a prompt?
Anything you are not allowed to share with a third party: customer personal data, passwords, unpublished financial results, or client confidential material. Check your organisation's policy and the tool's data terms. The PDPC's guidance on personal data in AI systems and on NRIC numbers, listed below, explains how the PDPA applies. Read the vendor's terms on data retention and on whether your inputs are used for training.
How do I build team habits?
- Keep a shared library of prompts that worked, with notes on what to check.
- Record which tool and version was used.
- Agree who reviews AI-assisted work before it leaves the organisation.
Is prompt engineering a long-term skill?
Models are getting better at understanding loose instructions, so the tricks matter less over time. The durable skills are clear thinking, knowing your subject well enough to spot errors, and good judgement about when not to use AI.
Keep learning
Our Academy lists courses including a Prompt Engineering Masterclass: see asia-ai.org for current availability (listed on the Academy page on 21 November 2026 as online, about 3.5 hours, with a certificate, at S$0). The free monthly newsletter is at asia-ai.org, and membership details are at asia-ai.org/join.
This article is general information, not legal advice.