Skip to content
← All articles

Is Wispr Flow Safe, What Its Own Pages Say About Your Voice

Where Wispr Flow sends your voice, which privacy switches start on, and how to turn them off, all from Wispr Flow's own pages and help centre.

By Nelson, maker of Aurator

On a Mac, every word you dictate in Wispr Flow goes to its servers to become text, and on the Free and Pro plans the switch that lets Wispr use your dictations to improve its models starts on. You can turn that switch off, and several others, but not the trip to its servers. So whether Wispr Flow is safe for you depends on what you dictate and which switches you turn off.

I make Aurator, so please read this knowing that. Everything I say about Wispr Flow comes from its own data controls page, privacy policy, pricing page and help centre, all checked on 9 October 2026, with the links at the end. Where its pages say something, I quote or follow them closely. Where they say nothing, I say nothing.

Where your voice goes when you use Wispr Flow

Wispr Flow's data controls page says it plainly. Transcription always occurs on the cloud. Its help centre says the same thing in other words. Its security overview says Wispr Flow processes dictated audio in the cloud to turn speech into text, and its getting started article says voice transcription needs internet access.

So when you press the key and speak, the sound of your voice travels from your Mac to Wispr's servers, becomes text there, and comes back. Its security overview says Wispr processes and stores customer data in the United States, and that the service is not end-to-end encrypted in the strict sense, because the service must process audio to produce a transcript. Your data is encrypted while it travels and while it is stored on Wispr's servers.

That is the first fact to know. There is no setting in Wispr Flow that keeps your voice on your Mac.

Does Wispr Flow train on your data

It can, unless you turn it off. The setting is called Improve the model for everyone, in Settings, Data and Privacy. When it is on, the data controls page says your audio, transcript and edits may be used to evaluate, train or improve AI models, by Wispr.

Here is the part many people miss. Wispr's security overview says this setting is on by default for Free and Pro, with the choice shown during onboarding. Its help centre says the same about two switches, Improve the model for everyone and Dictation cloud storage, which default to on unless restricted. On Enterprise plans, training is off and administrators cannot turn it back on.

Even with the switch off, the help centre says turning it off does not stop cloud processing. It says your audio, transcripts and edits are then held back from model improvement, but Flow still uploads dictation metadata. That metadata is the word count, duration, app, URL, timestamp, processing delay, app version, microphone, detected language and edit statistics.

Wispr's pages also say two things in its favour, and they are facts you should have. Wispr says it never sells your data. And it says its third-party AI providers cannot use your data to train their own models, whatever your setting.

What Wispr Flow keeps on its servers

Two settings and two exceptions decide this.

  1. Dictation Cloud Storage. Wispr's data controls page says this decides whether Wispr stores your transcripts, audio and dictation history on its servers. The help centre says it defaults to on unless restricted, and that signing out resets it to on, so check it again after you sign back in.
  2. Local data storage. This decides what stays on your own computer. Store data locally is the default. You can choose to delete local history every 24 hours, or never store it, and the help centre says either choice hides the voice profile features.
  3. Feedback. If you rate a dictation, use the thumbs, or report one, Wispr collects data from that session. The data controls page says it may include the audio recording, the transcript, your edits and the app, and that Wispr uses it to develop and train its AI models. The security overview says Wispr keeps a reported transcript, and that no privacy setting removes it.
  4. Meeting notes. If you use Wispr's notetaker, its security overview says transcripts and summaries are stored in Wispr's cloud, with no customer setting to opt out of that storage, and meeting audio is kept for 7 days.

Separately, the data controls page says Wispr may collect usage statistics such as the number of words you have dictated, regardless of your data controls.

Does Wispr Flow read your screen

It reads some of the text around you. The data controls page says Wispr always uses the name of the app you dictate in to set the tone, and takes into account the surrounding text box content for capitals and punctuation. Its privacy policy says it may collect the app you dictate in, and the help centre lists the URL among the metadata Flow uploads.

A setting called Context Awareness goes further. When it is on, relevant text from the active app window may be used to improve transcription accuracy, and the privacy policy describes this as limited, relevant content from the app in use, such as the text on the screen. You can turn it off in Settings, Data and Privacy.

A setting called Auto-add to Dictionary watches the text box after a paste, so that when you correct a word it adds your spelling to your dictionary. You can turn that off in Settings, Personalization.

Does Wispr Flow work offline or run locally

Not on a Mac. Its help centre says voice transcription needs internet access, and the data controls page says transcription always occurs on the cloud. Its hands-free article says being offline does not stop dictation on a computer. Flow warns you, and if transcription fails it saves the audio so you can try again later. The same article mentions a Backup or Offline Dictation option on iPhone for some accounts, and the pages I read do not say how it works.

How to set up Wispr Flow for the most privacy

If you already use Wispr Flow, these are the steps its own pages describe. Open Flow from the menu bar, then Settings, Data and Privacy.

  1. Turn off Improve the model for everyone. Wispr's help centre says to do this before recording anything confidential.
  2. Turn off Dictation cloud storage, if your account shows it.
  3. Turn off Context awareness if you do not want text from the app window used.
  4. Choose a Local data storage option if you do not want history kept on your computer.
  5. In Settings, Personalization, turn off Auto-add to Dictionary if you do not want the text box watched after a paste.
  6. Do not use Report or the thumbs on a dictation that holds anything private.
  7. After you sign out and back in, open Data and Privacy again and check the cloud storage switch.

Even with all of these off, your voice still goes to Wispr's servers to become text, and the metadata above still goes too. That is how the app is built.

A different design, voice to text on your Mac

Aurator starts from the other side. It turns your speech into text with Apple's speech model on your Mac, and the cleanup and the coach run on your Mac too. There is no account and no server of mine that could receive what you say. Your History and your recordings stay in one folder on your Mac, and recordings are kept 30 days unless you choose otherwise.

Wispr Flow, Free and ProAurator
Where speech becomes textWispr's servers, alwaysYour Mac, unless you turn on Cloud
Training switch at the startOn, you can turn it offAurator never trains on your words
HistoryOn your computer, and on Wispr's servers while cloud storage is onOn your Mac only
Works with no internetNoYes with Cloud off, once the speech model for your language has downloaded
AccountNeededNone
PriceFree plan, or Pro by subscriptionUS$29 once for one Mac

With Cloud off, Aurator goes online for two things only. It checks for updates, and about once a day it checks your licence. Neither carries your voice or your words. It never listens in a password field.

Cloud is there if you want it, and it is off until you turn it on in Settings with a key from your own Deepgram or Anthropic account, kept in your Mac's Keychain. Then your audio goes to your own Deepgram account and your text to your own Anthropic account, under your own terms with them, never to me. I wrote more about reading these policies in What a Privacy Policy Actually Promises About Your Voice, and the full comparison is in Why Aurator Is the Better Wispr Flow Alternative for Most Mac Users.

When Wispr Flow still fits

  1. Other devices. Wispr Flow runs on Mac, Windows, iPhone and Android. Aurator is for the Mac only, on macOS 26 or later.
  2. More languages. Wispr Flow says over 100. Aurator has ten.
  3. Teams with controls. Its Enterprise plan keeps training off for everyone and gives administrators settings to lock. Aurator is a dictation app for one person.
  4. Audits for a security review. Wispr's Trust Center offers a SOC 2 Type II report and an ISO/IEC 27001:2022 certificate.

What I would tell you to do

If what you dictate is private, your clients, your family, your money, start with an app that keeps your voice on your Mac. Give Aurator a week of your real work, and read tryaurator.app/privacy to see exactly what goes online and what does not. Aurator is at tryaurator.app, and you can download it here. Seven days free, with every feature, then US$29 once.

And if you think I got something wrong about Wispr Flow, please write to me at hello@tryaurator.app. I would really like to correct it.

Sources, all checked on 9 October 2026