ChatGPT Live Voice: voice mode learns to search the internet during conversation

A ZDNet reviewer tested the updated version of ChatGPT's voice mode — Live Voice — and noted that conversation with the assistant "almost" feels like communicating with a real person.

What's new about Live Voice

The main difference with the new ChatGPT voice models is that they can simultaneously listen to the user, respond by voice, and search for information on the internet without breaking the conversation into separate steps of "asked — waited — got an answer". Previously, voice dialogue with ChatGPT was structured differently: OpenAI showed Advanced Voice Mode back in 2024 along with the GPT-4o model — then the emphasis was on low response latency, natural intonations, and the ability to interrupt the assistant mid-sentence, but there was no live web search directly during the conversation, and the assistant still took a pause to find current facts.

• Live Voice can listen, speak, and search the internet simultaneously
• The feature is built into current ChatGPT voice models
• A ZDNet reviewer describes dialogue with the assistant as "almost" indistinguishable from talking to a person

How this feels in practice

According to the ZDNet reviewer who tested the feature, the naturalness of conversation is the main impression from the update: the assistant responds quickly enough and smoothly so that the pause before the answer doesn't give away the software in the interlocutor. At the same time, the material itself is framed as a review of personal experience and instructions on "how to try it", rather than an announcement of specific technical characteristics of the model — that is, we're talking about the subjective impression of a user, not measurements of latency or accuracy.

How this differs from competitors

The ability to conduct full-scale searches in real time directly during a voice call is a direction in which several major AI companies are moving simultaneously: a similar principle of "live" voice assistant with internet access is being developed by Google in the Gemini Live lineup. For the market, this means that the voice interface stops being simply a way to dictate a text query, which is then processed as a regular text chat, and becomes a separate channel for full-fledged interaction with the assistant — with its own pace, pauses, and ability to act during the conversation.

Where this can be useful in practice

A voice assistant that simultaneously listens and searches for answers on the internet is useful precisely in situations where your hands and eyes are busy with something else: while driving, in the kitchen while cooking, or simply when it's inconvenient to type on a phone. Previously in such scenarios, you either had to wait for the assistant to finish "thinking" after a question, or switch to text mode if you needed an exact current fact — the new version eliminates this gap between conversation and search.

What this means

Voice interfaces of AI assistants continue to move from simple speech-to-text conversion to models that can act during the conversation — search, clarify, and respond almost without delay — and it's precisely this smoothness, rather than a separate flashy feature, that makes an impression on users of an "almost real" interlocutor.