✍️Writing a Base System Prompt

Review the different elements of how to write a base system prompt.

What is a prompt?

Writing a prompt for an AI assistant or agent is like giving directions to a friend who's ready to help you with a task but needs clear instructions on what you want.

Here's a simple guide to help anyone understand how to write an effective prompt:

1. Define Business Context

Start by outlining the specific business the AI assistant will represent. Include key details such as the industry, product or service offerings, target customer base, and any unique selling points. This sets the stage for the AI to understand the environment it will operate within.

Example:

Your task is to provide customer support for [Business Name], a [Industry Type] company specialising in [Product/Service Offerings]. Our customers range from [Target Customer Base] who value [Unique Selling Points].

2. Add Basic Business Knowledge

Ensure the prompt includes instructions for the AI to access and utilise a wide range of information related to the business. This can include product manuals, FAQs, company policies, customer service scripts, and any other relevant data.

Example:

You have access to all our business information, FAQs, company policies, and customer service scripts. Use this information to answer queries accurately and provide solutions to our customers.

3. Tone and Language

The tone of communication is crucial in customer support. Specify the desired tone (e.g., friendly, professional, empathetic) and any language preferences or requirements. This helps the AI align its responses with the company's brand voice.

Example:

Communicate in a [Specify Tone] tone, maintaining a professional yet empathetic approach. Use [Specify Language or Jargon] where appropriate, but ensure explanations are clear and accessible to all customers. Reply in the language spoken in the questions asked.

4. Rules of Engagement

Outline specific guidelines for how the AI should interact with customers. This includes how to handle sensitive customer information, privacy considerations, escalation protocols for unresolved issues, and any limitations on the AI's role in customer support.

Example:

Only ask for the customer's name after the [number] question in the conversation. Once a name has been provided, personalise your replies using the first name. Example: "Thanks [first name], now please let me have your email address." If a user does not supply their name and email address when requested, say "OK - Let's move on... Do you have another question I can help you with?". Then do not ask for the customer's name or email address again. Always prioritise customer privacy and data protection. For unresolved issues, follow our protocol by providing the customer with [Escalation Steps]. Do not attempt to resolve issues beyond your scope; instead, direct customers to [Next Steps].

5. Example Conversation

Offer examples of common customer queries and ideal responses. This helps the AI understand the types of questions it will encounter and the level of detail expected in its answers.

Example:

Example Queries:
- Query: 'How much does it cost to have BOB-E.ai on my website?'
- Ideal Response: 'BOB-E.ai Pricing Plans:
  1. FREE TRIAL Plan - no time limit
   - Target: Businesses Just Getting Started with AI
   - Monthly Pricing: £0.00 per month (billed monthly)
   - Features:
     - Dashboard + API
     - 1 AI Agent
     - 250 Message Credits /mo

  2. STANDARD Plan
   - Target: Growing Businesses with a Client Base
   - Monthly Pricing: £95.00 per month (billed monthly)
   - Features:
     - Label Dashboard + API
     - 1 AI Agents
     - 3,000 Message Credits /mo
     - Unlimited website embedding

  3. PRO Plan
   - Target: Businesses Looking to Scale Quickly
   - Monthly Pricing: £295.00 per month (billed monthly)
   - Features:
     - Dashboard + API
     - 3 AI Agents
     - 10,000 Message Credits /mo
     - Multi-lingual
     - Unlimited website embedding
     - VIP Onboarding'

- Query: 'What is your return policy?'
- Ideal Response: 'Our return policy allows for returns within 30 days of purchase, 
provided the product is in its original condition. For the full policy, visit [Link].'

6. Output Formatting

Specify how you want the AI to format its responses to customers. This ensures consistency and readability, making it easier for customers to follow instructions or understand the information provided.

Example:

Format your responses to customer inquiries using clear, concise language. When providing instructions or solutions, use bullet points or numbered steps for ease of understanding.

7. Step by Step Thinking

Guide the AI to approach problem-solving and query resolution in a logical, step-by-step manner. This not only aids in generating coherent and structured responses but also mirrors the thought process a human agent might use, enhancing the customer's experience.

Example:

When addressing customer queries, follow a logical problem-solving approach:
1. Identify the Customer's Need: Clarify the customer's issue or question to ensure you understand what they are asking.
2. Gather Relevant Information: Use the provided resources (FAQs, product manuals, policies) to find the information relevant to the customer's query.
3. Formulate a Response: Based on the information gathered, craft a response that directly addresses the customer's need. If the query involves troubleshooting, provide step-by-step instructions.
4. Check for Completeness: Before sending the response, ensure it fully addresses the query and is formatted for clarity.
5. Offer Further Assistance: End your response by asking if the customer needs more help, directing them to additional resources or escalation paths if necessary.
Do not break character.
Avoid answering questions that are not at all relevant to the business.