It should sound natural, warm and easy to talk to. It shouldn't pretend to be a person. Those are two different things, and mixing them up is where AI receptionists go wrong. A clear, friendly voice that says early on that it's an AI gives callers the comfort of a good conversation and the honesty they need to decide how much to share. A voice designed so callers "can't tell" saves you a moment of awkwardness at the start and risks something much bigger later: the moment a caller realises, and wonders what else wasn't true.

Here's how to think about it from the caller's side.

What callers actually want from the voice

Picture someone ringing a business at 8.30pm to ask whether they can bring their dog to a grooming appointment on Saturday. What do they need from the voice that answers?

  • To understand it, and be understood, first time.
  • To not feel rushed or talked over.
  • To get a straight answer, or a clear "I'll make sure someone gets back to you".
  • To know where they stand: who or what they're talking to, and what happens next.

Notice what isn't on the list: "to be fooled". Callers want the call to be easy. Realism helps with the first two points. Honesty covers the last one.

Natural is good. Indistinguishable is a trap.

A robotic, flat or stilted voice makes people work harder. They speak slowly, repeat themselves and hang up sooner. So a natural-sounding voice with sensible pacing and a bit of warmth genuinely helps.

The trap is pushing past "natural" into "convincingly human" and then not saying so. A few things tend to happen:

  • Callers share more than they would have. People speak differently to a person than to a machine. Some will mention health details, family situations or money worries because they think a sympathetic receptionist is listening.
  • The reveal feels like a trick. Every AI hits a moment it can't handle. When that happens to a caller who thought they were talking to a person, the reaction is rarely "how clever". It's closer to "why didn't you tell me?"
  • Small oddities get magnified. A slightly wrong pause or a misheard name is forgivable from an assistant that's said it's an AI. From a "person", it's unsettling.

There's a long-standing idea in robotics, the "uncanny valley", that something almost but not quite human can feel more off-putting than something clearly artificial. Whether or not it applies neatly to voices, the practical lesson is the same: aim for pleasant and clear, not for a disguise.

What guidance says about telling callers it's AI

In Singapore, there's no specific rule we know of that requires a business to tell callers they're speaking to an AI. But the direction of official guidance is clear.

The PDPC and IMDA's Model AI Governance Framework (second edition) (checked 8 November 2026) encourages organisations to "provide general information on whether AI is used in their products and/or services", giving the example of telling users "they are interacting with an AI-powered chatbot" (paragraph 3.46). It also suggests organisations consider whether to offer people an option to opt out of the AI service, and where they don't, to consider other "modes of recourse" (paragraphs 3.52 to 3.53).

Outside Singapore, the EU has gone further. The European Commission's FAQ on Article 50 of the EU AI Act (last updated 24 July 2026, checked 8 November 2026) explains that AI systems that interact directly with people must be designed so people are informed "from the start of the first interaction", with an exception only where it's obvious, and that this duty applies from 2 August 2026. It may not apply to a Singapore business serving local callers, but it shows where expectations are heading.

This is general information, not legal advice. The point for most owners is simpler: telling callers up front is the honest default, and it's what guidance points towards.

How to sound warm and still be honest

Being upfront doesn't have to sound cold or corporate. What tends to work for callers:

  • Say it early and lightly. One short, friendly line near the start is enough. Something like: "Hi, you've reached [Business name]. I'm the AI assistant, and I can help with bookings and questions, or take a message for the team." (Hypothetical wording, for illustration.)
  • Then get on with helping. The disclosure isn't a speech. Callers mainly want to know what you can do for them.
  • Use the business's tone. A tuition centre, a clinic and a car workshop shouldn't sound identical.
  • Admit limits plainly. "I'm not sure about that, so I'll take your details and someone will call you back" builds more trust than a confident guess.
  • Answer the question "Are you a real person?" honestly, every time. No dodging, no jokes that leave it unclear.

Questions to ask about any AI voice that would answer your phone

Before you let an AI answer for your business, listen to it as a caller would and ask:

  1. Does it say it's an AI near the start of the call?
  2. If I ask "Am I talking to a person?", does it answer clearly?
  3. Does the voice sound pleasant and easy to follow, or flat and tiring?
  4. Does it pause naturally and let me interrupt, or talk over me?
  5. When it doesn't know something, does it say so, or bluff?
  6. Would I feel misled if I found out halfway through that it was an AI?

If the answer to the last question is yes, the voice is doing too much pretending.

When an AI voice is the wrong choice, however natural it sounds

Some calls need a person, and a better voice doesn't change that:

  • Emergencies. Anyone describing a medical emergency or fire should be told to call 995, or 999 for police. An AI shouldn't be their first stop, and nor should any receptionist.
  • Distress or bad news. Callers who are upset, grieving or frightened deserve a human.
  • Complaints and disputes. People want to be heard by someone who can act.
  • Judgement calls. Clinical, legal or financial advice, or decisions that commit the business, belong with a qualified person.
  • Callers who ask for a person. Respect it, and make sure there's a clear way to get a human back to them.
  • Customers who won't talk to automated systems. Some groups strongly prefer a person. If that's most of your callers, listen to them.

Our guide on whether a clinic should let an AI receptionist answer the phone goes deeper on where that line sits for sensitive calls.

Hear the difference for yourself

The quickest way to form a view isn't reading about voices. It's talking to one.

We built an AI receptionist for our own enquiries. Her name is Penny. She tells you she's an AI in her very first line, then gets on with helping, and at the end of each call the business gets a summary of who called, what they asked and the next step. You can talk to her by voice or type to her in your browser at penny.pencilglow.com, with no sign-up and no contact details needed. Ask her whether she's a real person. Try to trip her up with an awkward question. Then decide what you'd want your own callers to hear.