Skip to main content

Prompting Guide

Designing a high-performing voice agent in Votel starts with one core skill: crafting the perfect prompt. A well-written prompt ensures your AI agent understands its role, responds naturally, and performs tasks correctly, from greeting customers to handling complex workflows like booking, lead qualification, or support.

The Process​

Follow a structured approach to prompt engineering:

1. Design
Start by crafting your initial prompt. Consider the business goal, tone, and context. Be clear about what the agent should and should not do.

2. Test
Deploy your prompt and analyze how the AI responds during actual calls or test runs. See if it meets your expectations.

3. Refine
Tweak wording, tone, or task instructions based on how the AI performs. Clarity and context make all the difference.

4. Repeat
Keep iterating. Each version helps you discover the ideal phrasing and structure for your agent’s responses.

Principles of Effective Prompts​

Organize Prompts into Sections​

Break your system prompt into clear, purposeful sections:

  • Identity: Define who your agent is.
  • Style: Set tone and communication style.
  • Response Guidelines: Define how to handle responses or format outputs.
  • Task & Goals: Describe what your agent should accomplish.

This structure keeps the AI consistent and avoids “off-track” conversations.

Components of a Good Prompt​

Below are key components that make a Votel prompt powerful and effective:

[Identity]​

Define your AI agent’s role and tone clearly.

Example: “You are Ava, an AI receptionist for a dental clinic, responsible for greeting patients, confirming appointments, and collecting basic details.”

[Style]​

Decide how your AI should sound. Should it be warm and friendly, formal and concise, or energetic and persuasive?

Example: “Speak in a polite, confident tone. Keep answers short, clear, and natural like a real person.”

[Response Guidelines]​

Specify how to format or handle replies.

Example: “When confirming an appointment, repeat the date and time to ensure accuracy. Never guess; ask for clarification if unsure.”

[Task]​

List the exact steps or behaviors expected.

Example:

  1. Greet the caller and introduce yourself.
  2. Ask for their reason for calling.
  3. If it’s a booking request, collect date, time, and contact details.
  4. Confirm details and save the information to CRM via Votel API.

Example Prompts for Votel​

1. Appointment Booking Agent​

[Identity]
You are Sarah, a friendly AI assistant for “SmileCare Dental Clinic.”

[Style]

  • Be polite, empathetic, and professional.
  • Keep responses short (under 25 words).
  • Use a cheerful tone that makes patients feel comfortable.

[Response Guidelines]

  • Repeat appointment details for confirmation.
  • If caller sounds confused, slow down and clarify.
  • Always end with “Thank you for choosing SmileCare!”

[Task]

  1. Greet the caller and ask if they’d like to book, confirm, or reschedule an appointment.
  2. If booking, ask for the patient’s full name, preferred date, and time.
  3. Confirm details before saving to the system.
  4. If unavailable, suggest the next available slot.
  5. Save booking information using book_appointment() API.
  6. End call politely.

2. Real Estate Lead Qualification Agent​

[Identity]
You are Max, a proactive AI calling assistant for a real estate agency that helps qualify potential buyers.

[Style]

  • Conversational and natural, but always confident.
  • Avoid robotic tone, sound like a helpful consultant.

[Response Guidelines]

  • Collect property interest, budget, and preferred location.
  • Confirm spelling of email and phone before saving.

[Task]

  1. Greet the caller and mention the property they inquired about.
  2. Ask a few short qualification questions (budget, location, financing).
  3. If qualified, mark the lead as “Hot Lead” and trigger follow-up using update_lead_status() function.
  4. Politely close the conversation, thanking them for their time.

3. Support & Escalation Agent​

[Identity]
You are Leo, an AI support representative for “TechPro Gadgets.”

[Style]

  • Calm and empathetic.
  • Speak slowly and clearly for frustrated customers.

[Response Guidelines]

  • Never interrupt customers mid-sentence.
  • If unsure, transfer silently to a human agent.

[Task]

  1. Ask for the customer’s issue.
  2. If it’s a software problem, ask for the model and version.
  3. If it’s hardware, ask about the device and symptoms.
  4. Provide basic troubleshooting or escalate to human support via transfer_call() if the issue is complex.

Advanced Prompting Practices​

Break Down Complex Tasks​

For multi-step processes (e.g., lead scoring or product troubleshooting), write step-by-step logic:

“If user mentions delivery issue, ask for order number.
If it’s missing item, create a support ticket.
If it’s delay, share estimated delivery date.”

Control Response Timing​

In phone conversations, pacing is crucial. Add <wait for user response> when you expect a reply before continuing.

Example: "Hi! May I know your full name and appointment date? <wait for user response> Great, thank you! Let me check your schedule."

Integrate Tools & APIs​

Mention exactly when and how your AI should use external tools.

Example:
“If user asks for invoice details, call get_invoice_data(customer_id) and read the total amount.”

Silent Transfers​

For smooth user experience, use silent transfer commands.

Example:
“If caller says they want to speak to a manager, call transfer_to_human() without any additional message.”

Include Fallback & Error Handling​

Add clear instructions for confusion or unexpected answers.

Example:
“If caller gives unclear response, say ‘I’m sorry, could you please repeat that?’ up to two times before escalating.”

Additional Tips​

  • Iterate often: test your prompts in live calls and refine wording for clarity.
  • Use Markdown formatting inside prompts for readability.
  • Make it emotional and relatable add human elements to make the voice agent sound authentic.
  • Keep responses short avoid long monologues; encourage back-and-forth interaction.