ChatGPT Live Voice: voice mode can now search the web mid-conversation

A ZDNet columnist tested the updated version of ChatGPT's voice mode, Live Voice, and found that talking to the assistant "almost" feels like talking to a real person.

What's new in Live Voice

The key difference in ChatGPT's new voice models is that they can listen to the user, respond out loud, and search the web at the same time, without splitting the conversation into separate "ask — wait — get an answer" steps. Voice chat with ChatGPT used to work differently. OpenAI unveiled Advanced Voice Mode back in 2024 alongside GPT-4o. That release focused on low response latency, natural intonation, and the ability to interrupt the assistant mid-sentence, but it had no live web search during the conversation itself, and the assistant still had to pause to fetch current facts.

• Live Voice can listen, speak, and search the web simultaneously
• The feature is built into ChatGPT's current voice models
• The ZDNet columnist describes conversations with the assistant as "almost" indistinguishable from talking to a person

What it feels like in practice

According to the ZDNet columnist who tested the feature, the natural flow of conversation is the update's biggest selling point: the assistant responds quickly and smoothly enough that the pause before an answer doesn't give away that you're talking to software. That said, the piece is framed as a hands-on account and a "how to try it" guide, not as an announcement of specific technical specs for the model. In other words, these are one user's subjective impressions, not latency or accuracy measurements.

How it compares to competitors

Several major AI companies are pushing toward full real-time search during voice calls. Google is developing a similar "live" voice assistant with web access in its Gemini Live lineup. For the market, this means voice is no longer just a way to dictate a text query that then gets processed like an ordinary text chat. It is becoming a standalone channel for full interaction with the assistant, with its own pacing, pauses, and ability to act as the conversation unfolds.

Where it could be useful in practice

A voice assistant that listens and searches the web at the same time is most useful when your hands and eyes are busy: behind the wheel, in the kitchen while cooking, or simply when typing on your phone is inconvenient. In these situations, you previously had to either wait for the assistant to finish "thinking" after your question or switch to text mode if you needed a precise, up-to-date fact. The new version closes that gap between conversation and search.

What it means

Voice interfaces for AI assistants keep moving beyond simple speech-to-text toward models that can act during a conversation: searching, clarifying, and answering with almost no delay. It's this smoothness, rather than any single headline feature, that makes users feel they're talking to an "almost human" conversational partner.