Voice typing on Mac: what your dictation app can see, and how to keep it private

The short answer

Voice typing is private only if two things stay on your Mac

The first is your voice: the app should turn speech into text on your Mac rather than on a server. The second is everything else the app reads while you talk — the text around your cursor, what is on screen, your clipboard. A growing number of apps collect that “context” and send it with your words to a cloud AI. Wispr Flow's help page, for example, says its context is “on by default on Mac and Windows” and can include “a screenshot”.

So the useful questions for any dictation app are: where does recognition run, what does it read besides my voice, where does that go, and can I switch it off? The rest of this page answers those for macOS itself, Apple Dictation and six popular apps, including ours.

What is at stake

The four things a dictation app can touch

macOS permissions

What each permission actually allows

macOS makes every app ask before it gets these. They live in System Settings > Privacy & Security, and Apple's own descriptions are the best guide to what you are granting.

Source: Apple, Privacy & Security settings · Apple, menu bar indicators

Accessibility is the one to think hardest about. Apple warns that “If you give apps access to your Mac, you also give them access to your contacts, calendar, and other information, and are subject to their terms and privacy policies, and not the Apple Privacy Policy.” Source: Apple Support A dictation app genuinely needs it to type for you, so the question is not whether to grant it but whether you trust what that app says it does with it.

The built-in option

Apple Dictation: local sometimes, and one setting to check

Apple tells you to check rather than assume: “You can check Keyboard settings to see whether your voice inputs and transcripts for general text Dictation (for example, composing messages and notes, but not dictating in a search box) are processed on your device and not sent to Siri servers.” Source: Mac User Guide

When dictation is not on-device, Apple explains what happens instead: “the things you dictate are sent to and processed on the server, but will not be stored unless you opt in to Improve Siri and Dictation.” Requests are tied to “a random, device-generated identifier that rotates multiple times per hour and is not tied to your Apple Account or email address.” One line matters for third-party apps too: “if you choose to allow apps to use Speech Recognition for transcription, the audio data to be transcribed may be sent to Apple.” Source: Apple, Siri & Dictation privacy

The setting to check is Improve Siri & Dictation. Turned on, it lets Apple “store audio of your Siri and Dictation interactions from your Mac. Apple may review a sample of stored audio.” To change it: “choose Apple menu > System Settings, then click Privacy & Security in the sidebar. (You may need to scroll down.) Go to Analytics & Improvements, then turn Improve Siri & Dictation on or off.” Source: Mac User Guide You can also clear what is already stored: Apple says you can delete request history, “both audio recordings and transcripts”, that is “less than six months old” using Delete Siri & Dictation History. Source: Mac User Guide

How Apple Dictation compares with apps that clean up what you say is covered in Dictro versus Apple Dictation.

Context features

Apps that read your screen, and where it goes

“Context” is the newest privacy question in dictation. It makes output better — the app can spell a colleague's name from the email you are replying to — but it means the app reads far more than your voice. Here is what six apps say, in their own words.

Wispr Flow — context is on by default and sent with each dictation

Wispr's help page says “Flow reads text near your cursor and sorts your active app into one of four categories”, and “It is on by default on Mac and Windows.” It is specific about what is collected: “Context is collected locally and sent to Wispr with each dictation request unless Privacy Mode is on. It includes app info, textbox contents (before, selected, and after the cursor), on-screen text, variable and file names in coding apps, your user identifier within the app, the apps in your current session, a screenshot, and conversation history (participant IDs and message roles/content).” Source: Wispr Flow help

The same page describes Privacy Mode as “When enabled, no dictation or context data is retained or used for training” — which speaks to storage, not to whether context is sent. To turn the feature off, it says: “To disable Context Awareness for sensitive environments (legal, healthcare, or regulated data), go to Settings → Data and Privacy”. Look for the current name there: Wispr's data-controls page says one of its settings “was previously called ‘Privacy Mode.’” Source: wisprflow.ai/data-controls On permissions, Wispr notes that “Screen Recording permission is optional and is not requested during setup; macOS may request it when you use a feature that reads the screen.” Source: Wispr Flow help We compare the two apps in Dictro versus Wispr Flow.

superwhisper — rich context, and you choose where it is processed

“Super Mode uses accessibility APIs to gather three types of context”: “Text that is highlighted/selected in your active window when you begin recording”, “Text from active input fields, names, and title from your active window”, and “Content that has been copied to your clipboard before or during dictation.” “When Application Context is active, Superwhisper also includes system information: the current date and time, your full name, and your computer name.” Helpfully, “Open the dictation in History to review everything that was sent to the AI.” Source: superwhisper docs

Super Mode is “the only built-in mode with all three context types on by default”, and the docs advise: “Enable only the context types a mode needs.” Source: superwhisper docs Where that context goes depends on your models: “Pair a local voice model with a local language model, and every stage of a dictation stays on your machine.” Source: superwhisper docs

VoiceInk — context is opt-in per Mode, and OCR runs locally

VoiceInk is careful to separate the two steps: “For Screen context, VoiceInk captures the active window, runs OCR locally with Apple's Vision framework, and sends extracted text to the AI provider as part of the prompt context.” And the default is conservative: “Plain Dictation Modes do not send selected text, clipboard text, or screen text to an AI provider.” Source: VoiceInk docs It asks for Screen Recording only for this: it “Reads visible window text for Mode context when Screen context is enabled.” Source: VoiceInk docs

Spokenly — local transcription, but AI clean-up can leave the Mac

Spokenly lists the “Included context sent to the AI: clipboard, focused app, text around the cursor, active browser URL.” Source: Spokenly docs Its privacy page makes the point this guide keeps returning to: when a Mode uses AI Instructions, “your transcription and any optional context for that Mode may be sent to AI services for text processing, even if transcription itself runs locally.” Source: Spokenly privacy For the strictest setup, “Local Only Mode blocks all internet connections for extra privacy.” Source: Spokenly docs

Willow Voice — optional context, not retained

“If you turn it on, Willow can read small bits of text around your cursor,” and “When the feature is off, Willow does not read or use any on-screen text.” Source: Willow help Its privacy policy adds that “This contextual text is used only to provide the requested transcription, is not retained, and is not used for model training.” Source: Willow privacy policy

The overlooked one

The clipboard: where dictated text can quietly pile up

Many dictation apps — Dictro included — place your words by putting them on the clipboard and pressing ⌘V for you. That is reliable, but it has a side effect worth knowing: if you run a clipboard manager, it may save every dictation to its history. If that matters to you, add your dictation app to the clipboard manager's ignore list.

The clipboard can also flow the other way. superwhisper's Super Mode reads “Content that has been copied to your clipboard before or during dictation”, and Spokenly lists the clipboard as context. If you copy something sensitive, then dictate with one of those modes on, it may become part of what the app processes.

Held to the same standard

What Dictro can see

This guide is published by Dictro, so here is our own app on the same questions, with nothing left out.

• Permissions: Microphone and Accessibility, plus Automation for each browser only if you turn on website matching. No Screen Recording, and no Input Monitoring since version 0.5.91.
• Your voice and words: speech recognition and the clean-up pass both run on your Mac. There is no cloud mode, and no audio or transcript is uploaded.
• Your screen: Dictro notes which app is in front so it can use that app's writing style. It reads the text field you are typing into just before and just after pasting, on your Mac, to confirm your words arrived. It takes no screenshots and reads no other windows.
• Websites: only if you turn on website matching (off by default), Dictro reads the address of the page you are dictating into to pick that site's style. macOS asks your permission for each browser. The address is not stored or sent.
• Your clipboard: Dictro pastes through the clipboard and puts your previous clipboard back afterwards. A clipboard manager can still record the dictated text, as above.
• What does reach our server: you sign in, and the app sends an anonymous heartbeat with a random identifier, the app and macOS version, and a count of words dictated. Transcript history is off by default and stays on your Mac if you turn it on. The detail is in the privacy policy.

How Dictro compares with other Mac apps on where speech is recognised is in private dictation apps for Mac, compared; the architecture is explained in private dictation for Mac.

Do this today

A five-minute privacy checklist for voice typing on Mac

1. 1

   Review who has access

   In System Settings > Privacy & Security, open Microphone, Accessibility, Input Monitoring and Screen & System Audio Recording. Switch off any app you no longer use.

2. 2

   Decide on Improve Siri & Dictation

   Under Privacy & Security > Analytics & Improvements. Off means Apple does not store your dictation audio.

3. 3

   Turn off context you do not need

   In your dictation app, look for Context Awareness, Super Mode, screen or clipboard context, and switch off what you are not using — especially for legal, medical or client work.

4. 4

   Test with the network off

   Switch Wi-Fi off and dictate into Notes. Text that still appears was recognised on your Mac. For apps with AI clean-up, test the mode you actually use.

5. 5

   Watch the orange dot

   The dot next to Control Center means the microphone is on. Open Control Center to see which app is using it.

More on dictating without a connection in the offline dictation guide, and on using your voice across Mail, Slack and code editors in voice typing in any Mac app.