How to Read a Dictation Privacy Policy in 5 Minutes
Privacy policies are long, vague and written to be skimmed past. But when a dictation app hears every word you say, a five minute read can tell you whether your voice stays on your Mac or gets shipped to a server. Here is exactly what to look for, and what to ignore.
Key takeaways
- Use your browser find function and search for four words: processed, stored, shared and train.
- On-device means the model runs on your Mac and your audio is never uploaded.
- Cloud phrasing (servers, processors, partners) means your voice travels off your device.
- The fastest verification is a test: go offline and see if dictation still works.
Why a dictation privacy policy is worth five minutes
A dictation tool is not like a weather app. It has a live microphone feed of your meetings, client notes, medical drafts, passwords read aloud and half-formed ideas you would never publish. Where that audio goes is the single most important fact about the product, and it is almost never on the marketing page. It is buried in the privacy policy. The good news is that you do not need to read the whole thing. You need to find four specific claims. If you are new to the topic, our primer on whether Mac dictation is actually private gives useful background before you dive into the legal text.
The 5-minute method: four searches
Open the policy in your browser and press Cmd+F. Instead of reading top to bottom, search for the words that reveal where your voice actually goes. Each search takes under a minute.
Your 5-minute privacy policy checklist
- Search “processed” or “on-device”: does it say transcription happens locally on your Mac?
- Search “stored” or “retain”: are recordings kept, and for how long?
- Search “third party” or “processors”: who else touches your audio?
- Search “train” or “improve our models”: is your voice used to build their product?
- Search “delete”: can you remove your data, and is the process clear?
- Check the date: an unmaintained policy from years ago is its own red flag.
If the policy passes all six, you are looking at a tool that respects your voice. If it dodges any of them with vague language, treat that as a no until proven otherwise. For a deeper look at the terminology, our explainer on on-device versus encrypted cloud is worth ten minutes of its own.
Green flags and red flags, side by side
The same idea can be phrased to reassure you or to quietly warn you. Here is how the wording usually breaks down once you know what to look for.
| What the policy says | What it usually means | Verdict |
|---|---|---|
| “Transcription happens on-device” | The model runs on your Mac; audio is not uploaded | Green flag |
| “Audio is never stored or sent to our servers” | No cloud copy of your voice exists | Green flag |
| “We process your request in the cloud” | Your audio is uploaded to be transcribed | Red flag |
| “Shared with trusted third-party processors” | Other companies receive your data | Red flag |
| “Used to train and improve our models” | Your recordings become product fuel | Red flag |
| “We retain recordings for up to N days” | A stored copy exists for that window | Caution |
Notice that cloud tools are not necessarily malicious. Many encrypt data in transit and delete it quickly. But encrypted upload is still upload: your voice left the device and sat on someone else's computer, even briefly. That is a different privacy model from on-device processing, where the audio never leaves your Mac at all. If you want a concrete example of how these models diverge in practice, see our breakdown of cloud versus on-device dictation.
Verify the claim, do not just trust it
A policy can say on-device and still be wrong or outdated. The fastest way to check is a thirty second test. Turn off Wi-Fi, open any app, and try to dictate. If transcription works fully offline, the speech model is genuinely running on your Mac. If it stalls or errors out, it needed the network, which means it was sending your audio somewhere. This test is more honest than any paragraph of legalese, and it works for any tool. We cover it in more depth in our guide to whether on-device dictation is actually private.
It also helps to know that on-device speech recognition is real, mature technology, not a marketing gimmick. Open models like OpenAI's Whisper and NVIDIA's Parakeet run entirely on modern Apple Silicon, which is why a serious app can transcribe accurately without ever touching a server. When a policy claims on-device and the app passes the offline test, the two line up.
Where BlaBlaType lands
BlaBlaType is built so the privacy policy is short to read, because there is little to disclose. Speech recognition runs 100% on-device using local Whisper and Parakeet models, and the AI cleanup that removes filler and fixes punctuation runs on-device through Apple Intelligence. Your audio and transcripts never leave your Mac, so there is no cloud upload, no retention window and no third-party processor to worry about. It is macOS only and optimized for Apple Silicon, works system-wide in any app or text field, and offers a three day free trial with no card. You can read more about the approach on the product overview or compare plans on the pricing page.
Dictation with nothing to hide
Speech recognition and AI cleanup that run entirely on your Mac. Your voice never leaves the device. No card needed for the trial.
Download for macOSOnce you trust where your voice goes, dictation becomes genuinely useful for everyday work. Most people speak around three to four times faster than they type, so private, system-wide voice typing is a real productivity gain, not just a privacy one. For a practical starting point, try our walkthrough on how to dictate emails on Mac.
Frequently asked questions
What is the single most important line in a dictation privacy policy?
Look for a clear statement about where audio is processed. If the policy says transcription happens on your device and audio is never uploaded or stored on servers, that is the strongest signal. If it describes uploading audio to process your request, your voice leaves your Mac.
Does on-device dictation still need a privacy policy?
Yes, because the app may still collect other data such as crash logs, license checks or analytics. A good on-device policy will say audio and transcripts stay on your Mac while being transparent about any account or telemetry data it does handle.
What words in a privacy policy mean my voice is uploaded?
Watch for phrases like send to our servers, process in the cloud, third-party processors, share with partners, train our models, or retain recordings. Any of these usually means your audio leaves your device at some point.
How can I verify a dictation app really works on-device?
Read the policy for an explicit on-device claim, then test it: disconnect from the internet and try dictating. If transcription still works fully offline, the model is running locally on your Mac rather than in the cloud.
Is BlaBlaType on-device, and what does its policy say?
BlaBlaType runs speech recognition 100% on-device using local Whisper and Parakeet models, and the AI cleanup runs on-device through Apple Intelligence. Audio and transcripts never leave your Mac, so there is no cloud upload of your voice to read around.