Voice in ChatGPT Work
The least-covered part of the Work release. On desktop you can talk to the agent, interrupt it mid-sentence, and tell it to start or coordinate tasks — using whatever tools and permissions the selected experience already has.
VERIFIED 2026-08-23 · OPENAI HELP CENTERWhat it actually is
ChatGPT Voice adds a live voice interface to Work and Codex in the ChatGPT desktop app. You speak, you can interrupt naturally, and you can coordinate tasks using the tools and permissions available to the experience you selected.
That last clause is the whole point. This is not dictation into a chat box — Voice inherits the selected experience's tool access, which is what lets a spoken instruction turn into an agent actually starting work.
| Voice in Work / Codex | Ordinary Chat Voice | |
|---|---|---|
| Where | Desktop app, macOS & Windows | Standard ChatGPT surfaces |
| Can start tasks | Yes | No |
| Tool access | Inherits the selected experience | Chat's own |
| Settings, limits, retention | Per the selected experience | Separate, see ChatGPT Voice |
| Billing | Connected time metered separately; tasks hit the shared pool | Chat Voice limits |
Source: OpenAI Help Center, ChatGPT Work and Codex.
Turning it on
- Open the ChatGPT desktop app and choose Work or Codex.
- Select the Voice control, or use a shortcut you configured in Settings. There is no default shortcut documented — if you want one, you set it.
- Allow microphone access when prompted. Computer context may additionally require Screen & Audio Recording and Accessibility permissions.
- Begin speaking. You can interrupt naturally, follow the live text in the chat, and ask Voice to start or coordinate tasks.
The permission prompts are the real setup step. Microphone alone gets you talking; screen and accessibility permissions are what let it use computer context. If Voice seems oddly unaware of what is on screen, that is the permission you skipped.
Limits worth knowing before you rely on it
One conversation at a time. Only one Voice conversation can run at a time — you cannot have Voice supervising two long tasks in parallel.
Two meters are running. Connected Voice time is metered separately where flexible pricing applies, and any task you start through Voice draws from the same shared agentic usage and credit pool as Work and Codex. Leaving a Voice session connected while you think is not free.
Desktop only, eligible accounts only. macOS and Windows. Work on web and mobile is a separate thing, and Codex is not selectable there at all.
Where it earns its place
Voice suits the supervision half of agentic work rather than the briefing half. Writing a good brief rewards precision — outcome, context, format, constraints — and typing is better at precision. But answering a check-in, redirecting a task that has gone sideways, or approving an action mid-run are all short, reactive moments where reaching for the keyboard is the friction.
The practical pattern: type the brief, then keep Voice for the conversation that happens while it runs.
Common questions
Can I use Voice with Work on my phone?
No. ChatGPT Voice in Work and Codex is a desktop capability on macOS and Windows, for eligible accounts. Work on web and mobile is separate.
Is Voice in Work the same as ChatGPT Voice in Chat?
No. Ordinary Chat Voice has its own settings, limits, retention and data controls. Voice in Work and Codex uses the tools and permissions of the experience you selected, which is what lets it start and coordinate tasks.
What does Voice cost?
Connected Voice time is metered separately where flexible pricing applies. Tasks you start through Voice draw from the same shared agentic usage and credit pool as Work and Codex.
Can I run two voice conversations at once?
No. Only one Voice conversation can run at a time.