How to write effective prompts: 7 rules every beginner should know
Learn 7 simple rules that will make ChatGPT/Claude answer more accurately and on-target right away.
You’ve used ChatGPT/Claude a few times and felt “okay, but not wow”? The problem is usually NOT the model — it’s how you prompt. This article distills 7 simple rules to take you from “average user” to “top 10% user” in 15 minutes.
Rule 1: Tell the AI its ROLE
The difference:
❌ “Write an article about AI”
✅ “You are a tech editor with 10 years of experience, writing for a general Vietnamese audience aged 25-45. Write an article about AI…”
A persona helps the AI calibrate tone, vocabulary, and depth.
Rule 2: Describe the CONTEXT
❌ “Summarize this meeting”
✅ “This is a transcript of a meeting between the product team (already engaged) and the engineering team (newly joined, no product knowledge yet). Summarize for the eng team to ramp up fast, focusing on the technical decisions that need to be made.”
Same input, different goal → the AI needs to know to format differently.
Rule 3: Give EXAMPLES (few-shot prompting)
Show, don’t tell. When you need output in a specific format:
I need to translate English slogans into Vietnamese in a concise,
rhyming style. Examples:
EN: "Just do it" → VI: "Cứ làm đi"
EN: "Think different" → VI: "Nghĩ khác đi"
Now translate: "Stay hungry, stay foolish"Far more effective than 10 lines explaining “concise, rhyming style”.
Rule 4: Specify the output FORMAT
✅ “Reply in this format:
- H2 heading for each section
- Bullet list, one sentence per bullet
- End with a 50-word TL;DR”
→ The AI follows it exactly. If you don’t say, the format will vary every time.
Rule 5: Set concrete LIMITS
❌ “Keep it brief”
✅ “Maximum 200 words, no sentence longer than 20 words”
The AI understands numbers. “Brief” means something different to everyone.
Rule 6: Ask for STEP-BY-STEP REASONING
For complex questions (math, logic, planning), add:
“Think step by step before reaching a conclusion.”
This is the Chain of Thought technique — research has shown it significantly improves accuracy.
The latest generation of models (Claude 4.7, GPT-5) does this automatically → you don’t need to prompt for it anymore, but it doesn’t hurt.
Rule 7: ITERATE, don’t one-shot it
The first prompt is rarely the best. The standard workflow:
- Write a draft prompt
- Look at the output
- Spot issues: too long / wrong tone / missing point
- Edit the prompt → “Revise the output above: more formal tone, under 150 words”
- Repeat
After 2-3 rounds you’ll have a gold-standard prompt template for that task. Save it for later.
Bonus: Sample prompt template
Save this template, fill it in as needed:
You are a [ROLE + EXPERIENCE].
Task: [GOAL].
Audience: [WHO READS, BACKGROUND].
Format: [MARKDOWN / TABLE / JSON / ...].
Limits: [WORD COUNT / BULLET COUNT / TONE].
Example output:
[1-2 EXAMPLES]
Input:
[CONTENT TO PROCESS]What’s next
- Read more in detail about Prompt Engineering
- See Chain of Thought — a reasoning technique
- Compare tools: Claude, ChatGPT, Gemini