ChatGPT voice mode lets you talk to the chatbot out loud instead of typing. You ask a question with your voice, ChatGPT processes it and speaks the answer back. Since late 2025 the feature has been built into the normal chat screen, both in the app and on desktop, and it is available to all logged-in users. There is a difference, though, between what free users and paying subscribers get, and there are limitations you should know about before you rely on spoken output.
What is ChatGPT voice mode?
ChatGPT voice mode is a way to communicate with OpenAI’s chatbot using spoken language instead of typed messages. Voice mode is available in both the mobile app and the desktop version. You speak your question, ChatGPT generates an answer and reads it back to you in a synthetic voice. The feature works in more than forty languages and is available to all logged-in users on iOS, Android and desktop.
Until late 2025, you started a voice conversation in a separate mode with a blue orb-shaped screen. Since late 2025, voice has been built into the normal chat window by default. Your questions and the answers appear in your chat history while you talk, and you can keep referring to earlier messages and images during the conversation.
You do not need a separate subscription to use voice. There is a difference, however, between what free users and paying subscribers get, mainly in daily usage time and in the language model running under the bonnet. More on that below.
How to turn on voice mode in the ChatGPT app
Turning on voice mode in the ChatGPT app takes only a few taps. One important point: make sure you have installed the latest version of the app first. Older versions still show the standalone voice mode and lack recently added features such as live video sharing.
Open the ChatGPT app on your iPhone or Android device and log in. To the right of the input field you will see a sound wave icon. Tap it to start a voice conversation. The first time, the app asks for access to your microphone. Grant it, or the feature will not work.
In your first conversation you choose a voice. ChatGPT offers several options, such as Ember, Juniper, Breeze and Cove, each with its own tone and character. You can switch voices later via Settings, Voice. Then start talking in your own language. ChatGPT detects the language automatically and answers in the same language.
If you would rather talk in a separate screen without the chat history, you can turn that on via Settings, Voice, Separate mode. This brings back the old blue orb-shaped screen.
If you accidentally tap deny when asked for microphone access, you will need to turn that permission on manually later in your phone’s settings, not in ChatGPT itself. On iOS, go to Settings, ChatGPT, Microphone and turn the switch on. On Android, open Settings, Apps, ChatGPT, Permissions, Microphone and choose Allow. The exact route differs by brand (Samsung, Google Pixel, OnePlus), but the steps are similar.
If the microphone still does not work, check whether another app is holding on to it, for example Teams, Zoom or a browser tab with a video call. Only one application at a time can have exclusive access.
How to use ChatGPT voice mode on desktop
ChatGPT’s voice feature has worked on desktop since late 2024, and since late 2025 it has been built into the main screen there too. You do not need a separate application or browser extension. It only works, however, on chatgpt.com in the browser or via the official desktop app for macOS and Windows.
Open chatgpt.com in a modern browser such as Chrome, Edge, Safari or Firefox. Log in to your account. Next to the input field at the bottom is the same sound wave icon as in the app. Click it and give the browser permission to use your microphone. Some browsers ask for this permission in every new session; others remember it.
On desktop, permission is handled by the browser, not by your operating system. In Chrome or Edge, click the padlock to the left of the URL and set microphone access for chatgpt.com to Allow. In Safari, go to Safari, Settings, Websites, Microphone and choose Allow for chatgpt.com. For the official desktop app on macOS, go to System Settings, Privacy & Security, Microphone; on Windows, go to Settings, Privacy & security, Microphone.
The main advantage on desktop is the combination of speaking and reading along. You dictate a long instruction and watch the text appear at the same time, so you can check straight away whether ChatGPT has picked up your instructions correctly. For longer tasks that is faster than typing, provided you articulate clearly. If you want to speak but do not need a spoken answer, use dictation via the microphone icon: ChatGPT turns your recording into text that you can edit before you send the message. Since 26 June 2026, dictation has run on a new speech-to-text model on all plans.
A laptop microphone is generally good enough for recognisable speech, but in a room with background noise it pays to use a headset or external microphone. The reason is not the sound quality itself: with noise, ChatGPT more often misses parts of your question or transcribes them incorrectly.
The difference between Live, Advanced and Standard Voice
The difference between Live, Advanced and Standard Voice in ChatGPT lies in the underlying model and in how a conversation flows. You choose from these three options under Settings, Voice. According to OpenAI, which ones you see depends on your plan, your workspace settings, your region, your app version and parental controls.
Live is the newest option and is designed for a natural conversation in which you talk back and forth smoothly. Live listens and speaks at the same time, so you can interrupt an answer as you would in an ordinary conversation. During the conversation, Live can search the web and use the plugins and connected apps on your account. Since 15 October 2026, Live has run on GPT-Live-1.1 and GPT-Live-1.1 Quip, depending on your plan, and free users have limited access to Quip. Voice is now also available in Business, Enterprise, Edu and Healthcare workspaces, depending on the workspace settings. In each rolling 24-hour period, Go gets up to three hours with GPT-Live-1.1 Quip, Plus up to three hours with GPT-Live-1.1, and Pro up to fifteen hours or unlimited, depending on the Pro tier. In Business, a Standard seat gives three hours and a Premium seat fifteen hours per rolling 24 hours; extra usage after that costs 1.25 credits per minute. Since 9 September 2026, ChatGPT Voice can call on GPT-5.6 or GPT-6 Astra when a question needs searching or more reasoning, which has done away with the separate Instant, Medium and High levels for voice. Live does not support video or screen sharing.
Advanced is the previous real-time voice experience. Advanced runs on a speech-native model that processes audio input directly and responds in audio. As a result, the model responds quickly and recognises tone and emotion. Choose Advanced when you need video or screen sharing. Those features remain available to eligible subscribers in the ChatGPT iOS and Android apps, and only via Advanced.
Standard works turn by turn. ChatGPT first converts your spoken question into text, sends that text to the language model and then reads the answer aloud in a synthetic voice. Choose Standard if you do not want ChatGPT to keep audio recordings of your conversations, because with Standard the audio is deleted after transcription.
In practice you notice the difference most in fast, natural conversations and in languages other than English. In those languages Advanced sounds smoother than Standard, and the chatbot can pick up a question halfway through without you having to finish speaking first. If you use voice occasionally for a short question, the difference is small. If you dictate or hold long conversations every day, it is noticeable.
Limitations of ChatGPT voice mode and what to watch out for
ChatGPT voice mode comes with limitations that are less obvious in written use. The most important one: everything ChatGPT gets wrong in text, it also gets wrong in speech. The difference from written output is that you do not read along in a voice conversation. A mistake in a spoken answer slips past more easily than the same mistake in a chat window that you go through word by word.
During the period when I used ChatGPT intensively for SEO work and content production, I consistently found that the model makes up sources, quotes and figures that sound plausible. In a written conversation that stood out as soon as I clicked on the source; in a voice conversation you hear a name or a percentage go by, and the temptation to accept it without thinking is much greater. If you use the output to support decisions or to publish something under your own name, that is a real risk.
A second point to watch is summarising. When you summarise with AI by voice, whether a document or a long explanation, you get back a fluent, convincing text. Its content may differ from the source in ways you cannot verify in spoken form. For LearnLLM learners the rule is: treat a spoken summary as a first impression, not as the factual content.
Voice also has practical limitations. Background noise and unclear articulation lower the transcription quality. Jargon and proper names are regularly misrecognised. Spoken answers in languages other than English sometimes have a slight American accent, which takes some getting used to in a business context. And you can only hold one voice conversation at a time per account.
The way to use voice professionally is to apply the same checkpoints as with written AI output: you reread an important outcome in the transcript, you verify a fact against an external source, and a text that goes out always ends with human editing. In my own way of working, that last step is not optional but part of the workflow. You will find a good explanation of that way of working in LearnLLM’s ChatGPT training for professionals.
Frequently asked questions about ChatGPT voice mode
Most follow-up questions about ChatGPT voice mode concern privacy and use on the go.
Does ChatGPT keep your voice recordings?
Yes, ChatGPT keeps voice recordings from Live and Advanced for 30 days, together with the transcript in your chat history. If you delete the chat, the associated recordings are removed within 30 days, according to OpenAI, except where they are needed for safety or legal obligations. Archiving a chat does not delete the recordings. With Standard, the audio is deleted after transcription, and OpenAI only uses recordings for training if you choose to share your audio.
Can you use ChatGPT voice mode in the car?
Yes, ChatGPT’s voice feature works in the car via Apple CarPlay on supported iPhones. From the CarPlay screen you start a voice conversation or continue a recent or pinned chat. Under Settings, Voice you can set ChatGPT to start with voice automatically in CarPlay; that option only appears after you have used ChatGPT in CarPlay once. OpenAI recommends setting up the app before you start driving.
How do you change the language in ChatGPT voice mode?
You change the language under Settings, Voice, Language: choose the language you speak most often. According to OpenAI, that choice helps process your speech more accurately when voice picks up the wrong language. During a conversation you can also ask ChatGPT to speak another language. On an iPhone, if there is distracting noise, you can turn on Voice Isolation under Mic Mode in Control Centre.
How to use ChatGPT voice mode at work
In practice, using voice professionally means using it where it adds value and avoiding it where it creates risk. Good uses: thinking out loud on a walk, dictating a first outline while driving, having a complicated concept explained while your hands are busy with something else, or practising a foreign language. For uses like these, speaking often feels more comfortable than typing.
Voice is less suitable for tasks where factual accuracy counts: checking figures, going through legal texts or summarising policy documents. On those tasks voice is technically no worse than text, but the combination of a fluent spoken voice and the lack of visual checking makes it easier to accept something unchecked. The four-eyes principle from the financial sector is useful here: nothing goes out without a second check. That principle applies to AI output too, and to spoken output even more so. A broader overview of what you can do with ChatGPT in your work helps you make that trade-off.
LearnLLM’s ChatGPT course teaches you how to use the tool in your daily work without compromising on reliability. You learn to build a repeatable workflow, set up fixed checkpoints and choose deliberately when voice or text is the better mode. You finish the course with a work portfolio and a working workflow based on your own tasks. Enrol in the ChatGPT Course and use voice as part of a responsible approach.
