← Mac dictation privacy research

Free privacy-path checklist

Trace where your words go.

Answer six concrete questions about the dictation mode you actually use. The checker separates speech processing, connected features, offline use, retention, accounts, and the final text destination—without collapsing them into a misleading privacy badge.

No signup · no saved answers

Six boundaries · one active setup

Build the evidence path.

Answered0/6
SpeechWhere does microphone audio become text?

Check the active model or mode. A product can offer both local and cloud speech paths.

Connected featuresCan a later feature receive text or context?

Look beyond transcription: cleanup, rewriting, assistants, vocabulary, sync, and agent handoff can create separate paths.

OfflineDoes the selected speech path work without internet?

Evaluate after any required model download, app setup, or license check. Offline speech and fully disconnected software are different claims.

RetentionWhat does the current policy say is retained?

Separate audio recordings, transient processing, transcript history, account data, logs, and the destination copy.

AccountWhich identity or license boundary participates?

An account, subscription, license check, provider key, or synced history can exist even when speech runs locally.

DestinationWhere will the finished text live?

The receiving editor, document, CRM, chat, or AI agent creates a second policy boundary after dictation.

Questions to carry forward

Turn labels into verifiable questions.

The list updates with your choices. Use it when reading a privacy policy, evaluating a new mode, or asking a publisher for a precise answer.

  1. Find a current first-party source that names where microphone audio becomes a transcript for the selected mode.
  2. Check whether cleanup, rewriting, assistants, vocabulary, sync, or agent features send audio, transcript text, selected text, context, or metadata.
  3. Test the selected speech mode after required setup and document which setup, license, sync, or feature connections still apply.
  4. Read the current retention terms for audio, transcripts, account history, logs, subprocessors, and deletion—not only one no-retention headline.
  5. Identify every participating account, subscription, license check, provider key, and synced-history boundary.
  6. Name the receiving app or service and check its storage, sharing, training, retention, and organization policies separately.

Why this checker has no privacy score

A score would hide the decision you actually need to make. Local speech, remote speech with no stated retention, a connected cleanup feature, a license check, and a cloud document are different facts. Adding them into one number would invent tradeoffs the evidence cannot justify.

The result therefore stays descriptive. It names the path, exposes what is unknown, and gives you questions to verify. This is not a security certification, and it does not call an app secure, compliant, or universally private.

Use the active mode—not the product headline

A dictation product can offer an on-device model and an optional remote model. It can transcribe locally and later send transcript text to a rewrite service. It can keep audio on the Mac while the receiving document syncs the resulting text.

Record the model, feature, account, and destination you actually selected. Repeat the check when any of those choices—or the product policy—changes.

Start with source-reviewed product evidence

The open Mac Dictation Privacy Matrix separates the same boundaries across current products and links each row to its reviewed first-party sources. Use the checker for your configuration, then inspect the dated matrix for product evidence and limitations.

Open the Mac Dictation Privacy Matrix →

Questions

Read every boundary precisely.

What is a dictation privacy path?

It is the sequence of systems that microphone audio, transcript text, optional cleanup or assistant content, and the final text destination pass through. Each step can have different processors, accounts, and retention terms.

Is on-device dictation always private?

On-device describes where a specific speech-to-text step runs. A later rewrite, sync, assistant, provider, or destination can still send transcript text or context to a connected service.

Does zero retention mean audio stays on my Mac?

No. A remote service can process audio while promising not to retain it. Local processing and remote processing with a retention limit are different architectures.

Does this checker test network traffic or app security?

No. It organizes documentation and configuration evidence. It does not inspect packets, audit source code, certify a binary, or prove that implementation matches policy.

Does IraVoice receive or save my answers?

No. The answers stay in temporary page memory and disappear when the page is reloaded or closed. The site records only the same cookie-free aggregate page request described in the privacy policy.

Continue the audit

Evidence before recommendation.

18 documented products

Inspect speech, connected features, and source notes.

Open the matrix →Measure your setup

Compare speed and word accuracy without audio upload.

Run the speed test →Learn the distinction

Understand local, cloud, retention, and destination.

Read the guide →

macOS 26+ · Apple-silicon

Test IraVoice against the same six boundaries.

Speech recognition and supported formatting run on the Mac. Speech models download on first use, and paid activation sends the license key to Gumroad. The trial requires no card or account.

Start 14-day trialRead the full privacy policy