How to Write Prompts: Core Rules and Real Examples for AI
Two people give an AI the exact same task. One gets a generic draft they have to rewrite entirely. The other gets output they can use almost immediately.
The difference isn’t the tool, luck, or any “tech talent.” The difference is the prompt — how the task was framed.
The good news: writing prompts isn’t programming. It’s an ordinary communication skill you can pick up in an afternoon. This article gives you everything you need: a structure to follow, ready-to-use templates, and a rundown of mistakes to avoid.
What is a prompt, and why does it matter so much?
A prompt is the message you send to an AI — a question, instruction, or task description. (For more terminology, see the AI glossary.)
Why does it matter so much? Because AI can’t read your mind. The way it works, it generates the most probable response to what you actually wrote — not what you meant. When the request is vague, the AI fills in the missing details from averages. And average is exactly what you get.
There’s a direct relationship: the quality of your prompt determines the quality of the output.
The basic structure of a good prompt
The simplest framework — one that works for most tasks — has four parts:
Tell the AI who to be. "You are an experienced copywriter" sets a very different tone than "You are a patient teacher.
What exactly should the AI do — write, summarize, translate, suggest, compare. One verb, one task.
Who the output is for, what it will be used for, and what the AI needs to know (pasted text, situation, goal).
What the result should look like — length, tone, bullet points vs. paragraphs, what to leave out.
In practice, it looks like this:
❌ Weak: “Write some text about our new product.”
✅ Strong: “You are a copywriter for a sporting goods online store. Write a product description for a running jacket aimed at beginner runners — highlight the waterproofing and reflective elements. Maximum 80 words, friendly tone, no superlatives like ‘best on the market’.”
You don’t always need to fill in all four parts. But when the output misses the mark, almost invariably one of them is missing.
Five techniques that take prompts to the next level
Anyone can master the basic structure in five minutes. These techniques are what separate a usable result from an excellent one:
1. Give the AI an example (few-shot prompting)
The single most effective trick. Instead of describing a style, show a sample:
“Write three email subject lines in the same style as this one: ‘Your order is on its way 📦’.”
The AI will replicate the tone, length, and structure of the example far more accurately than if you described the style in words.
2. Ask the AI to think step by step
For more complex tasks (decisions, calculations, analysis), add:
“Work through this step by step and explain your reasoning at each stage.”
The output will be more structured, and it’s easier to spot where the AI may have gone wrong.
3. Say what you don’t want
Constraints work just as well as instructions:
“Avoid technical jargon. Don’t use passive voice. No exclamation marks.”
4. Ask the AI to help improve your prompt
The AI can help with the prompt itself:
“I want you to [task]. What information do you need from me to make the result as good as possible? Ask me before you begin.”
5. Request variations
The first draft rarely lands perfectly. Instead of a single answer, ask for a range:
“Give me three versions: one conservative, one playful, and one completely unconventional.”
Ready-to-use templates for common situations
Copy, fill in the brackets, and send:
You are a professional yet personable email writer. Write an email to [recipient] with the goal of [goal]. Maximum 6 sentences, polite and to the point.
Summarize the following text into 5 bullet points for someone who has 30 seconds. Most important information first. Text: [paste here]
Write a post for [platform] about [topic] for [target audience]. Length up to [X] characters, tone [describe], end with a question for the reader.
Explain [topic] to me as if I'm a complete beginner. Use an analogy from everyday life and finish with a summary of the three most important points.
Compare [option A] and [option B] from the perspective of [criteria]. Create a pros-and-cons table and finish with a recommendation for [situation].
Rewrite the following text to make it [more concise / clearer / more formal]. Keep all the facts. Text: [paste here]
The most common prompting mistakes
Click a card to find out how to fix the mistake.
That last point is more important than it looks: even a perfect prompt can lead to a confident but incorrect answer. Why this happens and when to watch out is covered in How AI works.
How to get better at prompting
Prompting is a skill — and like any skill, it improves with practice, not reading.
A practical routine for your first week:
- Pick one recurring task from your work (emails, summaries, descriptions).
- Write a prompt using the structure Role + Task + Context + Format.
- Iterate — two or three rounds of feedback on the output.
- Save the prompt that worked. Next time, you’ll just tweak it. Over time you’ll build up your own personal template library.
After a few days you’ll find yourself writing good prompts automatically — and that skill will pay off with every AI tool you ever use. You can start experimenting right now in the free AI chat, no sign-up needed.
Try the templates in action
Copy any template from this article into AI Chat and watch the difference compared to a vague request. GuideGlare AI Chat works in your language, right in the browser.
Test yourself: Can you write a good prompt?
Prompt basics
Now that you know how to give good instructions, it’s worth knowing what AI shouldn’t be given — and how to verify its outputs. That’s covered in Safe AI use.