Latest in AI
Granola Voice Typing: Free Unlimited Dictation on Mac, With a Privacy Tradeoff
Granola is moving beyond meeting notes into everyday writing, where context helps but changes what a dictation tool needs to read.
Granola launched free, unlimited voice typing on October 29, letting Mac users dictate into other apps. The useful change is that its meeting assistant can now help with the writing between meetings, although the contextual features deserve a closer look before using them with sensitive work.
What Granola Voice Typing includes
The official announcement introduces voice typing as part of Granola rather than a separate subscription. The product page says it works across apps, adds punctuation and paragraph structure, and is available to free users without a word limit.
At launch, voice typing is available on Mac. Windows, iOS and Android versions are listed as coming soon. That distinction matters because availability of Granola's meeting product on a platform does not establish availability of this new feature there.
Granola also says it can use meeting context to recognize names and terminology. For someone writing follow-ups after several technical calls, getting the same project names right could be more valuable than simply transcribing faster.
The company's four-times-faster claim compares 160 spoken words per minute with 40 typed words per minute. It is a general speaking-versus-typing comparison, not an independent benchmark of how quickly someone finishes an edited document.
Local history does not mean local processing
The dictation privacy documentation makes an important distinction: dictation history stays on the Mac, but speech processing is not entirely local. Audio, dictionary entries and the chosen language go to transcription and AI providers.
With screen context enabled, limited text from the destination app is sent too. Granola reads through macOS Accessibility rather than taking screenshots. The context can include surrounding text, a window title and nearby conversation content; password fields are skipped.
Screen context can be disabled, although that also turns off styles and app-specific formatting. The documentation says transcription and AI providers do not train their models on dictations.
By default, local text history remains until deleted. Failed dictations can retain encrypted audio locally for retrying, with that audio deleted after 24 hours. Successful dictation audio is not kept. Users can instead select 24-hour history retention or no local storage.
The useful test is correction effort
For an existing Granola user, the appeal is straightforward: another writing step moves into a tool already used during the working day. Free unlimited usage also removes the need to estimate a monthly word allowance before trying it.
But raw transcription speed is only part of the workflow. A tool can produce text quickly and still leave the user correcting terminology, restoring a sentence's intended meaning or changing an overly polished tone. The useful comparison is the time from speaking to sending an accurate message.
For confidential work, the corresponding question is whether the extra context is appropriate to transmit. Turning it off may be the right choice even if the output needs more editing. Context is a product advantage and a data decision at the same time.
Granola's launch is worth watching because it connects meeting context to everyday composition. Whether that connection saves work will depend less on the length of the transcript than on how little the user has to repair afterward.
I’m Paras. I work at the intersection of AI, engineering and distribution.
I help technical founders explain product changes clearly and turn useful launches into content developers actually read.