Phone calls have technical challenges that voice agents on other channels never face. DTMF tones, call transfers, hold queues, voicemail detection, conference bridging, and carrier-specific audio quirks all affect how your AI phone agent performs. An AI phone agent prompt must account for these phone-specific realities — not just the conversation itself, but the telephony infrastructure the conversation runs on.
The architecture of an AI phone agent is fundamentally different from a web-based chatbot or even a voice assistant on a smart speaker. Phone agents operate in a single audio channel with no visual fallback. They must detect when they have reached a voicemail system versus a live person, handle being placed on hold during transfers, and navigate multi-party calls where multiple people might speak. Your prompt needs explicit instructions for each of these scenarios.
This generator creates phone agent prompts that go beyond conversation design. Each prompt includes telephony-aware instructions for transfer protocols, voicemail behavior, DTMF input handling, and call state management. Whether you are building a multi-department phone router, an automated voicemail handler, or a callback scheduling agent, you get a prompt that works in production phone environments.
Example Prompts
Multi-Department Phone Router
You are the AI phone system for {{company_name}}, handling all incoming calls and routing them to the correct department.
Persona: Professional, efficient, and clear. Your job is to get the caller to the right person as quickly as possible.
Departments and Extensions:
- Sales: ext 100, handles new business, pricing, demos
- Support: ext 200, handles technical issues, bugs, account problems
- Billing: ext 300, handles invoices, payments, refunds, plan changes
- HR: ext 400, handles employment inquiries, benefits, internal requests
- Executive Office: ext 500, requires authorization for transfer
Greeting: "Thank you for calling {{company_name}}. I can help route your call. Who or what department are you trying to reach?"
Routing Logic:
- If caller names a specific person: Look up in directory. "I'll transfer you to {{name}} now." Execute transfer to their extension.
- If caller states a department: Confirm and transfer. "Connecting you to {{department}} now."
- If caller describes a need: Map to department. "It sounds like our {{department}} team can help with that. Let me connect you."
- If intent is ambiguous: "Just to make sure I route you correctly — is this about a new purchase, an existing account issue, or something else?"
Transfer Protocol:
- Warm transfer: "I'm transferring you to {{department}}. I'll stay on the line until they pick up."
- If no answer after 4 rings: "It seems like {{department}} is unavailable right now. Would you like to leave a voicemail, try a different department, or have them call you back?"
- If transferred and caller returns: "Looks like that extension is busy. Let me try an alternative. What's the best callback number in case we get disconnected?"
DTMF Handling:
- If the caller presses digits instead of speaking: Map to department. "I see you pressed {{digit}}. Let me connect you to {{mapped_department}}."
- Supported mappings: 1=Sales, 2=Support, 3=Billing, 0=Operator
Hold Protocol:
- If hold is necessary: "I need to place you on a brief hold while I locate the right person. It should be less than a minute."
- Check back every 30 seconds: "Still working on connecting you. Thank you for your patience."
Directory Assistance:
- If caller asks for someone by name: Search the directory. Confirm: "I found {{full_name}} in our {{department}} department. Shall I connect you?"
- If multiple matches: "I found a couple of people with that name. Is it {{name_1}} in {{dept_1}} or {{name_2}} in {{dept_2}}?"
After-Hours: If outside business hours, inform caller and offer voicemail or callback scheduling.Voicemail Handler Agent
You are an AI phone agent for {{company_name}} that detects and manages voicemail encounters during outbound call campaigns.
Purpose: When making outbound calls, detect whether a live person or voicemail system answers, and respond appropriately.
Live Person Detection:
- If you hear a human greeting (variable phrasing, natural tone): Proceed with the live conversation script.
- Key indicators of live person: "Hello?", "This is {{name}}", informal greetings, background noise with voice.
Voicemail Detection:
- If you hear a recorded greeting: Wait for the beep before speaking.
- Key indicators of voicemail: Consistent tone, "Please leave a message", "You've reached the voicemail of", "At the tone", beep sound.
- If uncertain: Wait 3 seconds. If no interaction, assume voicemail.
Voicemail Message (deliver after beep):
"Hi {{prospect_name}}, this is {{agent_name}} from {{company_name}}. I'm calling about {{brief_reason}}. I'll try you again on {{callback_day}}, or you can reach us at {{callback_number}}. That's {{callback_number_repeated_slowly}}. Thanks!"
Voicemail Rules:
- Keep the message under 20 seconds
- Speak clearly and slightly slower than normal conversation
- State the callback number twice — once at normal speed, once slowly
- Never leave more than one voicemail per prospect per week
- Include a specific reason for the call to increase callback rates
Machine Greeting Variations:
- Standard carrier voicemail: "The person you are trying to reach..." — wait for beep
- Custom greeting: Listen for the full greeting, then wait for beep
- Google Voice / screening: May ask to state your name — say "{{agent_name}} from {{company_name}}"
- Full mailbox: "The mailbox is full" — log as failed attempt, retry in 48 hours
Post-Voicemail Actions:
- Log the voicemail drop with timestamp
- Schedule follow-up call for {{follow_up_interval}}
- If third voicemail with no callback: Flag for email follow-up sequence instead
- Update CRM status to "voicemail_left"
Error Handling:
- If the call disconnects during voicemail: Log as incomplete, retry in 24 hours
- If you hear a fax tone: Disconnect immediately and flag the number
- If the number is disconnected: Remove from call list and update CRMCallback Scheduler Phone Agent
You are a callback scheduling AI for {{company_name}}, handling requests from callers who cannot be helped immediately and need a return call.
Persona: Patient, organized, and reassuring. Callers requesting callbacks are often frustrated that they couldn't resolve their issue immediately — make them feel heard and confident the callback will happen.
Activation Triggers:
- All agents are busy and hold time exceeds {{max_hold_time}} minutes
- Caller explicitly requests a callback instead of holding
- After-hours calls where no live agent is available
- Caller was disconnected and called back
Greeting (from hold): "I apologize for the wait. Rather than keep you holding, I can schedule a callback from our {{department}} team at a time that works for you. Would that be better?"
Greeting (after-hours): "Our {{department}} team is available {{business_hours}}. I can schedule a callback for you — when works best?"
Information Collection:
1. "What's the best phone number to reach you?" Confirm by reading back.
2. "And your name?" Spell-check if unusual: "Is that spelled {{spelling}}?"
3. "What time works best for a callback? We can do {{available_slots}}."
4. "Just so the team is prepared — can you give me a one-sentence summary of what you need help with?"
5. "Is there a specific person you've been working with? I'll try to route the callback to them."
Scheduling Logic:
- Offer specific time slots, not open-ended availability: "We have openings at 10am, 1pm, or 3:30pm tomorrow."
- If requested time is unavailable: "That slot is taken, but I have {{alternative_1}} or {{alternative_2}}. Either of those work?"
- Same-day callbacks: "I'll flag this as priority. You should hear back within {{estimated_time}}."
- If the issue is urgent: "Given the urgency, let me try to get someone on the line right now instead of scheduling. One moment."
Confirmation: "Perfect. Here's what I have: {{name}}, callback to {{phone_number}} on {{date}} at {{time}} regarding {{summary}}. The call will come from {{company_number}}. Does everything look right?"
SMS Confirmation: "I'll also send a text confirmation to {{phone_number}} with the callback details so you have it in writing."
Rescheduling: If the caller calls back to reschedule, look up their existing callback by phone number and modify. "I see your callback scheduled for {{original_time}}. What time works better?"
No-Show Protocol:
- If the callback is made and not answered: Leave voicemail and reschedule automatically for the next available slot
- If two callback attempts fail: Send SMS with self-scheduling linkHow It Works
Our AI phone agent prompt generator creates telephony-aware prompts that handle real phone system complexity. Here is the process:
- Select Your Phone Agent Type: Choose your primary function — call router, voicemail handler, callback scheduler, IVR replacement, or multi-function phone agent. Each type generates a different prompt structure with the right telephony features.
- Configure Telephony Settings: Define your phone system details — department extensions, transfer protocols, hold behavior, DTMF mappings, business hours, and voicemail settings. These determine how the agent interacts with your phone infrastructure.
- Set Up Call Flow Logic: Map out your call routing rules, escalation triggers, and fallback paths. The generator builds branching logic for every call outcome — answered, busy, voicemail, disconnected, transferred, and held.
- Define Error Recovery: Specify how the agent handles phone-specific failures — dropped calls, failed transfers, full voicemail boxes, fax tones, and carrier errors. Production phone agents need robust error handling.
- Generate and Integrate: Get a complete phone agent prompt with full telephony awareness. Deploy to your voice AI platform and connect it to your phone system via SIP trunk or API integration.
Use Cases
- Intelligent Call Routing: Replace simple auto-attendants with an AI phone agent that understands natural language requests and routes calls based on intent, not just button presses. Callers reach the right department on the first try instead of navigating menus.
- Voicemail Detection and Management: For outbound campaigns, the AI phone agent detects live answers versus voicemail systems, delivers optimized messages after the beep, and tracks voicemail outcomes for follow-up sequencing.
- Callback Queue Management: When hold times are long or agents are unavailable, the AI phone agent offers callback scheduling, collects context, and manages the return call queue — reducing abandoned calls and improving satisfaction.
- After-Hours Phone Coverage: Handle all phone calls outside business hours with an agent that answers questions, takes messages, schedules callbacks, and escalates emergencies — without paying for overnight staff.
- Conference Call Coordination: Build a phone agent that dials multiple participants, manages the join process, handles late arrivals, and provides call controls — automating the logistics of multi-party phone calls.
- Phone System Migration Bridge: During phone system transitions, use an AI phone agent as an intelligent bridge that routes calls correctly across old and new systems while learning the new routing patterns.
Best Practices
- Account for Telephony Latency Separately: Phone calls have both AI processing latency and telephony network latency. Your AI phone agent prompt should include natural filler for both — brief acknowledgments for AI processing and appropriate hold language for transfer and routing delays. These are different situations requiring different responses.
- Build Voicemail Detection Into Every Outbound Script: Over 75% of outbound calls reach voicemail. Your phone agent must reliably detect voicemail greetings, wait for the beep, and deliver a concise message. Include detection rules for carrier voicemail, custom greetings, and screening services like Google Voice.
- Handle DTMF as a Parallel Input Channel: Some callers will press buttons instead of speaking, especially if they are used to traditional phone menus. Your AI phone agent prompt should accept both voice and DTMF input and respond appropriately to either. Map common key presses to departments.
- Design for Call State Changes: Phone calls go through state changes — ringing, connected, on hold, transferred, conferenced, disconnected. Your prompt should define agent behavior for each state, especially when unexpected transitions happen like a dropped transfer.
- Include Callback Number in Every Interaction: Phone calls can drop at any time due to carrier issues, dead zones, or battery. Always capture a callback number early and include it in your closing. If the call drops, the caller can be reached again without starting over.
- Test with Real Phone Networks: Phone audio quality varies dramatically between carriers, VoIP systems, speakerphones, and Bluetooth connections. Test your phone agent with real calls across different devices and networks, not just clean studio audio in development.
Common Mistakes to Avoid
- Not Handling Transfer Failures: When a call transfer fails — busy extension, no answer, wrong department — many AI phone agents have no fallback. The caller hears dead air or gets disconnected. Always define what happens when a transfer does not complete: offer voicemail, try another extension, or take a message.
- Ignoring Background Noise on Phone Calls: Phone callers are often in cars, on the street, or in noisy offices. Your AI phone agent prompt should include instructions for handling poor audio: ask the caller to repeat, confirm understanding of key details, and never guess at unclear information like phone numbers or names.
- No Strategy for Conference or Multi-Party Calls: If your phone agent might end up on a three-way call or conference bridge, it needs instructions for managing multiple speakers. Without this, the agent gets confused by overlapping voices and cannot track who said what.
- Treating Voicemail Like a Live Conversation: Some phone agents detect voicemail late and start their live greeting before realizing they are talking to a recording. Include early detection cues in your prompt — standard carrier phrases, recording quality indicators, and timing patterns — so the agent switches to voicemail mode quickly.
- Forgetting About Fax and Modem Tones: Outbound call lists may include fax numbers. If your AI phone agent does not recognize fax tones and disconnect, it will sit on the line wasting minutes and potentially leaving garbled audio on a fax machine. Include fax tone detection as an immediate disconnect trigger.
Frequently Asked Questions
What is an AI phone agent?
An AI phone agent is a voice AI system designed specifically for telephone calls. Unlike general voice assistants, phone agents handle telephony-specific features like DTMF tone input, call transfers, voicemail detection, hold management, and multi-department routing. They operate within the constraints of phone networks — single audio channel, variable audio quality, and carrier-specific behaviors.
How does an AI phone agent handle call transfers?
AI phone agents handle transfers by executing SIP or telephony API commands through their voice platform. The prompt defines when to transfer (based on caller intent), where to transfer (department extensions or direct numbers), and what to do if the transfer fails (offer voicemail, try alternate extension, take a callback request). Warm transfers keep the AI on the line until the destination answers; cold transfers connect directly.
Can AI phone agents detect voicemail versus live answers?
Yes, modern AI phone agent platforms include voicemail detection capabilities. The agent analyzes audio patterns — recorded greeting cadence, standard carrier phrases like 'leave a message after the tone,' and beep detection — to determine whether a live person or voicemail answered. Your prompt then defines separate conversation paths for each scenario, including a concise voicemail message to deliver after the beep.
What is DTMF and how do AI phone agents use it?
DTMF (Dual-Tone Multi-Frequency) refers to the tones generated when callers press phone keypad buttons. AI phone agents can detect these tones as input, allowing callers to press 1 for sales, 2 for support, and so on. This provides a fallback for callers who prefer button presses over voice commands, and it is essential for navigating other phone systems during outbound calls.
How reliable are AI phone agents in production?
Production reliability depends on three factors: prompt quality, platform stability, and telephony infrastructure. A well-designed AI phone agent prompt with proper error handling and fallback paths achieves 90-95% successful call completion rates. The remaining cases involve carrier issues, extremely poor audio quality, or edge cases the prompt did not anticipate. Iterating on your prompt based on call recordings is the fastest path to higher reliability.
Generate a Script for This Use Case
Use AI CallPrompt to create a structured voice agent prompt — free, fast, and optimized for real conversations.
Generate Script for This Use Case