Voice turns ChatGPT into a spoken conversation. The important change: voice is now integrated into the main chat page for most people, rather than the separate full-screen blue orb that older guides describe. Both still exist, and you can choose.
1. Start it
- Web and desktop: click the Voice icon on the right side of the prompt window. The first time, your browser asks for microphone access.
- iOS and Android: tap and hold the large microphone button to talk in real time, and release to stop.
On first use you pick a voice. Change it later in settings, or from the customization menu at the top right of voice mode.
2. Choose integrated or separate mode
Most iOS and Android accounts get the integrated experience by default; some still see only separate mode while the rollout finishes. To switch either way:
Settings → Voice → Separate Mode
What it does: toggles between voice inside the chat page and the standalone voice screen. If your app looks nothing like a guide you’re reading, this setting is usually why.
3. Talk over it
Interrupting works — start speaking during a reply and it stops to listen, so you needn’t wait out a paragraph you already disagree with. The conversation and its memory are the same as the text chat, so you can switch modes mid-thread.
When it’s worth it, and when it isn’t
Voice earns its place hands-busy — cooking, walking, driving as a passenger — and for practising a language or rehearsing an argument out loud.
Skip it for anything you need to copy, edit precisely, or find again later: text is faster to scan and trivially reusable. For code or tables, ask it to switch to text for that part. Voice is one face of a multimodal model — same model, different channel.
If it fails
No icon at all usually means an unsupported browser or a blocked microphone permission; check the browser’s site settings first. If you want a transcript you can keep, record locally and run Whisper transcription instead. For the reverse direction, see text to speech locally.
Source: ChatGPT Voice FAQ.