How the voices for ChatGPT were chosen

Turns out picking a voice for an AI is a lot like casting a movie. OpenAI says it worked with professional casting and directing talent to sift through more than 400 submissions before landing on the five voices that now read back your ChatGPT conversations. That's a surprisingly analog process for a company built on training massive neural networks.

The detail that stands out here is the scale of the funnel. Four hundred submissions is a real casting call, the kind you'd expect for a national ad campaign or an animated feature, not a software feature. It suggests OpenAI treated voice as a product decision with real stakes, not an afterthought bolted onto the text interface.

It also hints at why voice AI has become such a sensitive area for the company. After the Sky voice controversy last year, where comparisons to Scarlett Johansson sparked public backlash and a legal letter, OpenAI has clearly gotten more careful about how it sources and vets the people behind these voices. Bringing in industry casting professionals, rather than handling selection purely in-house, reads like a direct response to that episode.

What OpenAI hasn't detailed is the criteria that mattered most in the final cut, warmth, clarity, distinctiveness, or something else entirely. But the fact that they're willing to talk about the process at all signals a company trying to make voice selection feel less like a black box and more like a craft, complete with the same rigor Hollywood applies to picking who gets to play the lead.