The thing that used to give phone AI away was the voice. Flat, robotic, a half-second too slow. In my newest video I put ElevenLabs’ new V4 voice on my actual business line, called in like a customer, and recorded exactly what happened.
Fast enough to feel like a conversation
Two things make a voice work on a phone: speed and emotion. The new Turbo model answers in around 100 milliseconds, which is the gap between a natural reply and that awkward pause that tells a caller they’re talking to a machine. It also clones a voice from about 10 seconds of audio.
I gave it ten seconds of mine and set my expectations low. It came back close enough to be useful, not a perfect copy of me. Being bilingual, I ran it in Spanish too, where it leaned so hard into emotion it nearly sounded like a telenovela. The point isn’t that the clone is flawless. It’s that a believable voice is now cheap and fast, so the voice stops being the hard part.
Watching Anna book a real appointment
Demos are easy, so I skipped the demo. I put Anna, our AI employee, on the real line and called in myself. A caller pushed her for a number, and instead of inventing one she offered a free assessment, which is exactly the limit I’d set for her. Then she offered two Thursday slots, took the 10:30, asked for my full name and email, and read the email back to confirm before locking it in.
Small thing, that read-back, but it’s the difference between a booking and a wrong address. By the time I hung up, I had a summary of the call and a meeting link waiting. The loop actually closed, which is the whole promise behind the front desk that never misses a call.
The voice is the easy part now
Here’s the honest takeaway. The voice stopped being the hard problem. What makes an AI employee worth putting on your line is everything around the voice: that it knows your services, that it refuses to quote a figure it shouldn’t, that it books into your real calendar, and that it hands off to a person when a call needs one.
A great voice on top of a shaky setup is just a stiff reply that sounds nicer. That’s the same lesson behind the call that came in at 9:47 pm: answering is only worth something if what happens next actually closes the loop. We build that whole setup, voice included, around how your business works. If you’re curious what it would sound like, start with a free assessment.