Skip to content

Hotkeys

Voice Mode uses three independent hotkeys.

Hold to record, release to transcribe. The transcript is pasted into the focused app. Free, always available. Default: Right Option.

Select text in any app, then hold this hotkey and speak — the model rewrites the selection in place. Also runs the AI Correction pass on fresh dictation when used without a selection. Free. Default: Shift + Right Option.

Hold to record, release to send the transcript to the assistant backend instead of pasting. The reply is spoken back (TTS) and, for drafted content, can also be pasted. Pro-only. Default: Command + Option.

All three hotkeys live in one place: Settings → General → Hotkeys. The General row shows the current bindings inline (something like Dictation: Right ⌥ · AI Writing: ⇧Right⌥ · Assistant: ⌘⌥); click Change… to open a single sheet that lets the user rebind any of the three.

The first time a user changes a hotkey or launches the app, macOS will prompt for Accessibility and Input Monitoring permissions — these are required for Voice Mode to detect global key presses. Without them, the hotkeys will silently do nothing.

In addition to the assistant hotkey, the assistant can also be triggered by speaking a trigger word at the start of a normal dictation (see assistant.md). This lets users stay on the dictation hotkey and switch into assistant mode just by saying, e.g., “hey Claude, …”.

  • Pick keys your other apps don’t already use. Right Option is the default because almost nothing else on macOS binds it; remapping to F-row keys or modifier combinations works too, but check that your IDE or chat app isn’t already grabbing them.
  • Use the trigger word for occasional assistant turns. If most of the user’s voice work is dictation and only sometimes the assistant, sticking with one hotkey plus a trigger word (“hey Claude, …”) is less to remember than juggling two.
  • A dedicated assistant hotkey is worth it for power use. Users who lean heavily on the assistant prefer a dedicated key — no trigger word to remember, faster to start, less chance of the trigger phrase landing in the dictation pipeline by accident.