The Uncanny Valley - Why Early AI Voice Felt Wrong, and Why They Don’t Now

There's a concept in robotics called the uncanny valley. It describes the strange discomfort we feel when something looks almost human — but not quite. A robot with a vaguely human face triggers unease in a way that a clearly mechanical robot never does. The closer it gets to human without arriving, the more unsettling it becomes.

The same phenomenon exists in sound. And for several years, AI voices lived right in the middle of it.

The Problem With "Almost Right"

Early AI voices weren't bad in an obvious way. They pronounced words correctly. They had rhythm and pace. On the surface, they functioned. But something was off — and that something was deeply felt, even if most listeners couldn't name it.

The breath was in the wrong place. The emphasis landed a half-beat late. The emotional tone of a sentence didn't quite match its meaning. Everything was technically present, but the human truth behind it was missing. The voice knew what to say, but not what it meant.

That gap — small, specific, hard to articulate — was more disturbing than silence would have been. It's why "almost human" is creepier than "not at all human." When something is clearly artificial, your brain accepts it on those terms. When something is almost right, your brain keeps searching for what's wrong — and can't stop.

This is what gave AI voice its early reputation. Not that it was terrible. That it was uncanny.

What Changed

The journey from uncanny to indistinguishable wasn't a single breakthrough. It was a gradual accumulation of improvements — each one small, collectively transformative.

Modern AI voice models were trained on vastly larger datasets, capturing not just how words are pronounced but how people actually speak — the micro-variations in pace, the natural irregularities, the subtle emotional colouring that makes a voice feel inhabited. The technology learned that real speech is imperfect in specific, predictable ways — and that those imperfections are precisely what makes it feel human.

The result is something that has crossed the valley. Not every AI voice, and not in every application — but the best of what's available today doesn't trigger the uncanny response. It simply sounds like a person speaking. Listeners don't notice it, which is the only benchmark that matters.

Why This Matters for Advertising

Radio is an unforgiving medium for the uncanny valley. There's no visual to distract from it, no image to carry the weight. The voice is the entire experience — and if something is even slightly off, the listener feels it before they can explain why.

That vague wrongness doesn't just fail to engage. It actively creates distrust. The brain's response to the uncanny is protective — something isn't right here, don't relax. That's not a feeling you want associated with your brand.

This is why the production process around AI voice matters as much as the technology itself. The gap between an AI voice that sounds natural and one that still lives in the valley isn't the model — it's the craft applied to it. Voice selection, script writing, pacing, production quality: these are the decisions that determine which side of the line the finished ad sits on.

The Valley Is Behind Us — For Those Who Know How to Use the Technology

Early AI voice had an excuse. The technology was genuinely limited, and the uncanny response was a reasonable one.

That excuse no longer exists. The tools available today are capable of producing voices that are, in the right hands, completely indistinguishable from a human in a professional studio. The listeners in your audience won't know. They'll just hear an ad that sounds right — and that's the point.

The uncanny valley was a phase, not a destination. We're past it. The only question now is whether the people producing your advertising know that — and know how to prove it.

At Brand New Day, we've been watching this technology develop for years, and we know exactly where the line is. Get in touch and hear it for yourself.

For a complete overview of radio advertising and what makes it work, see our full guide to radio advertising in Australia.