Gemini for Mac is expanding beyond a desktop chatbot with a new system-wide voice feature that works inside other Mac applications. By pressing and holding the Fn key, users can speak naturally while working in an email, document, browser, messaging app or another text field.
Google calls the feature Speak to Window. Releasing the key submits the request, and Gemini places the response at the active cursor or applies an instruction to selected content. The same interaction can serve as ordinary dictation, a rewriting tool or a screen-aware assistant depending on what the user says and which permissions have been enabled.
The update changes the position of Gemini on macOS. The original app provided quick access through Option + Space and allowed users to share a window for additional context. Voice input now reaches beyond the Gemini interface, bringing the assistant into the application already being used.
Gemini for Mac Turns the Fn Key Into an AI Shortcut
Pressing and holding Fn begins recording immediately. A user can dictate a sentence, paragraph, email or longer passage without opening the main Gemini window first.
Gemini does more than convert each spoken word directly into text. Google says intelligent dictation removes filler words, recognizes corrections made in the middle of a sentence and formats the output into cleaner paragraphs or structured content.
Someone could begin a message by saying one phrase, correct the wording aloud and continue without manually returning to the mistake. Gemini is designed to interpret the correction rather than place both versions into the document.
This creates a different experience from basic macOS Dictation. Apple already allows users to enter text through the microphone, but Google is combining transcription with a generative model that can reorganize the spoken material before inserting it.
The distinction may be useful for people who think more quickly through speech than through typing. A loosely spoken explanation can become a more organized paragraph instead of a literal transcript containing pauses, repeated words and abandoned sentences.
Longer sessions are available through a hands-free option. Double-pressing the Fn key begins recording without requiring the key to remain held. Pressing it again ends the recording and sends the request.
Dictation and Commands Share the Same Interface
Speak to Window determines whether the user is entering text or asking Gemini to perform a task.
A person composing an email can dictate the complete message. The same person can select an existing paragraph and ask Gemini to make it shorter, more formal or easier to understand.
The response is inserted directly into the active application rather than returned only inside a separate chat. This reduces the repeated copying and pasting usually required when an AI tool is used alongside writing software.
Gemini can also interpret formatting instructions. Spoken requests can turn rough notes into paragraphs, reorganize information or produce a structured version of material already visible on the screen.
The assistant does not gain unrestricted control over every Mac application. Its behavior depends on the active text field, selected content and permissions granted through macOS. Some applications may also handle inserted text differently or limit the operations available to external accessibility tools.
Google is initially rolling out the new voice feature in English. The Gemini app itself supports additional languages and regions, but Speak to Window availability may not match the complete service at launch.
Screen-Aware Commands Require Permission
Basic intelligent dictation is available by default, but commands that analyze and modify information on the screen require an additional setting.
Google places this option under the Speak to Window section of the Gemini app settings. Once reasoning is enabled, the assistant can distinguish between straightforward voice-to-text requests and instructions that require understanding selected or visible material.
A user could highlight a paragraph and ask Gemini to rewrite it without first describing the text. Selecting a document, PDF or image can also provide context for a spoken request to summarize or extract information.
Gemini can generate an image from a voice description and place the result into a supported workspace. That capability extends the feature beyond dictation, though the usefulness of direct insertion will depend on the application and document format.
For wider access to window content, macOS requires the user to grant Accessibility permission to Gemini. Screen Recording or related privacy permissions may also appear according to the type of context being shared.
These permissions deserve attention because they can allow the app to read information from the active window or interact with interface elements. Users handling confidential business documents, personal communications or financial information should decide when screen context is appropriate rather than enabling every option without review.
Google says Gemini uses the foreground window as context when window sharing is active. Access to local files is limited to folders or items deliberately shared with the app, while the broader Gemini Spark agent can connect only to directories selected by the user.
Selected Content Makes Commands More Precise
The most controlled way to use screen-aware commands is to select the exact material Gemini should process.
A user can highlight text, hold Fn and ask for a correction, translation, summary or change in tone. Gemini receives a narrower target instead of interpreting the complete window.
Google advises users to deselect the information after Speak to Window begins recording when they do not want the selected material replaced directly. That interaction may require some practice because the same selection can function as context or as the destination for Gemini output.
This system gives the feature considerable flexibility but also introduces the possibility of accidental replacement. Important text should be saved before experimenting with commands in applications that lack strong version history or undo support.
Screen awareness also does not guarantee accuracy. Gemini can misinterpret a document, overlook a visual detail or rewrite language in a way that changes its intended meaning. Inserted responses should be reviewed before an email is sent, a file is shared or a final document is submitted.
A More Direct Rival to Apple Intelligence
Gemini for Mac is entering an area where Apple is also integrating generative tools into writing and system workflows. Apple Writing Tools can proofread, summarize and rewrite selected text across supported applications, while Siri and Apple Intelligence provide varying levels of screen and personal context.
Google is taking a more explicit shortcut-driven approach. Holding Fn establishes one consistent voice action across the desktop, regardless of whether the user wants dictation, rewriting or assistance with visible content.
The Mac app remains limited to Apple silicon models running macOS Sequoia 15 or later. Google also requires at least 8GB of memory, a personal Google Account or an approved work or school account, and an internet connection.
Users can change the Fn behavior and other keyboard shortcuts inside Gemini settings. The existing Option + Space shortcut continues to open the compact chat window, while Option + Shift + Space opens the full Gemini interface.
The new voice layer gives Google a way to participate in Mac workflows without controlling macOS itself. Its success will depend less on the quality of isolated chatbot answers and more on whether Speak to Window can reliably place useful text in the correct location without interrupting the work already underway.
