Voice Mode

Talking with ChatGPT

ChatGPT Voice turns the assistant into a spoken conversational partner. You speak, it answers out loud, and the entire conversation lives in a single chat thread: you can follow the response in text, type when you can’t speak, and review past exchanges without starting over. This feature changes how you use the tool more than any other because it removes the keyboard—and with it, the need to sit in front of a screen.

Starting a voice conversation

On iPhone and Android, tap the Voice icon in the input bar, grant microphone access if the app asks, and choose a voice the first time you use it. Then speak naturally. During the conversation, the microphone button mutes or unmutes you, and an exit button ends the call. On the web, the steps are the same: tap the Voice icon in the chat input area at chatgpt.com, then allow microphone access in your browser.

In the desktop app, Voice is designed primarily to drive Work and Codex: launch a task, track its progress, and coordinate multiple agents by voice. For spoken conversations in Chat, use the web or mobile apps.

Don’t confuse Voice with dictation, the separate microphone icon. Dictation transcribes a message that you review and edit before sending. If you need word-for-word text, use dictation; Voice is built for conversation.

Three modes, one setting

Under Settings > Voice, you’ll find three options, depending on your plan, region, app, and workspace.

• Live is the most natural. Powered by GPT-Live-1.1 or its lighter variant GPT-Live-1.1 Quip, depending on your plan, it listens and speaks simultaneously: you can interrupt it just like a human interlocutor. It can search the web, rely on memory, display visual results, use your plugins, and accept both text and images in the same thread.
• Advanced is the previous real-time experience. It remains useful on mobile for tasks Live doesn’t support: sharing live video from your camera or your screen, for eligible subscribers.
• Standard works turn-by-turn: your speech is transcribed, then ChatGPT responds.

When Live needs to take action in a connected app, approval happens on-screen: voice confirmation is not accepted on web or mobile.

How much voice time do you get?

In Chat, Live usage is measured on a rolling 24-hour basis, and ChatGPT notifies you when you approach your limit.

What voice makes possible

The conversation is truly bidirectional: with Live, you can cut in to rephrase or change direction, and the flow stays smooth. Switching languages happens on demand: start in French, ask for a phrase in English, then continue in Spanish.

“How do you say ‘I would like to reschedule our meeting to next week’ in professional English?”

ChatGPT gives you the phrase orally, and you immediately ask it to repeat, rephrase, or offer a more formal version. The back-and-forth takes ten seconds, whereas it would have required three written messages.

Brainstorming is another ideal use case for voice because it tolerates imprecision and thrives on momentum.

“I need ideas for our training school’s back-to-school campaign. We’re targeting career-changers in management. Throw out some ideas, and I’ll tell you which ones I like.”

The same mechanism works for practice: “Run a job interview simulation with me for a digital project manager role. Ask me one question at a time and give me feedback after each answer.” Or for language practice: “Let’s practice Spanish. You speak to me in Spanish, I’ll answer in Spanish. Correct my mistakes at the end of each exchange.” In both cases, speaking isn’t just a convenience; it’s the core mode of the exercise.

When to stick to the keyboard, and best practices

Voice wins when you’re on the move, when your hands are full, or when thinking out loud comes more naturally than writing. It loses in a quiet or shared office, and whenever the task involves code, tables, or text you need to copy exactly.

A few habits significantly improve the experience. Speak normally, without over-articulating. Stay precise, just as you would in writing: a vague request yields a vague answer. Interrupt immediately if the response goes off track. For time-sensitive questions, note that Voice relies on your device’s time zone to understand “today” or “tomorrow”: specify the date, time zone, or location if the response seems off, and verify any critical information, just as you would in writing.

On the data side, audio has its own setting: under Settings > Data controls, the Include your audio recordings option determines whether your voice and dictation recordings can be used to improve the models. Turning off Improve the model for everyone also stops this sharing.

Your turn

Ten minutes spoken. Open Voice on your phone or on the web and check the active option under Settings > Voice. Ask for an explanation of a concept from your field, interrupt it with a follow-up question, have it translate a sentence into English and then German, and finish with three minutes of brainstorming on a current project. Then review the written thread: it keeps a record of everything that was said.

Official sources: ChatGPT Voice (OpenAI Help Center: Live, Advanced, and Standard options, getting started, limits by plan, video and screen sharing) · ChatGPT Work and Codex (Voice in Work and the desktop app) · Voice Dictation FAQ (dictation and audio sharing)

Key takeaways

• Launch Voice via the Voice icon in the input bar on web and mobile; on desktop, it primarily drives Work and Codex
• Three options in Settings > Voice: Live (most natural), Advanced (video and screen sharing on mobile), and Standard (turn-by-turn)
• Voice time depends on your plan and is measured on a rolling 24-hour basis
• Ideal for brainstorming, language practice, interview prep, and mobility
• For word-for-word text, use dictation (microphone icon); for conversation, use Voice