Updated August 2026. Applies to ChatGPT, Gemini, Claude and every current AI assistant.
Prompt writing is the one AI skill that transfers to every tool and survives every model update — which is why it is the first real module in any serious AI course. This prompt engineering guide covers the structure that works, in five parts you can apply today.
The five-part prompt
Most people write one line and get a generic answer. A working prompt has five components:
| Part | What it does | Example |
|---|---|---|
| Role | Sets expertise and perspective | “You are a Nepali tax consultant…” |
| Task | The specific thing to do | “…explain the process for…” |
| Context | Everything the AI cannot know | “I run a small mobile shop in Pokhara, annual turnover…” |
| Format | How the answer should look | “…as a numbered checklist, under 200 words” |
| Constraints | What to avoid or include | “Use simple Nepali. Do not state fees — I will verify those.” |
Combine them and the answer transforms. The single biggest improvement most people can make is adding context — AI cannot read your situation, and vague inputs produce vague outputs.
The five mistakes that produce bad answers
1. No context. “Write a business plan” produces a generic template. “Write a business plan for a mobile accessories shop in Pokhara with NPR 5 lakh capital targeting students” produces something usable.
2. Asking for everything at once. Break complex tasks into steps. Ask for an outline, approve it, then ask for each section.
3. Accepting the first answer. The first output is a draft. Say “make it shorter”, “more specific”, “less formal” — iteration is where quality comes from.
4. Not showing an example. Pasting two sentences of the style you want beats any number of adjectives. This is the single most effective technique in the entire discipline.
5. Trusting facts. AI fabricates confidently. Use it for language and structure; verify every fact, especially Nepal-specific ones.
Nepali-specific prompt techniques
Force natural Nepali. The default output reads like a government notice. Fix it: “सजिलो बोलचालको भाषामा लेख्नुस्” or “write in simple spoken Nepali, like explaining to a friend”. Our Nepali content guide covers this in depth.
Type in Nepanglish. “Malai euta application lekhna help garnus” works perfectly — no need to switch keyboards.
Name your audience. “Explain to my aama who has never used a bank app” produces genuinely different output than “explain simply”.
Block fabrication explicitly. “If you are not certain about a Nepali government fee or rule, say so instead of guessing.” This instruction measurably reduces confident errors.
Prompts you can copy today
For a Facebook post:
You are a Nepali social media writer. Write a Facebook post for my [shop type] in [city] announcing [thing]. Simple spoken Nepali, 3 lines, one clear call to action. No hashtag spam.
For understanding a document:
Explain this document in simple Nepali as if to someone with no legal background. List what I must do, in order. If anything is ambiguous, say so rather than guessing.
For studying:
You are a patient tutor. Explain [topic] to a Class 12 student in Nepal using a local example. Then ask me three questions to check I understood. Do not give the answers until I try.
For writing:
Here are two sentences in my style: [paste]. Match this style and write [task]. Keep it under [X] words.
Frequently asked questions
What is prompt engineering? Writing inputs that get useful outputs from AI. It is a practical skill, not a technical one — no coding involved.
Do I need to learn prompting for every AI tool? No — the five-part structure works across ChatGPT, Gemini, Claude and others. That transferability is why it is worth learning properly.
Does prompting work in Nepali? Yes. Type in Nepali or Nepanglish; the same structure applies. See our ChatGPT in Nepali guide.
How long should a prompt be? As long as needed to include role, task, context, format and constraints — typically three to six sentences. Longer is usually better than shorter.
Where can I learn this properly? Structured training covers prompting as a full module before any tool is introduced, because it transfers to all of them — see the AI training page.
The bottom line
A working prompt has five parts: role, task, context, format, constraints. Add context, show an example of the style you want, iterate rather than restart, and verify every fact. Master that and every AI tool you ever use gets better — which is exactly why prompting is taught before tools, not after.