Google adds voice-powered editing and summaries to Gemini for macOS

Google is rolling out a new voice mode for the Gemini app on macOS that works directly inside other desktop applications. Users can hold the Fn key, speak naturally and place the resulting text at the active cursor. The feature is available globally in English to all Gemini for macOS users, with more languages planned.

By default, the tool acts as intelligent dictation. Google says it removes verbal fillers such as “ums” and “ahs,” follows corrections made halfway through a sentence and formats the spoken input before inserting it. That makes it different from literal speech-to-text: the system is intended to produce a cleaner written version of what someone meant rather than a word-for-word transcript.

Users can separately enable Gemini reasoning in the app’s settings. With that option switched on, Gemini can use information visible on the screen to perform broader tasks. Google gives examples including selecting local documents and asking for a summary, highlighting notes and turning them into an executive briefing, or requesting a rewrite in a different tone. The same voice interface can also create images or request edits to a visual shown on the desktop.

The distinction between default dictation and optional reasoning matters. Simple text entry requires less context, while actions involving files, images or screen content expose more information to the assistant. Google’s announcement does not provide detailed new privacy controls for this feature. Users and organisations should therefore review the app’s permissions and avoid selecting confidential material unless its handling fits their policies.

For writers, employees and people who find typing difficult, the practical gain is speed. A spoken draft can appear in an email, document or chat without first moving through a separate chatbot window. Corrections and formatting may also reduce the editing normally required after conventional dictation. Accuracy still needs checking, especially for names, numbers and specialist vocabulary.

For the AI market, the release shows assistants moving from destination apps toward an input layer that follows people across their existing software. That puts Gemini in closer competition with dedicated voice-writing tools and system-level assistants. The useful test is not whether the transcription sounds polished in a demonstration, but whether it remains accurate, predictable and appropriately private across everyday applications. Mac users can download the app now, although Dutch voice support has not yet been announced.