The same AI model can give you a vague, unhelpful response or a sharp, useful one. The difference is often not the model. It is how you asked.
Prompting is the skill of communicating clearly with an AI. You do not need to learn special syntax or magic phrases. You just need to give the model what it needs to aim in the right direction.
Think of it like briefing a contractor
If you hire someone to renovate a room and say “make it look nice,” you will get whatever they think looks nice. If you say “paint the walls light grey, keep the original floorboards, add a shelf at eye level on the north wall,” you get something much closer to what you actually wanted.
AI works the same way. Vague instructions produce generic output. Specific instructions produce useful output.
The four things a good prompt usually includes
1. What you want
State the task directly. Do not hint at it or assume the model will infer it.
| Instead of | Try |
|---|---|
| ”Tell me about emails" | "Write a short professional email declining a meeting invitation" |
| "Summarise this" | "Summarise this article in three bullet points for a non-technical reader” |
2. Who you are or who this is for
Context about the audience shapes the answer significantly.
“Explain neural networks to a 12-year-old” gets a very different response than “Explain neural networks to a software engineer switching into ML.”
Both are valid. Neither is better than the other. The right one depends on who needs the answer.
3. The format you want
Tell the model how to structure its output if it matters.
- “Give me a numbered list”
- “Write this as three short paragraphs”
- “Use a table with two columns”
- “Keep it under 100 words”
4. Constraints or things to avoid
If there are things you do not want, say so.
- “Do not use jargon”
- “Avoid bullet points, write in prose”
- “Do not suggest paid tools”
Show an example when you can
If you want a specific style or format, show one. AI models are good at picking up patterns from examples.
“Write a product description in this style: [paste an example]”
This is often more effective than a long description of the style in words.
Iterate, do not restart
If the first answer is not quite right, do not start over. Tell the model what to change.
- “Make it shorter”
- “The tone is too formal. Make it friendlier”
- “The second paragraph is off-topic. Remove it and expand the third”
The conversation is your editor relationship. Use it.
What not to worry about
You do not need to say “please” to get better results (though it does not hurt). You do not need to use specific keywords or incantations. You do not need to be a developer or know anything about how the model works.
Clear, specific, direct language works best. Write the prompt you would write to a smart capable person who does not know you, does not know your project, and needs to understand what you want from scratch.
Better prompts come from being specific about what you want, who it is for, how you want it formatted, and what to avoid. Show examples when style matters. Treat the conversation as an iterative process and ask for changes rather than starting over. You do not need special syntax, just clear instructions.