AI Voices for IVR and Virtual Receptionists: Scripts, Voice Choice and Setup
A phone menu is the first thing many customers hear from your business, and most of them are recorded badly — a rushed take from whoever was in the office, or a stock voice that doesn’t match anything else about the brand.
AI voices fix the practical problem: you can update a prompt in a minute instead of rebooking a session, and every prompt in the system sounds like the same person. What they don’t fix is a badly written script. This guide covers both.
Key Takeaways
- Write the menu before you choose a voice. Bad structure is what frustrates callers, not voice quality.
- Keep options to four or fewer per menu, and put the most-used one first.
- Phone audio is narrowband — choose a clear voice over a characterful one.
- Check your phone system’s required audio format before generating anything.
- Confirm what your plan’s licence permits for commercial use before deploying to a customer-facing line.
Write the menu first
Callers don’t complain about synthetic voices. They complain about being read seven options when the one they needed was the seventh.
Four options maximum per menu. Anything more and callers stop listening and start pressing zero.
Most common option first. Look at what your team actually gets asked and lead with that.
Action before number. Say “For billing, press 2”, not “Press 2 for billing”. The caller hears what they need before the instruction, so they don’t have to hold two things in memory.
Give a way out. Always include a route to a person. Menus that trap callers generate more complaints than any other design flaw.
Say the wait honestly. If it’s eight minutes, say eight minutes. Callers tolerate waiting far better than they tolerate being misled.
Script templates
Adapt these, then generate all of them with the same voice so the system sounds coherent.
Main greeting
Thank you for calling [Company]. For sales, press 1. For billing, press 2. For technical support, press 3. To speak with someone, press 0, or stay on the line.
Out of hours
Thank you for calling [Company]. Our office is closed. We’re open Monday to Friday, 9am to 5pm [time zone]. You can reach us any time at [website], or leave a message after the tone and we’ll return your call on the next working day.
Hold message
Thanks for waiting. Your call is important and we’ll be with you shortly. If it’s easier, you can also reach us at [website] or email [address].
Voicemail
You’ve reached [Company]. We can’t take your call right now. Please leave your name, number and a short message, and we’ll get back to you within one working day.
Queue position
You’re number [N] in the queue. The current wait is about [N] minutes. Press 1 at any time to leave a message instead, and we’ll call you back.
Choosing a voice for phone audio
Phone calls run through a narrow frequency band that strips out most of what makes a voice sound rich. A voice that’s warm and detailed in headphones can turn muddy and hard to follow on a call.
What actually matters on a phone line:
- Clear consonants. Consonant definition survives compression; tonal warmth doesn’t.
- Measured pace. Slightly slower than conversational. Callers are often writing down a number or navigating a menu at the same time.
- Neutral delivery. Characterful voices are distracting across dozens of short prompts.
- Accent matched to your callers. Not to your brand aspirations — to the people who actually call you.
Generate your longest menu prompt and listen to it through an actual phone call, not through your computer speakers. It’s the only test that counts.
The IVR voice guide covers voice options for phone systems in more detail.
Audio format requirements
This is where most deployments stall. Phone systems are specific, and they generally reject anything that isn’t exactly right.
| System type | Typical requirement |
|---|---|
| Traditional PBX | 8kHz, 8-bit mono, µ-law WAV |
| Asterisk / FreePBX | 8kHz mono WAV or GSM |
| Cloud platforms (Twilio, RingCentral, 8x8) | MP3 or 16-bit mono WAV, usually resampled automatically |
| Microsoft Teams Phone | MP3 or WAV, under 5MB |
Check your provider’s documentation before you generate a full set of prompts. Converting afterwards is straightforward with a tool like Audacity or ffmpeg, but it’s easier to produce the right format once.
Two more practical details: normalise levels across every prompt so callers aren’t adjusting volume between menus, and leave about half a second of silence at the start and end of each file so the system doesn’t clip the first word.
Keeping a phone system consistent over time
The advantage of generated audio shows up at the third update, not the first.
- Keep the scripts in a document, versioned, with the voice name recorded next to each one.
- Use one voice across everything — greeting, hold, voicemail, out of hours, department menus.
- Regenerate the whole set when you change voice, rather than mixing old and new.
- Review quarterly, or whenever hours, departments or pricing change. Out-of-date phone prompts are a common and avoidable source of complaints.
Before you deploy: licensing
Customer-facing phone systems are commercial use, and rights differ by tool and by plan. Before you put generated audio on a business line:
- Check what your specific plan permits for commercial deployment.
- Confirm whether any attribution is required.
- If you’re deploying on behalf of a client, confirm whether the licence covers third-party use.
If you can’t find a clear answer in the terms, ask the vendor in writing before you deploy. Kveeky’s plan details are on the pricing page; for licence questions specific to your deployment, contact support and ask directly.
Frequently asked questions
Can I use an AI voice for my business phone system?
Technically yes — the audio is a standard file your phone system plays like any recording. The question is licensing, which depends on your tool and plan. Confirm commercial use is permitted before deploying to a customer-facing line.
What’s the difference between IVR and a virtual receptionist?
IVR is a menu that routes calls based on keypresses. A virtual receptionist is a broader term covering automated greeting and routing, sometimes with speech recognition. Both need the same thing from you: clearly written prompts in one consistent voice.
How long should a phone menu be?
The greeting and options together should run under 30 seconds. If yours is longer, you have too many options — split them into a second-level menu rather than reading a long list.
What audio format do phone systems need?
Most traditional systems want 8kHz mono WAV; cloud platforms usually accept MP3. Check your specific provider’s documentation before generating a full set, since requirements vary more than you’d expect.
Should the voice match my brand or my callers?
Your callers. A phone line is a utility, and comprehension beats brand personality. Pick the accent and pace your actual callers will find easiest to follow.
Getting started
Write the menu first. Cut it to four options. Read it aloud and time it. Then generate one prompt, listen to it over a real call, and only then produce the full set.
Generate a test prompt free — no card required. The IVR voiceover page covers voice selection, and if you’re producing training or product audio alongside your phone system, the training video guide uses the same workflow.