How to actually use the new ChatGPT voice upgrade

ChatGPT voice just had its biggest upgrade so far. It’s no longer just a talking search bar that answers a question and stops. Used properly, it can now start real tasks, use plugins, create documents, and keep working even after you hang up.

This guide walks you through what changed, how to turn it on, the limits that will trip you up if you don’t know about them, and a few power-user tips to make voice part of your daily workflow.

What changed in the new ChatGPT voice update

The latest update (September 23, 2026) turns ChatGPT voice into something much closer to a hands-free control panel. Two core changes matter most:

1. Live voice can now use plugins. Live is the new voice mode that listens and speaks at the same time. It works on web, iOS, and Android. It can now talk to the apps and services you’ve connected to ChatGPT (previously called apps, now plugins), so during a voice call it can reach into those tools and actually do work for you.

2. Voice now works inside ChatGPT Work. Work is the mode designed for longer, more serious jobs: research, analysis, and finished files like documents, presentations, and spreadsheets. With the update, you can speak and have ChatGPT build those assets directly inside Work.

The key shift: before, voice was just conversation. You asked, it answered. Now, your voice can kick off real tasks that run in the background and produce files you can use.

Why this matters for your daily workflow

Think about all the “dead time” in your day: walking, commuting, washing dishes, waiting in line. Until now, that time was mostly useless for deep work.

With Live voice + Work, you can:

• Talk through a one-page summary and come back to a finished draft.
• Dictate the outline of a presentation and let ChatGPT build the slides.
• Describe a simple spreadsheet you want and have it created while you’re on the move.
• Trigger plugins to pull data, run searches, or connect to other tools, all by voice.

And here’s a crucial detail: if you end the call while a task is still running, it doesn’t stop. The work continues in the text chat. When you return, the output is waiting for you. That’s what makes this usable in real life instead of just a cool demo.

How to start a ChatGPT voice call

Getting into voice is straightforward, but the power is in the settings you pick afterward.

On mobile (iOS and Android)

1. Open the ChatGPT app.
2. Tap the voice icon in the message bar.
3. Allow microphone access if your phone asks.

On web

1. Go to chatgpt.com.
2. Open a new chat.
3. Click the voice icon in the prompt box.
4. Allow your browser to use the microphone.

That starts a call, but to unlock the new behavior you need to pick the right voice mode.

The one setting most people miss: choosing the right voice mode

Inside ChatGPT settings, there’s a Voice section with three options. They behave very differently:

1. Live
• Listens and speaks at the same time.
• Can use plugins.
• Ideal for hands-free control and running tasks in Work.

2. Advanced
• The older real-time mode.
• Supports video and screen sharing (mobile).
• Use this if you need to show your screen or work with video.

3. Standard
• Transcribes what you say first, then responds in text and/or voice.
• Slower but more like classic “dictation plus answer.”

To use the new capabilities, go to Settings → Voice and select Live. That’s the only mode that can use plugins during a call.

Using plugins with Live voice

Plugins (formerly called apps) connect ChatGPT to other tools and services. With Live, you can now trigger those connections by voice.

Here’s how it works in practice:

1. Make sure the plugin is already installed and connected in your ChatGPT account.
2. Start a Live voice call.
3. Say what you want, and explicitly tell it which plugin to use if needed.

If you ask it to use a plugin that isn’t connected yet, ChatGPT will prompt you to sign in and approve access. Do this before you start a serious voice session or you’ll break your flow halfway through.

All your existing permissions and workspace restrictions still apply. Voice doesn’t bypass anything—you only get access to what you already have in your account.

The silent approval trap that stops tasks

There is one behavior that catches almost everyone the first time they try this.

When ChatGPT wants to do something that needs your explicit approval (for example, sending an email, posting something, or taking a sensitive action), it asks for confirmation on-screen, not out loud.

Important details:

• You must tap Approve or Decline yourself.
• Spoken approval (like “yes, go ahead”) is not supported for these actions.
• If your phone is in your pocket and you ask it to send something, it will wait silently on the approval prompt until you look at the screen.

Know this before you rely on voice to send or publish anything. If a task seems “stuck,” check the screen for an approval request.

What a real voice-to-work run looks like

Here’s an example of how a full run might work inside Work using Live:

1. Open ChatGPT, start a voice call, and choose Live.
2. Say something like: “Create a one-page summary document on [topic]. Pull out the main points and keep it concise.”
3. Because you’re in Work, ChatGPT treats that as a real document task and starts building it.
4. While it talks, you’ll see the written response appear in the chat in real time, so you can read instead of trying to remember everything it says.
5. If you hang up before it finishes, the task continues in text. You can come back later and pick up from the completed draft.

This same pattern works for presentations and spreadsheets too, as long as those tools are available in your Work environment.

Models behind Live voice and how they switch

Live calls run on a model called GPT Live 1 or GPT Live 1 mini, depending on your plan. These are optimized for real-time conversation.

When a question needs heavier reasoning or deeper search, voice can also reach for OpenAI’s larger models. You control this with the same model and reasoning settings you already use in text chat. So your voice sessions benefit from the same intelligence you’re used to when typing.

You can also mix modes during a call:

• Send text messages in the same chat while voice is active.
• Attach images using the add button in the message bar and talk about them during the call.

Current limitations you should know about

Live is powerful, but it’s not unlimited. A few constraints matter if you’re planning to use it for serious work:

• No video or screen sharing in Live. If you need to show your screen or use video, switch to Advanced on mobile.
• No automatic access to your ChatGPT library files. Live can’t currently pull files directly from your library, though you can usually attach files to the chat manually.
• Only one voice conversation at a time. You can’t run multiple parallel voice calls.
• Audio clips are stored. Audio from your voice calls is stored with the transcript in your chat history for up to 30 days. If you delete the chat, those clips are also removed within 30 days. Keep this in mind for sensitive topics.

Who gets access to the new voice features

Access depends on your plan and workspace settings.

Individual plans

• Free and Go: Can use voice in chat, with the plugins supported by your plan.
• Voice usage runs on a rolling 24-hour window. When you hit your limit, ChatGPT will tell you.

Approximate Live usage:

• Free: Limited access to the mini Live model.
• Go: Around 3 hours on the mini model.
• Plus: Around 3 hours on the full Live model.
• Pro and higher: Significantly higher limits.

Your Work usage still counts the same whether you talk or type. Voice doesn’t give you “extra” model time; it just changes how you use it.

Business, enterprise, EDU, and healthcare

• Voice in Work requires both voice access and Work access on your account.
• Voice is available in eligible business, enterprise, education, and healthcare workspaces, but your admin controls whether it’s enabled.

Turning voice into a real productivity tool

Talking is faster than typing. The real advantage comes from pairing that speed with tasks that produce actual output: documents, research summaries, slides, spreadsheets, and more.

But speed only matters if the work you create is seen and used. That’s especially important now that search is changing: people are getting answers from ChatGPT, Perplexity, Gemini, and AI overviews at the top of search results pages instead of scrolling through links. If you’re producing content, you need to understand how AI-driven discovery is reshaping where attention goes. For a broader view on this shift, it’s worth reading how Apple’s on-device AI bet is changing the game and how enterprise tools like Odysseus build on top of ChatGPT-style systems.

7 practical tips so you don’t waste your first week

To make the new ChatGPT voice actually usable day to day, a few simple habits help a lot.

1. Use headphones.
Live listens while it speaks. If your phone’s speaker is feeding back into the mic, you’ll get interruptions and misheard phrases. Headphones fix most of that.

2. Tell it when you want to think out loud.
If you’re going to ramble or brainstorm, say something like, “Let me think out loud, don’t reply until I ask.” It’s not perfect, but it cuts down on interruptions.

3. Connect plugins before you start.
Sign into and authorize any plugins you plan to use before your call. Nothing kills momentum like having to log into a service mid-sentence.

4. Turn on background conversations.
In Settings → Voice, enable background conversations if you want the call to continue while you switch apps or lock your phone.

5. Give it one clear job per call.
Voice works best when you give it a single, well-defined task: “Draft a one-page summary,” “Outline a 10-slide deck,” “Compile a comparison table.” It struggles more when you pile five different jobs into one conversation and expect it to organize everything perfectly.

6. Check the text transcript.
Transcripts are not word-for-word and can drift, especially if people talk over each other. Always skim what it wrote instead of trusting only what you heard.

7. Try different voices.
In Settings → Voice, you can choose from several voices (like Arbor, Cove, Juniper, and Sol). The one you pick genuinely affects how long you’re willing to stay in a call, so experiment until you find one that feels comfortable.

Two hidden settings that make you use voice more

There are two easy-to-miss options that dramatically change how often you actually reach for voice.

Start with voice
• Found in Settings → Voice.
• When turned on, opening a new chat drops you straight into a voice call—no need to tap the mic icon.

If you’re trying to build a habit of talking instead of typing, this one switch does most of the work. The thing you want is already open when you arrive.

CarPlay support
• ChatGPT works in Apple CarPlay on supported iPhones.
• From the CarPlay screen, you can start a new voice conversation, continue a recent chat, or open a conversation inside one of your projects.
• There’s also a setting to start voice automatically when CarPlay connects.

Set this up before you drive and only use it where it’s legal and safe. Done right, it can turn commute time into planning, outlining, or light research time.

Language and speaking style controls

Voice is more flexible than it first appears:

• You can set your default language in Settings → Voice.
• During a call, you can ask it to switch languages on the fly.
• You can also ask it to slow down, speed up, or change its tone (more formal, more casual, more concise, etc.).

There isn’t a precise speed slider yet, so you guide it by asking for adjustments in natural language.

Turning speed into an advantage

The real opportunity with the new ChatGPT voice isn’t just that it feels futuristic. It’s that talking is faster than typing, and now your voice can trigger real work: documents, research, structured data, and plugin-driven tasks.

If you pair that speed with clear workflows—like “every commute, I talk through one article outline” or “every walk, I dictate a research summary”—you can produce more and better work without needing more screen time.

Just remember the main constraints: plugin approvals happen on-screen, Live has usage limits, and it’s best at one clear job per call. Once you build around those rules, the new ChatGPT voice becomes much more than a talking search bar—it becomes a practical, everyday assistant you can actually rely on.