Learn · Privacy
There is no single answer, because “Mac dictation” covers three different things that behave three different ways. Apple’s built-in Dictation can be processed entirely on your Mac, or sent to Apple’s servers, and which one you get depends on your chip and your language. Local-model apps keep the audio on the machine. Dictate sends it to our servers, transcribes it there, and discards it. Dictate is our app.
The short version
Read the one that matches what is installed on your Mac. If more than one is installed, read both, because they do not behave alike.
Built into macOS
Depends on your chip and your language.
On an Apple silicon Mac dictating in a supported language, Apple says the processing happens on the device and no connection is needed. On an Intel Mac, or in a language without on-device support, the audio is processed on Apple’s servers instead.
Superwhisper · VoiceInk · MacWhisper
In local mode, nothing leaves the Mac.
These run a speech model on your own machine. VoiceInk uses local Whisper models by default and offers cloud providers as an option, so the mode you are actually in is worth checking rather than assuming.
$8 a month
The audio leaves, is transcribed, then is discarded.
Audio goes over TLS to our servers, comes back as text, and is dropped when the request completes. History keeps the transcript and metadata, never the recording. Dictate is not local and it is not on-device.
Apple’s split
Apple documents the same feature two ways, and both are true. On an Apple silicon Mac, dictating in a supported language, the processing happens on the device and no connection is required. On an Intel Mac, or in a language that is not supported for on-device processing, the audio is processed on Apple’s servers.
The supported-language list lives on Apple’s feature availability page and it changes between releases, so it is worth checking against the macOS version you are running rather than against a blog post.
Two footnotes worth knowing
Search fields are an exception Apple names.
Dictating into a search box can send the resulting text on to the search provider, even where the speech itself was handled on the device. On-device processing is a statement about the audio, not about everything that happens to the words afterwards.
We are not going to paraphrase Apple’s policy.
What Apple does with audio its servers do process is described in Apple’s own privacy documentation, and that is the version to read. We can tell you which path your Mac takes. We cannot tell you what happens at the far end of Apple’s.
The connection side of the same question is worked through in does Mac dictation work offline.
Our side of it
Four stops, and the fourth is the one that matters. Nothing here is a promise about intent; it is the route the recording takes.
Recorded on your Mac for as long as you hold the key. Releasing it ends the recording.
Sent over TLS to the Dictate API, forwarded to the transcription provider, returned as text.
Placed where your cursor already was: Accessibility API first, paste second, clipboard last.
The audio is dropped when the request completes. Nothing is retained to replay.
History is account-backed and holds transcript text plus metadata such as duration, model, and timestamp. It never holds the recording.
The mechanics behind that route, including the Secure Enclave key each request is signed with, are written out on the security page. What we collect on the website and in the app is on the privacy page. Both are meant to be read before you buy, not after.
The ledger
The row printed in black is the row we lose. It is printed that way on purpose, because for a reader asking this question it is probably the only row that counts.
| Dimension | Dictate | Local-model apps |
|---|---|---|
| Where the model runs VoiceInk ships local Whisper models by default and offers cloud providers as an option, so "local app" describes the mode, not always the setting you are on. | Our servers | Your Mac, in local mode |
| Audio never leaves your Mac This is the row that decides it if your standard is that nothing may leave the machine. Superwhisper, VoiceInk, and MacWhisper win it in their local modes. Dictate loses it, and no amount of TLS changes that. | No | Yes |
| Transcribes with no connection Dictate needs a network to transcribe. Only the signed entitlement ticket is checked locally, so a dead connection never signs you out of the app you paid for. | No | Yes |
| Audio kept after transcription Dictate discards the recording when the request completes. File-first tools such as MacWhisper work from audio files you keep on disk, which is what they are for. | No | Varies by app |
| Recordings stored in history Dictate history holds transcript text plus metadata: duration, model, timestamp. There is no audio in it. | No | Varies by app |
| Transcript text stored on a server Account-backed history is why the text follows you between Macs. If you would rather nothing at all sat on someone else’s server, that is a fair reason to choose a local tool. | Yes, in your account | No, on your Mac |
| Transport | TLS to the Dictate API | None required in local mode |
| Model to choose, download, and maintain This is the trade running the other way. Managed transcription is the reason there is nothing to configure, and it is also the reason the audio has to leave. | None | Yours to pick and keep current |
| Pricing VoiceInk licenses started around $25 for one Mac when checked in mid 2026, and Superwhisper’s lifetime tier has been reported at about $249.99 as of mid 2026. Both can change. | $8/month, one plan | One-time or lifetime, varies |
Cells marked “Unknown, see vendor docs” mean we couldn’t verify the capability from public sources at time of writing. We’d rather say “we don’t know” than guess.
The honest call
Some work has a rule attached, and the rule is absolute: the recording does not leave the machine, full stop. If that is your situation, a local-model app is the correct tool and Dictate is not. Superwhisper, VoiceInk, and MacWhisper all run the model on your Mac, and Apple’s built-in Dictation does the same on Apple silicon in a supported language, at no cost.
What Dictate offers instead is a managed path with the handling stated plainly: audio over TLS, transcribed on our servers, discarded when the request completes, and history that stores text rather than sound. That is a reasonable trade for a great deal of work. It is not a substitute for a rule that says nothing leaves the machine, and we will not pretend it is.
We hold no compliance certification and make no regulatory claim on this page or anywhere else. If your obligations are set by someone other than you, ask them the server-side question first and buy nothing until you have the answer. The same reasoning, aimed at practice work, is on dictation for lawyers.
If you are evaluating one
Ask these of every app on your shortlist, Dictate included. An app that cannot answer all four in plain sentences has told you something.
01
Not “is it encrypted”, not “is it secure”. Does it leave. Dictate’s answer is yes, over TLS. A local tool in local mode answers no.
02
Dictate discards the audio when the request completes and keeps transcript text and metadata in your account. An app that keeps recordings should say so in the same sentence it says “private”.
03
Dictate forwards audio to a transcription provider from our API. Any managed app has a chain like that. The question is whether it is written down where you can find it.
04
Dictate stops transcribing, because there is nothing local to fall back to. It does not sign you out: the entitlement ticket is signed and checked on your Mac.
If server-side is fine with you
Flat $8 a month, monthly only, no tiers. English only, and live speech only: Dictate does not transcribe existing audio files. Unused subscriptions are refunded in full; once anything has been transcribed, that month is not refundable.
Asked constantly
Where the audio goes, what is kept, and who wins the strict version of the question.
It depends which dictation you mean. Apple’s built-in Dictation is processed on the device on an Apple silicon Mac in a supported language, so the audio does not leave. On an Intel Mac, or in a language without on-device support, the audio is processed on Apple’s servers. Dictate, the app this site sells, transcribes on our servers over TLS and then discards the audio.
Sometimes. On an Apple silicon Mac dictating in a supported language, Apple says processing happens on the device and no connection is required. On an Intel Mac, or in a language that is not supported for on-device processing, the audio goes to Apple’s servers. Apple keeps the current language list on its macOS feature availability page and it changes between releases, so check it against the version you are running.
No. Audio travels over TLS to our servers, is transcribed there, and is discarded when the request completes. Transcription needs a network connection. The only part that works offline is the signed entitlement ticket that proves your subscription, which is verified on your Mac.
No. The audio is discarded when the transcription request completes. History is account-backed and stores transcript text plus metadata such as duration, model, and timestamp. There is no recording in it to play back, export, or hand over.
If the standard is that the audio must never leave the machine, a local-model app is the honest answer: Superwhisper, VoiceInk, and MacWhisper run the model on your Mac in their local modes. Apple’s built-in Dictation also qualifies on an Apple silicon Mac in a supported language. Dictate does not qualify, because it transcribes on our servers.
That is a decision for you and whoever writes your rules, and it turns on one question: whether server-side processing is permitted at all. What we can state exactly is the mechanism. Audio goes over TLS, is transcribed, and is discarded; history keeps text and metadata, never recordings. If your rules say audio may not leave the machine, use a local-model tool instead.
