Chatbot AI performs more optimally when given clear and well-structured instructions. This guide explains how to design Chatbot AI prompts and instructions that align with operational needs, brand voice, and customer service workflows.
By following this guide, you can ensure the chatbot delivers consistent responses and understands when to hand over conversations to a human Agent.
A. Understanding the Basic Concepts
To ensure a chatbot follows defined rules, Custom Prompts should be treated as a Standard Operating Procedure (SOP) for new employees.
The Knowledge Base serves to provide information or facts (the “what”), while Custom Prompts give guidance on how to behave, how to respond, and how to handle specific scenarios (the “how”).
A well-structured prompt consists of three main elements provided to the AI:
Knowledge Base: Documents containing factual information, such as frequently asked questions (FAQs) and product lists.
Custom Prompt: Configuration of the communication style and interaction guidelines.
-
Human Handover Conditions: Rules that define when a conversation should be handed over to a human Agent.
Learn more about how to configure AI in the Chatbot here.
B. Components of a Well-Structured Prompt
When configuring the AI, you need to define the following components to ensure consistent behavior.
1. Identity
Define who the chatbot is. This helps the AI stay in the correct “role.”
Name: Assign a specific name to the chatbot (for example: MinBos, MekaBot, Usman).
Role: Define the chatbot’s position or function (for example: Customer Service Agent, Sales Representative, Virtual Assistant).
Customer Greeting Style: Clearly define how the chatbot should greet customers to match the brand voice (for example, using “Kakak” instead of “You”, or addressing customers as “Mekarians”).
2. Tone of Voice
Choose a communication style that aligns with the brand identity. Based on common references, some frequently used tones include:
Casual and conversational: Feels light and easy to approach.
Professional and formal: Polite and business-oriented.
Empathetic and friendly: Shows care and a personal connection.
Playful and fun: Entertaining and enjoyable.
3. Operational Goals and Behavior
This section is the core of the AI instructions—the “brain” of its operations. Write it with clear and well-structured logic.
Write Instructions Clearly and Specifically
Do not assume the AI can understand implicit rules. Explain things as if you were teaching a five-year-old child.
- Poor prompt: “Give a discount if the purchase quantity is large.”
- Good prompt: “If a customer purchases 12–59 pcs, apply a 5% discount. If they purchase 100 pcs or more, apply a 15% discount.”
Use a Systematic Format
AI understands structured text more easily than long paragraphs without separation. Use headings, bullet points, and numbering.
Example structure: (heading → subheading → bullet points)
Define ‘If–Then’ Scenarios
Provide clear logic for the AI to follow when handling more complex cases.
Example: “A discount applies IF AND ONLY IF the product is a custom jersey AND the customer pays in full upfront. If the payment uses a Down Payment (DP), then the discount does not apply.”
Provide Examples (Few-Shot Prompting)
To help the AI understand the expected response format, include concrete examples in the prompt.
Format: “If the customer asks X, then respond with Y.”
Set Clear Limitations to Prevent Hallucinations
Explicitly define what the AI is not allowed to do. If a product has commonly used nicknames (for example, ‘Qontak’ for ‘Mekari Qontak’), instruct the AI to recognize them to avoid confusion.
C. Designing Human Handover Rules
You need to define clear rules for when the AI should stop responding and hand over the chat to a human agent. This is crucial for maintaining customer satisfaction.
Some common conditions that should be included:
Complaint Handling: When the customer appears angry, uses all capital letters or offensive language, or reports billing errors.
Sales Closure: When the customer wants to negotiate a special deal or place a large or group order.
Unanswerable Questions: When the AI fails to understand the question after two attempts, or when the information is not available in the Knowledge Base.
Specific Keywords: When the user types “CS”, “Admin”, “Agent”, or “Orang”.
D. Best Practices Summary
Iteration: Prompting is not a one-time task. Continuously test, observe the AI’s responses, and refine the instructions on a regular basis.
Contextualization: If the Knowledge Base contains media (images or videos), include textual descriptions in the prompt or Knowledge Base so the AI can understand the context.
Avoid Ambiguity: Do not leave rules open to interpretation. If discounts apply only to specific products, list them clearly. If a service is available only offline, state it explicitly.
This concludes the explanation of the guide for creating effective Chatbot AI prompts and instructions. Next, explore examples of use cases that demonstrate effective prompt implementation here.