What happens to your voice in an AI journal app
Where your recording goes when an AI voice journal transcribes it, the three retention models apps actually use, how to read a privacy policy in five minutes, and what Dearly does with your audio.
The question nobody’s listing answers
Every AI voice journal app will tell you it’s “private” or “secure”. Almost none will tell you, in the store listing, the one thing that matters: after you stop talking, where is the recording, and for how long?
A diary is the most sensitive thing most of us will ever put on a phone. Before you trust an app with it, you deserve a straight answer. This guide explains what actually happens to your voice, the three models apps use, how to find out which one you’re looking at, and what we do in Dearly.
What has to happen for voice to become a journal page
To turn speech into text, something has to run a speech-recognition model over your audio. There are only two places that can happen:
- On your phone. The model runs locally. Your audio never leaves the device. This is the most private option, and also the most limited: on-device models are smaller, handle accents and background noise worse, and are often tied to specific hardware (Google’s Recorder app is Pixel-only for exactly this reason).
- On a server. Your audio is sent to a transcription service, converted, and the text comes back. Better accuracy, works on any phone, and the privacy question becomes: what does the server do with the file once it has the text?
Anything labelled “AI insights” — summaries, mood, patterns — is almost always server-side too, but it runs on the text of your entries, not on your audio. Those are separate questions, and a good policy answers both.
The three audio retention models
1. Uploaded and kept
The app stores your recordings in its cloud, usually so you can play them back later and sync across devices. Day One works this way (with end-to-end encryption, which is the best version of this model). The trade-off is permanent: your voice exists on someone else’s infrastructure for as long as your account does, and the app’s breach surface includes your audio.
2. Uploaded, then deleted
The recording is sent for transcription, the text comes back, and the audio is deleted from the server — immediately or within a stated window. AudioPen states this explicitly; so does Dearly. This is the practical middle: server-grade accuracy without long-term audio retention. The things to check are whether “deleted” is stated in the policy (not just the marketing) and whether a third-party transcription provider is named.
3. Never uploaded
Fully on-device. The most private and the rarest, because it constrains accuracy and hardware. If an app claims this, it should also work in airplane mode — that is the easy test.
The second question: is the AI trained on your entries?
Separate from audio, ask what happens to your text. Some services reserve the right to use your content to improve their models. For a diary that should be a hard no. Look for a plain sentence: “we do not use your content to train AI models.” If the policy doesn’t say it, assume it isn’t promised.
How to read a journal app’s privacy policy in five minutes
Search the policy (Ctrl/Cmd-F) for these words and read the sentence around each:
| Search for | You’re looking for |
|---|---|
| ”audio”, “recording”, “voice” | Is audio stored? For how long? Where? |
| “delete”, “retain”, “retention” | A stated window, not “as long as necessary" |
| "train”, “improve our models” | An explicit promise that your content is not used for training |
| ”third part”, “processor”, “provider” | Who actually runs transcription and AI; they should be named |
| ”encrypt” | In transit (the minimum) and ideally at rest |
| ”export”, “delete your account” | Can you take everything out and remove everything, from inside the app? |
If a five-minute search can’t answer the audio question, that is itself the answer.
Beyond the policy: the things on the phone
- A lock. Biometric lock (Face ID, Touch ID, fingerprint) on the journal itself, not just the phone. Your phone gets handed to people.
- Notifications that don’t leak. A reminder that previews your last entry on the lock screen undoes the lock.
- Analytics scope. Usage analytics are normal; your entries should never be in them. The policy should say what the analytics vendor receives.
What Dearly does
We built Dearly because every voice journal we tried was model #1 and wouldn’t say so. Here is our answer, in the same terms:
- Your recording stays on your phone. It travels over an encrypted connection only to be transcribed, and is never stored on a server. Model #2, stated in the privacy policy, not just here.
- Transcription and insights run on Google’s Gemini, named in the policy. Your content is not used to train models.
- Only the transcribed text syncs, so your journal is on your other device but your voice isn’t.
- Biometric lock is in the free tier.
- Export and permanent deletion are inside the app — how to delete your account.
- The trade-off, honestly: transcription needs a connection. Recording works offline; the page appears when you’re back online.
A checklist you can copy
Before trusting any AI voice journal:
- Does the policy say whether audio is stored, and for how long?
- Does it say your content is not used for training?
- Is the transcription/AI provider named?
- Can you lock the journal with biometrics?
- Can you export and delete everything from inside the app?
Five yeses is the bar. Related: what a private voice journal requires · best AI voice journal apps compared.