Column · @practicalaipath
Writing Clearer Prompts for Better AI Results in Singapore
Why Prompt Clarity Matters
Getting useful output from a language model starts with how you ask. A vague or crowded prompt often produces a vague or crowded answer. The same model, given a clear instruction, returns something far more targeted. This is especially important if you are taking a prompt engineering course, where the focus is on learning to communicate with AI systems in a structured way.
Many people treat prompts like casual chat. That works for simple questions, but for complex tasks the model needs more guidance. Think of the prompt as a brief for a busy assistant. The more specific you are about what you want, the less guesswork the model has to do. Over time, building a habit of clarity saves you from editing long, irrelevant responses.
Step 1: Define Your Goal First
Before you type anything, be clear on what you want the model to produce. A summary? A list of pros and cons? A draft email? A creative story? The same starting question can lead to very different outputs depending on how you frame it. If you just say “tell me about solar panels”, you might get a history of solar technology, a cost comparison, or installation advice. None of these is wrong, but only one matches what you needed.
State the purpose at the start. For example: “Write a short comparison of monocrystalline and polycrystalline solar panels for a homeowner in Singapore. Focus on efficiency, cost, and space requirements.” That prompt gives the model a clear audience, a format, and specific criteria. The output will be far more useful right away.
What to watch: If you get an answer that covers the wrong angle, your prompt probably lacked a goal. Try adding one sentence that tells the model what to do with the information.
Step 2: Structure the Prompt Logically
Once you know your goal, arrange the prompt so the model can follow it easily. Start with the main instruction, then add context, then specify any constraints. This is where prompt structure comes into play. A well-structured prompt separates the instruction from the background details.
For example, instead of writing: “I need a recipe for dinner, but I have chicken and broccoli and I want something healthy and quick and it should be Asian-style.” Try: “Give me a healthy, quick Asian-style recipe using chicken and broccoli. Keep the cooking time under 30 minutes.” The second version places the request first, then adds the specifics. The model can process it more directly.
What to watch: Long, run-on sentences make the model guess which part is most important. Break your prompt into short, clear sentences. If you have multiple requirements, list them. A bullet list in the prompt often helps the model check each point.
Step 3: Use a Role or Persona
Role prompting is a simple technique that often improves output quality. By telling the model who it is acting as, you give it a frame of reference. For instance, “You are a career coach. Give me three questions to ask in a job interview for a marketing role.” The model will tailor the language and depth to that persona.
This works because the model has been trained on text where different roles use different styles. A role prompt narrows the range of possible responses. You can also combine roles: “You are a Singaporean chef explaining how to cook laksa to a beginner. Keep the steps simple.”
What to watch: If the role is too broad, the model might still give generic advice. Be specific about the audience and the context. “You are a teacher explaining to a 10-year-old” is very different from “You are a teacher explaining to a university student.”
Step 4: Give Examples When Possible
Few shot examples are a powerful way to show the model what you want. Instead of describing the output format, you give one or two examples. The model picks up the pattern and follows it. This is especially useful for tasks like writing product descriptions, social media posts, or data extraction.
For instance, if you want a list of pros and cons in a table, show one example: “Product: Electric kettle. Pros: Heats water fast, easy to clean. Cons: No temperature control, cord is short.” Then ask for the same format for a different product. The model will match the style and structure.
What to watch: Keep examples short and relevant. Too many examples can confuse the model or make the prompt too long. One or two well-chosen examples usually do the trick. Also, make sure the example demonstrates the exact format you want.
Step 5: Specify the Output Format
Tell the model how you want the answer. Output format matters because the model can produce a paragraph, a list, a table, a JSON object, or a set of bullet points. If you do not specify, the model chooses a default, which might not suit your use case.
Examples: “List the steps as numbered points.” “Give the answer as a table with two columns.” “Write a short paragraph of no more than 100 words.” These instructions are easy to add and they save you from reformatting the output later.

What to watch: If you ask for a format the model cannot produce well (like a complex graph or a specific diagram), it will approximate or give text. Stick to formats the model handles reliably: text, lists, tables, and simple structured data.
Step 6: Set Constraints and Evaluation Criteria
Constraints help the model stay on track. You can set a word limit, a tone (formal, friendly, professional), or a specific angle. For example: “Explain the difference between machine learning and deep learning in under 150 words. Use simple language for a non-technical reader.” The model then knows to avoid jargon and keep it short.
Evaluation criteria are the standards you want the answer to meet. You can ask the model to include certain points or avoid others. For instance: “In your answer, mention at least two real-world applications and do not use technical acronyms.” This gives the model a checklist to work through.
What to watch: Too many constraints can make the output feel forced or incomplete. Pick the three or four most important limits. The rest you can handle by editing the output or by refining the prompt after seeing the first result.
Step 7: Iterate and Refine
Rarely does the first prompt produce a perfect answer. That is normal. Good prompting is an iterative process. You look at the output, see what is missing or off, and adjust the prompt. This is where the concept of iteration becomes practical. You might add more context, change the role, or rephrase the instruction.
For example, if the model lists five steps but you wanted ten, you can try: “List ten steps instead of five. Be more detailed.” If the tone is too casual, add: “Use a formal tone.” Each small tweak brings the output closer to what you need.
This back-and-forth is a form of guided practice for better prompts. Each round teaches you how the model responds to different phrasings. Over time, you learn which words and structures produce the best results for your specific tasks.
Step 8: Watch the Context Window
The context window is the amount of text the model can see at once. If your prompt is very long, or if you include a lot of conversation history, the model might lose track of the beginning. This is especially relevant when you are working on a complex task that spans multiple exchanges.
To avoid this, keep the prompt concise. If you need to refer to earlier information, restate it briefly. For instance, instead of relying on the model to remember a previous instruction, say: “As I mentioned earlier, the target audience is young professionals. Now write a product description for a travel backpack.” This refreshes the context.
What to watch: If the model starts repeating itself or ignoring earlier instructions, you might be hitting the context limit. Shorten the prompt or start a fresh conversation.
Putting It All Together
Clear prompting is a skill you build with practice. Start with a clear goal, structure the prompt, use a role when helpful, give examples, specify the output format, set constraints, and refine based on results. This entire process is what a good prompt engineering course teaches, and it applies whether you are using AI for work, study, or personal projects.
This kind of guided practice for better prompts turns vague requests into precise instructions. The model becomes a more reliable tool, and you spend less time editing its output. Over time, you develop an instinct for what works and what does not.
If you are looking to develop these skills further, United Ceres College at 37 Middle Rd, #05-00, Singapore 188946 offers programmes that include practical training in AI and communication. You can reach them at +65 8699 3609 for more information.
United Ceres College is also on Instagram.