Dictation app privacy comparison: where your voice actually goes

The single biggest privacy difference between dictation apps is whether your audio leaves the Mac. Apple Dictation (for many languages on Apple silicon), Superwhisper's offline models and Sayburst transcribe on the device; Wispr Flow, Microsoft's Dictate and browser voice typing send your speech to a server, and some also send text from your screen.

We make Sayburst, which is a local app, so read this with that in mind. Every competitor statement below comes from the vendor's own privacy or product page, checked November 2026, and links to it.

#The comparison

Sources: Apple's Ask Siri, Dictation & Privacy, Microsoft's Dictate in Outlook, MDN's SpeechRecognition notes, Wispr Flow's data controls, and superwhisper.com. "Not stated" means we couldn't find it on the official page, not that the answer is bad.

#What each vendor says

Apple Dictation. Your Mac shows in System Settings › Keyboard whether dictation is processed on the device. When it isn't, Apple says what you say is sent to and processed by Apple, tied to a random device identifier that rotates multiple times per hour rather than your Apple Account. Audio is only kept if you opt in to Improve Siri & Dictation, and Apple may keep related request data for up to two years.

Microsoft Dictate (Word, Outlook, PowerPoint). Microsoft says your speech is sent to Microsoft and used only to give you the text, and that the service doesn't store your audio or the transcribed text. It needs a connection and a Microsoft 365 subscription.

Browser voice typing (Google Docs, voice typing websites, many extensions). MDN notes that in browsers like Chrome, speech recognition on a web page sends audio to a web service, so it doesn't work offline. Extensions such as Voice In also say they need an internet connection.

Wispr Flow. Its data controls page says "Transcription always occurs on the cloud." You choose whether your audio, transcripts and edits can be used to improve models (the old Privacy Mode is now "Improve the model for everyone"), whether dictations are stored in its cloud, and whether text from the active app window is used for context. It states SOC 2 Type 2 compliance.

Superwhisper. Its site says it works offline, with offline models running best on Apple silicon Macs, and it also offers cloud AI models and your own API keys on Pro. Which you use decides where your words go.

Sayburst. Speech recognition always runs on the Mac and Sayburst never uploads audio. After the model downloads once, it works with no internet. If you connect a cloud writing provider with your own API key, only the transcribed text (plus any context you switch on) goes to that provider, never audio, and Sayburst shows exactly what will be sent before you enable it.

#Check your own setup in five minutes

1. Run the Wi-Fi test. Turn Wi-Fi off and dictate a sentence into Notes. If text appears, recognition is local. If it fails or hangs, your audio was going to a server.
2. Check Apple's indicator. Open System Settings › Keyboard › Dictation and read the note about where your language is processed.
3. Check the improvement opt-in. Go to System Settings › Privacy & Security › Analytics & Improvements and see whether Improve Siri & Dictation is on. Turn it off if you'd rather Apple didn't keep audio samples.
4. Open your dictation app's privacy settings. In Wispr Flow that's Settings › Data and Privacy: look at model improvement, cloud storage of dictations, and Context Awareness. In any app, look for anything that reads the screen, clipboard or selection.
5. Look at history. Most apps keep a dictation history. Decide how long that should be. In Sayburst, retention runs from forever down to one day, or off, and recordings are deleted after transcription unless you choose to keep them.

#The part dictation can't make private

There's always a second step: the app you dictate into. Local recognition keeps your voice off the internet, but a Gmail draft is still in Gmail, a Slack message is still in Slack, and a Google Doc is still a Google document. The local vs cloud dictation guide separates the stages in more detail.

And no dictation app makes your organisation compliant with anything on its own. If you handle patient, client or legal material, your employer decides which tools are allowed.

#When the built-in option is private enough

If you're on Apple silicon and your language runs on-device, Apple Dictation is already a private choice and costs nothing. Confirm with the Wi-Fi test and leave Improve Siri & Dictation off.

Where it falls short is everything after recognition: no cleanup, no vocabulary, and it stops after 30 seconds of silence. That's when people reach for an app, and often a cloud one, trading privacy for polish.

#Polish without uploading anything

Sayburst is built so you don't have to make that trade. Hold fn, talk, let go. Recognition runs on the Neural Engine, cleanup runs on the Mac, and the optional Sayburst Writer (a 2.5 GB on-device model) turns talk into finished writing without an account, an API key or a connection.

Here's a note for an HR file, the kind of thing you wouldn't want on someone else's server:

"Note for Sam's file. Um, we discussed the missed deadlines on the Halvorsen account. Sam said the, the delays came from the handover in March, sorry, in April. Agreed next step, uh, weekly check-ins for six weeks"

What lands in the document:

Note for Sam's file. We discussed the missed deadlines on the Halvorsen account. Sam said the delays came from the handover in April. Agreed next step: weekly check-ins for six weeks.

The fillers, the stutter and the wrong month are gone, and none of it touched a network. Context such as selected text, clipboard or window text is off by default, and if you turn any of it on, it goes only to the writer you picked, which can be the local one.

Our privacy page has the details, and offline dictation on Mac shows how it works with no connection at all. If you're weighing Wispr Flow specifically, see is Wispr Flow private? and Sayburst vs Wispr Flow.

Sayburst needs an Apple silicon Mac with macOS 14 or later and costs $29 once. It's Mac-only, so on Windows or a phone, the cloud-based tools above are what's available, and their privacy settings are worth the same five-minute check.

Questions people ask

Which dictation apps keep audio on the device?

On a Mac, Apple Dictation processes many languages on-device on Apple silicon, Superwhisper offers offline models, and Sayburst always transcribes on the Mac. Wispr Flow, Microsoft's Dictate and browser-based voice typing send speech to a server.

Is Wispr Flow private?

Wispr Flow's data controls page says transcription always happens in the cloud. You can control whether your data helps train models, whether dictations are stored on its servers, and whether text from the active window is used for context.

Does Apple keep my dictation recordings?

Apple says it doesn't keep audio unless you opt in to Improve Siri & Dictation. When speech isn't processed on your Mac, it's sent to Apple and tied to a random identifier that rotates several times an hour rather than your Apple Account.

How can I tell if my dictation app works offline?

Turn off Wi-Fi and dictate a sentence. If the text still appears, recognition is happening on your Mac. If it fails or stalls, your audio was going to a server.