A prompt is the set of instructions that guides the model. The clearer and more specific it is, the more accurate and less random the replies. This five-part framework works in any language, and you can apply it right away inside AI Studio in Kwamle Media.
1. Define the role and identity
Start by telling the model who it is. For example: "You are the customer-care assistant of a clothing store. You write in simple, warm, professional language." The role sets the style of every reply that follows.
2. Set the tone and the dialect
Be explicit: formal or casual? Short or detailed? For Arabic, standard or a specific dialect? For example: "Use polite Gulf Arabic, short sentences, and avoid technical terms." Without this instruction the model may drift between styles from one reply to the next.
3. Add clear constraints
Constraints prevent embarrassing mistakes. Some that work well:
- "Never mention a price that is not in the product list."
- "If you do not know the answer, ask the customer to contact support instead of guessing."
- "Keep each reply to three sentences unless the customer asks for details."
4. Give examples
The strongest way to raise quality is to show one or two examples of a question and an ideal reply. The model imitates the pattern it sees, so the output lands much closer to what you expect.
Customer: Do you deliver to Dubai?
Ideal reply: Yes, we deliver across Dubai within 24 hours. Shall I prepare the order for you?
5. Ask for a specific output format
When you need structured information, such as collecting an order, ask for a clear shape: "Return the reply as: product name, quantity, total price." That makes it easy to connect the reply to the next steps of the bot.
A template to copy
You are [role] for [your business].
Write in [dialect / tone], with [short / detailed] sentences.
Rules: [constraint 1], [constraint 2], [constraint 3].
If you do not know the answer: [fallback behaviour].
Example. Question: "..." Reply: "..."Test the prompt on real questions from your customers and adjust it gradually. Small, steady improvements make a bot that feels like your best employee.
