Will It Sound Like a Robot? What Shapes How the Receptionist Talks on a Call

What makes a call feel robotic

Four habits do most of the damage. Speeches where one short answer would do. Several questions in one breath. Asking for something the caller already said. And running ahead, answering its own question on the caller’s behalf.

The rules written into the prompt

Replies are kept short, because it is a phone call and not an essay. It asks at most two related questions in a turn, like brand and model, and then waits. It keeps a running list of what the caller has said and never asks for a field it already has.

• It speaks only its own side of the call. It never writes or assumes what the caller will say.
• It never uses a name the caller has not given. Greeting someone as “John” when they never said John is a tell.
• Dates come from the calendar’s own labels, not from arithmetic, so it does not mix up Thursday and Friday.
• A caller who says “I don’t know” to a model number is not blocked. It moves on.

It stays in the caller’s language

If the caller opens in French, the receptionist continues in French through the booking and the text that follows. It restates the recording notice in French the first time it switches, since the greeting was spoken in English before it knew who was calling.

What the voice vendor controls

We choose the voice and write the conversation rules. The speech timing sits in the voice pipeline underneath: when the receptionist starts talking after a pause, and how it handles a caller who interrupts. We configure that and test it, but we do not control it line by line.

The rules above are written in the prompt, not enforced in code, so they can slip. We read transcripts for that reason. “Natural, human-paced” is a fair description. “Never talks over a caller” is not a promise anyone should make.

It says what it is

Asked whether it is a person, a robot, or an AI, the receptionist always says it is an AI assistant. A shop can also have it say so in the greeting. A recording notice is spoken when you have configured one. Sounding natural does not change that.