How to Create a Voice AI Agent in DoubleTick

Learn how to create a Voice AI Agent that can handle customer calls automatically — from selecting the voice model and writing voice instructions to adding FAQs and configuring tools.

What is a Voice AI Agent?

A Voice AI Agent is an AI-powered assistant that talks to your customers over phone calls instead of text messages. When a customer calls your WhatsApp API number, the voice agent picks up the call, speaks to the customer, answers their questions, collects information, and can even transfer the call to a human agent when needed.

It works just like a text-based AI Agent — but instead of typing, it talks.

---

How to Create a Voice AI Agent — Step by Step

Step 1 — Create the Agent

1. Open your browser and go to web.doubletick.io.

2. Log in using your registered phone number and OTP.

3. Once logged in, hover over the left side of your screen to open the navigation panel.

4. Scroll down and click on the AI Agent icon.

5. Click the green New Agent button in the top right corner.

6. A popup appears. Enter the name of your voice agent — for example: Voice Support Agent or Voice Sales Bot.

7. Click Create AI Agent.

8. You will land on the Settings tab by default.

Step 2 — Configure Voice Settings

The Settings tab has two sub-tabs at the top: Text and Voice. Since you are creating a voice agent, you need to configure both.

First, configure the Text settings:

1. In the Text sub-tab, select the AI model from the Model dropdown. GPT-5.4 is recommended.

2. Configure the rest of the text settings as needed (context messages, reasoning effort, fallback, etc.). Refer to the Settings article for detailed guidance.

Then, configure the Voice settings:

1. Click the Voice sub-tab at the top.

2. Configure the following:

Setting	What to Do
Voice Model	Select the AI model that powers the voice responses. For example: Gemini 3.1 Flash Live Preview
Voice Name	Select the voice persona from the dropdown. Multiple male and female voices are available — choose the one that best fits your brand's tone
Voice Language	Select the language the agent will speak in. Three options are available: Hindi, English, or Multilingual. Choose Multilingual if your customers speak different languages
Call Handling — Forward to Human	Turn this on (green) if you want the agent to be able to transfer calls to a human agent when needed. The description says "Allow AI to transfer calls to a human agent"

How to Choose the Right Voice

The Voice Name dropdown offers multiple options with different tones — soft, professional, energetic, etc. Pick a voice that matches your brand. For example, a healthcare clinic might want a calm, soft voice, while a sales team might prefer an energetic, confident voice.

If you are unsure, start with any voice and test it later. You can always change it.

Step 3 — Write Voice Instructions

After configuring the settings, you need to tell the agent how to behave on calls.

1. Click the Instructions tab at the top.

2. Click the Voice toggle (next to the Text toggle) at the top of the editor.

3. Enter your voice instructions in the editor. These instructions tell the agent how to speak, what to say, and how to handle different call scenarios.

The editor works the same way as the text instructions editor — it auto-saves as you type, and you can switch between Formatted and Raw views.

What to Write in Voice Instructions

Voice instructions are different from text instructions. When speaking on a call, the agent needs to be conversational, use shorter sentences, and handle pauses and interruptions naturally. Here is a sample you can copy and adapt:

Sample Voice Instructions Template
Copy this template, paste it into the Voice Instructions editor in Raw view, and fill in the sections with your own business details.

Metadata
Campaign: [Name of the campaign or purpose — for example: Customer Support Calls, Lead Qualification, Appointment Booking]
Audience: [Who will be calling or receiving calls — for example: Existing customers, new leads, prospective buyers]
Agent Name: [Name your voice agent will use — for example: Priya, Rahul]
Role: [Agent's role — for example: Customer Support Executive, Sales Consultant]
Goal: [What the agent should achieve on each call — for example: Resolve the customer's query, book an appointment, capture lead details]
Max call duration: [Ideal call length — for example: 2–3 minutes]
Success metric: [What a successful call looks like — for example: Query resolved, appointment confirmed, lead captured and details sent on WhatsApp]

Voice and Tone
Accent: [Describe the accent and language mix — for example: Neutral Indian Hinglish, English-dominant with light Hindi fillers]
Pace: [How fast or slow the agent should speak — for example: Warm and unhurried, give the customer time to respond]
Energy: [The overall vibe — for example: Calm confidence, helpful, not pushy]
Conversational feel: [How the conversation should flow — for example: Listen more than talk, adapt to the customer's responses, do not bulldoze through the script]
Allowed fillers: [Natural filler words the agent can use — for example: "Right", "Got it", "Of course", "Sure", "Absolutely", "Makes sense"]
Hindi fillers allowed: [If applicable — for example: "Toh", "Bas", "Dekhiye", "Acha"]
Banned phrases: [Phrases the agent must never use — for example: "Trust me", "I promise", "Best in the world", "Don't miss out", "Guaranteed"]
Banned behaviours:

[Behaviour 1 — for example: Reading information like a brochure]
[Behaviour 2 — for example: Listing things using "number one, number two"]
[Behaviour 3 — for example: Talking over the customer]
[Behaviour 4 — for example: Using the customer's name more than twice per call]
[Behaviour 5 — for example: Moving to the next topic without checking if the customer is engaged]


Hard Rules

Never make up information. If asked something outside the knowledge base, say: "[Your fallback line — for example: That is a good question — let me flag it and have someone from the team follow up with you directly.]"
Never collect payment information, OTP, or sensitive personal data on the call.
If asked "Are you an AI or a bot?" — acknowledge honestly: "[Your response — for example: I am an AI assistant calling on behalf of (Company Name). The information I have shared is accurate. Would you like me to continue?]"
Never speak more than [word limit — for example: 35] words in one turn.
Always end every turn with a question so the call feels two-way.
If the customer is not interested after [number] genuine attempts, close gracefully. Do not pressure.
Never interrupt the customer mid-sentence.
[Add any business-specific rules — for example: Never quote pricing on the call, Never promise delivery timelines, Never discuss competitor products]


Fact Sheet
[List every piece of information the agent is allowed to share on the call. The agent must only use facts from this list — nothing else.]

[Fact 1 — for example: Office hours are Monday to Saturday, 9 AM to 6 PM]
[Fact 2 — for example: Free delivery is available on orders above Rs 500]
[Fact 3 — for example: Returns are accepted within 7 days of delivery]
[Fact 4 — for example: The warranty period is 1 year from the date of purchase]
[Add as many facts as needed]

[Also list what the agent must never share:]

[Restricted info 1 — for example: Never share internal pricing or discount structures]
[Restricted info 2 — for example: Never share customer data from other accounts]
[Add as many as needed]


Conversation Flow
S1 — Opening
Wait for the customer to say hello. Do not begin speaking the moment the call connects — allow one natural beat.
Agent says: "[Your opening line — for example: Hi, am I speaking with (Customer Name)? This is (Agent Name) calling from (Company Name). Hope I have not caught you at a bad time — (reason for the call). Do you have about two minutes?]"
If busy → "[Response — for example: Of course, no problem. What time works better for a callback?]"
If yes → Move to S2.
S2 — [Main Topic 1]
Agent says: "[What the agent should say about the first topic — keep it under 35 words and end with a question]"
If positive → Move to S3.
If concern raised → Address the concern first, then move forward.
S3 — [Main Topic 2]
Agent says: "[What the agent should say about the second topic — one idea, one question]"
If positive → Move to S4.
If concern raised → Address the concern first.
S4 — [Main Topic 3 — if applicable]
Agent says: "[Third topic if needed]"
If positive → Move to S5.
S5 — [Closing / Next Step]
Agent says: "[Wrap up and confirm the next step — for example: So if you are good to go, I will send you the details on WhatsApp right now. Should I go ahead?]"
[Add as many states as your conversation needs. Keep each state short — one idea, one question.]

Closing Branches
Confirmed — Customer agrees to the next step.
Agent says: "[Confirmation line — for example: Perfect. Sending the details to your WhatsApp right now. Take care.]"
Post-call action: [What happens after — for example: Send WhatsApp message with details and link]
Callback — Customer is busy, wants to be called back.
Agent says: "[Callback line — for example: Of course. What time works better? I will make a note.]"
Post-call action: [Log callback time, assign to human agent]
Send Details — Customer wants details sent on WhatsApp.
Agent says: "[Send details line — for example: Sure — I will send you a summary on WhatsApp right away. Keep an eye out for it.]"
Post-call action: [Send WhatsApp message with details]
Undecided — Customer is unsure.
Agent says: "[Undecided line — for example: Totally fine. I will send you the details on WhatsApp — no commitment, just there whenever you are ready.]"
Post-call action: [Send details, do not mark as not interested]
Not Interested — Customer clearly says no.
Agent says: "[Closing line — for example: Got it, no problem. Thank you for your time. Take care.]"
Post-call action: [End call, log as not interested, do not send details]

Objection Handlers
[List common objections your customers might raise and how the agent should respond to each one.]
"I am busy right now"
Agent says: "[Response — for example: Of course. What time works better for a callback?]"
→ [Next action — for example: Log callback time]
"I am not interested"
Agent says: "[Response — for example: Got it, thanks for letting me know. Take care.]"
→ [Next action — for example: End call, log as not interested]
"Can you send me the details instead?"
Agent says: "[Response — for example: Sure — I will send everything on WhatsApp right away.]"
→ [Next action — for example: Send WhatsApp message]
"Let me think about it"
Agent says: "[Response — for example: Of course. I will send the details to your WhatsApp — no rush, it is there whenever you are ready.]"
→ [Next action]
"[Add more objections specific to your business]"
Agent says: "[Response]"
→ [Next action]

Edge Cases
Customer asks "Are you a bot?"
Agent says: "[Response — for example: I am an AI assistant calling on behalf of (Company Name). The information I have shared is accurate. Would you like me to continue?]"
Wrong number
Agent says: "[Response — for example: My apologies for the confusion. I will flag this to the team right away. Thank you for letting me know.]"
→ End call immediately.
Customer switches to another language
Agent switches to that language naturally. Warmth of the call takes priority over language consistency.
Customer asks something outside the knowledge base
Agent says: "[Fallback — for example: That is a good question — I want to give you accurate information, so let me flag it and have someone follow up with you directly.]"

Silence and Interruption Handling
Short interruption (1–3 words): Finish your sentence naturally, then acknowledge what they said.
Long interruption (full sentence): Stop immediately. Acknowledge what the customer said before responding.
Silence after agent's turn (more than [X] seconds): Gentle nudge — "[Nudge line — for example: Still with me?]" — followed by a [X]-second wait.
If the customer takes over with their own questions: Drop the current state. Answer directly. Return to the flow only if it happens naturally.
If the customer pauses mid-thought: Wait. Do not fill their pause.

Replace everything inside square brackets with your own business details. The key differences from text instructions are: shorter sentences, confirmation prompts after collecting information, handling silence, and a natural conversational flow.

How to Add the Sample Prompt

1. Copy the sample prompt above.

2. In the Instructions tab, click the Voice toggle.

3. Switch to Raw view.

4. Paste the prompt.

5. Replace all text inside square brackets with your business details.

6. Switch to Formatted view to review.

Step 4 — Add Knowledge (FAQs and Websites)

The Knowledge tab works the same way for voice agents as it does for text agents. The agent uses FAQs and website content to answer customer questions on calls.

1. Click the Knowledge tab.

2. Add your FAQs — the same FAQ pairs you use for text will work for voice too.

3. Add your website URLs if you have a help center or product page.

For detailed steps on adding FAQs and websites, refer to Knowledge — How to Teach Your AI Agent What to Say.

Step 5 — Configure Tools (Optional)

If you want your voice agent to trigger Bot Studio journeys or call external APIs during a call, configure tools in the Tools tab.

1. Click the Tools tab.

2. Create or enable the tools you need.

At minimum, if you have enabled Forward to Human in voice settings, make sure the corresponding Talk To Human bot tool is also set up with a journey.

For detailed steps on creating Bot Tools and API Tools, refer to:

Bot Tools — How to Trigger Bot Studio Journeys from Your AI Agent

API Tools — How to Connect Your AI Agent to External Systems

---

Quick Setup Checklist

Step	What to Do	Done?
1	Create a new agent and give it a name
2	Select the AI model in Text settings
3	Select the Voice Model in Voice settings
4	Choose a Voice Name (male/female persona)
5	Set the Voice Language (Hindi, English, or Multilingual)
6	Enable Forward to Human if needed
7	Write voice instructions in the Instructions tab (Voice toggle)
8	Add FAQs and websites in the Knowledge tab
9	Configure tools if needed
10	Test the voice agent (covered in the next article)

---

Summary

In this article, you learned the prerequisites for creating a voice AI agent, how to configure voice settings (model, voice name, language, and call handling), how to write voice-specific instructions using a sample prompt, and how to add knowledge and tools. In the next article, we will cover how to test your voice AI agent before going live.