Enhancing AI Conversations with Effective Prompting

Properly fine-tuned prompts help AI agents provide clearer, more engaging responses, ensuring:

✔ Better customer interactions with structured conversations.
✔ Consistent messaging in both inbound and outbound calls.
✔ Improved AI adaptability to objections and follow-ups.
✔ More natural, brand-aligned communication.

This guide walks you through how to analyze, edit, and test AI prompts for improved response quality.

Step 1: Analyze Current Prompt Performance

Before adjusting prompts, it's essential to evaluate how the AI currently performs.

🔹 How to Assess AI Responses

Run test calls in the Test & Train panel to review responses.
Check if the AI follows the script correctly or deviates.
Identify areas where the AI misunderstands questions or provides unclear responses.
Note objections or dropped calls where the AI struggles.

💡 Tip: If the AI sounds robotic or lacks engagement, adjust its responses in Freestyle Prompting for a more natural conversation flow.

Step 2: Edit Prompt Content with Clearer Instructions

🔹 Example Prompt for a Business

Business: Horizon Travel Agency – AI agent handling inbound travel inquiries and booking assistance.

✅ Freestyle Prompting Example for Horizon Travel AI Agent:

"Speak with a warm and professional tone, like a knowledgeable travel concierge. Be helpful and enthusiastic while answering customer inquiries.

When the customer asks about destinations, provide recommendations based on seasonality and budget.

If they ask about pricing, confirm whether they have preferred dates and suggest a few options.

If the customer is hesitant, reassure them about our flexible cancellation policies. If they decline, offer an exclusive discount for first-time bookings.

Ensure that before ending the call, you ask: 'Would you like me to send the itinerary details via email or SMS?'"

💡 Tip: This structured prompt helps the AI guide the conversation smoothly while ensuring it asks key questions to move the customer closer to a booking.

Step 3: Save Changes & Evaluate Improvements

Once prompt adjustments are made, test them to ensure better AI response quality.

🔹 How to Test & Refine AI Prompts

Click Save Changes after updating the prompts.
Run a test call simulation with different customer scenarios.

Evaluate if the AI:
- Stays on script while allowing natural flexibility.
- Handles objections properly without dropping the conversation.
- Uses the correct tone and approach for the customer type.
Make further refinements if needed.

💡 Tip: Continuous testing and fine-tuning lead to more effective AI conversations over time.

📌 Basic Prompt Templates for Different Use Cases

1️⃣ Template for Inbound Calls (Customer Support AI)

Scenario: AI handles customer service inquiries for a SaaS company.

✅ Freestyle Prompting Example:

"Speak in a calm and professional tone, like a friendly customer service representative. Always greet the customer warmly and introduce yourself.

If the customer has a technical issue, follow a structured troubleshooting guide.

Ask clarifying questions to gather details about the issue: 'Can you describe what you were trying to do before the issue occurred?'

If the issue is unresolved, offer to escalate to a human agent.

If the customer sounds frustrated, acknowledge their concern: 'I understand how frustrating that can be. Let me assist you in resolving this.'"

💡 Tip: This ensures the AI agent provides clear guidance while maintaining a reassuring presence.

📌 Want to refine AI voice settings? Read our Configuring Voice Options Guide.

2️⃣ Template for Outbound Calls (Sales or Promotional AI)

Scenario: AI calls previous customers with an exclusive renewal offer.

✅ Freestyle Prompting Example:

"Use an engaging yet professional tone, similar to a helpful sales consultant. Be direct but not pushy.

Greet the customer by name and reintroduce the company: 'Hi {{contact_name}}, this is [AI Agent] from [Company]. We’re offering a special loyalty discount for returning customers. Would you be interested in hearing more?'

If the customer asks for details, explain the benefits briefly and clearly.

If they hesitate, offer a limited-time incentive: 'For a limited time, we're giving an extra 10% off renewals made today.'

If they decline, be polite and ask if they’d like a follow-up later: 'No worries! Would you like me to notify you when we have future offers?'

Before ending the call, confirm next steps: 'Would you like me to send the details via email or SMS for reference?'"

💡 Tip: This ensures the AI remains engaging yet respectful, increasing the chances of a successful sale or follow-up.

📌 Want to connect AI calls with scheduling? Read our Zapier Integration Guide.

Important Note: Prompt Token Limits

Awaz.ai uses an advanced LLM framework with a default maximum token limit of 8,192 tokens per AI agent prompt.

This is roughly equivalent to about 7,000 words.
All prompts combined (Freestyle, Instructions, etc.) must stay within this limit.
If the total prompt length exceeds 8,192 tokens, the system will reject it and show an error.

💡 Tip: To avoid errors, keep your prompts concise, clear, and focused. Break complex instructions into smaller, more manageable prompts if necessary.