Designing an effective prompt is key to making your AI-powered VoiceAgent sound natural, stay on-brand, and handle different situations smoothly. This guide walks you through the structure of a great prompt and provides best practices for creating one that works.
What Makes a Good Prompt?
Each prompt should give the AI enough context to:
Understand its role (e.g., support agent, sales rep, survey bot)
Speak in a way that matches your company tone
Follow a specific flow and respond intelligently to unexpected inputs
A good prompt is clear, structured, and tailored to the type of interaction you want to automate.
Prompt Structure: What to Include
A good VoiceAgent prompt includes four essential parts:
1. Identity Section
The identity section tells the VoiceAgent who it is, what it’s doing, and how it should behave. This keeps conversations on-brand and ensures the tone is consistent.
Why it matters: Without identity, the AI may default to a generic tone or take liberties with its role. Setting a role helps it stay focused and appropriate.
Best Practices
Define who the agent is (e.g., a support assistant, sales rep, onboarding guide).
Mention the goal of the call (e.g., helping with an integration, qualifying a lead, booking a demo).
Example:
“You are a helpful and friendly sales agent assisting customers with product questions."
Why it matters: Giving your VoiceAgent a clear role helps it stay in character and maintain a consistent tone.
Tip: Be specific about the VoiceAgent’s purpose and attitude - e.g., sales assistant, onboarding guide, or support agent.
2. Style Guardrails
Style guardrails define the "personality" of your VoiceAgent and help it sound consistent, human, and on-brand.
Common Style Guardrail Instructions
Instruction | What It Means | Example Behavior |
Be concise | Avoid long-winded answers. Stick to the point. | ✅ “We integrate with HubSpot and Salesforce.” |
Sound conversational | Use natural, friendly language. No jargon. | ✅ “Sure! That’s a great question.” |
Use varied language | Avoid repeating the same phrases. Paraphrase when needed. | ✅ “Let me check that for you… Okay, I found it!” |
Be empathetic | Acknowledge frustration or confusion in a caring way. | ✅ “Totally understand - it can be frustrating when that happens." |
Ask one question at a time | Don’t overwhelm the user. Keep questions simple. | ✅ “Which CRM do you use?” |
Use casual time references | Refer to time like a human would. | ✅ “Would next Friday work?” |
Don’t repeat exact phrases | Avoid robotic repetition. | ✅ Vary greetings like: “Hi there!” / “Hey, how’s it going?” / “Good to speak with you today!” |
Tone Examples Based on Use Case
Use Case | Suggested Tone | Example Language |
Sales follow-up | Friendly, upbeat, confident | “Hey {{name}}, I’m glad we got connected! Mind if I ask a couple of quick questions to understand your needs?” |
Support call | Calm, reassuring, helpful | “Let’s see what’s going on - I’ll do my best to help sort it out with you.” |
Survey/CSAT bot | Polite, neutral, efficient | “Thanks for your time! On a scale from 1 to 5, how satisfied were you with our service today?” |
Demo booking agent | Warm, helpful, persuasive | “Great! Sounds like [Your Company] could be a good fit. Can I book a quick demo with one of our specialists for you?” |
3. Response Guidelines
These guidelines help the VoiceAgent stay in character and handle edge cases smoothly.
Why it matters: Calls don’t always go as planned. These rules teach the agent how to deal with confusion, interruptions, and unclear inputs.
Best Practices
Define how the agent should react to unclear, partial, or awkward responses.
Avoid overly technical or robotic fallback replies.
Keep it human - even when things don’t go as expected.
Example Instructions
Instruction | Example |
If the user only partially answers, politely ask for clarification. | “Just to make sure I understand…” |
If the user asks about AI, express excitement. | “Yes, I’m powered by AI! Pretty cool, right?” |
If interrupted, pause and wait for them to finish. | Let the user speak, then respond naturally. |
Don’t say "transcription error" | Instead: “Sorry, I didn’t catch that -could you repeat it?" |
4. Task Flow & Branching Logic
This section defines the structure of the conversation - the questions to ask, possible answers, and what to do next based on user responses.
Why it matters: Without branching, the agent can sound repetitive or jump around. Clear logic makes the conversation smoother and more natural.
Best Practices
Use "If X, then Y" logic for decision-making.
Keep each step focused on a single goal.
Set clear exit conditions (e.g., when to end the call or transfer to a human).
Sample Call Flow
Greet the user:
→ “Hi {{name}}, this is Emma from [Your Company]. You were interested in a strategy call - is that right?”If yes → continue
If no → politely end callAsk about business goals:
→ “What motivated you to look into a new phone system?”Ask about team size
→ If team size ≤ 2, thank and close the call
→ If team size ≥ 3, offer a demoAsk for preferred date/time
→ Confirm and wrap up
Example Use Case: Support Scenario Prompt
If you’re designing a VoiceAgent to simulate or handle basic support conversations, your prompt should prepare the agent to guide users through common issues with clarity and empathy. Below is an example prompt structure and sample scenarios that can help shape your own use case.
Prompt Example (Support Agent)
You are a friendly and helpful technical support assistant working for a SaaS company. Your goal is to help customers resolve issues with their virtual phone system. If needed, escalate to the support team and guide the customer on what information to provide.
Your Agent Should:
Be non-technical but solution-oriented
Ask clarifying questions if the user is vague
Stay calm and constructive, even if the user is frustrated
Avoid robotic language or deflection
Wrap up with next steps or escalation instructions if needed
Scenario: Call Routing & API Use
Customer Intent:
Wants to route incoming calls based on booking number
Unsure if an API is required
Asks about handling VIP callers
VoiceAgent should ask or confirm:
"Are you currently using any booking system or CRM?"
"Do you want to route calls based on specific tags or contact info?"
"Would you like me to check if this requires custom API setup?"
If escalation is needed:
"To help you faster, I recommend sharing your booking workflow with our support team. Would you like me to guide you through that?"
Scenario: Follow-Up Automations
Customer Intent:
Wants to send automatic SMS after calls
Asks about Slack alerts or CRM integrations
VoiceAgent should clarify:
"Do you want to trigger the SMS for all calls or just missed ones?"
"Which CRM are you using? I can check if call notes can be synced automatically."
"Would you like help setting up workflow rules?"
Scenario: Outbound Calls Are Failing
Customer Intent:
Can't place outbound calls
Unsure whether it’s a network, configuration, or account issue
VoiceAgent should ask:
"Is the number showing any error when you try to call?"
"Has anything recently changed in your setup or internet connection?"
"I can help check your number configuration - do you know if it's linked to a call flow?"
Example Pushback Lines
Sometimes customers express confusion or frustration. Your VoiceAgent can handle these naturally with phrases like:
"No problem, let me simplify that."
"Thanks for your patience - here’s what we can do."
"If that doesn’t work, I can help you escalate this."
"Let’s double-check that together."
When writing support-style prompts:
Simulate a real user with a real problem
Use phrasing that’s realistic and empathetic
Include fallback options and escalation pathways
If you need further assistance or have any questions, you can contact our Support team. We are always here to help you!