The case for audio that deletes itself

Most clinical software treats your recordings as an asset to retain. The default is to keep: keep the audio, keep the transcript, keep a copy on a server “just in case.” But the recording of a therapy session is not an asset. It is a liability that happens to be temporarily useful. The most defensible thing you can do with session audio is to auto delete session audio the moment it has served its single purpose — producing a draft note you then review and sign. Everything you keep past that point is something that can be subpoenaed, breached, or misplaced. Nothing you no longer hold can be.

Session audio lifecycle: capture, transcribe on device, auto-delete A left-to-right loop: an active recording is captured, transcribed on the device, then the audio is automatically deleted while the draft note is kept. Capture Active recording of the session Transcribe On device only — no cloud, no upload Auto-delete Audio removed on finalize, by setting The draft note remains; the audio does not.
Audio exists only long enough to produce a draft note you then review and keep.

Minimization is a security control, not a chore

In information security there is a principle older than any of the privacy laws built on top of it: data minimization. The least risky data is the data you never collected. The second least risky is the data you deleted. You cannot lose, leak, or be compelled to produce a recording that no longer exists. This is not a clever trick — it is the cleanest control available, because it removes the threat surface rather than guarding it.

Session audio is an unusually high-stakes form of data. It is your client’s voice, often discussing the most private parts of their life, captured with a fidelity that no progress note ever matches. A transcript can be redacted; a note is already a clinical abstraction. Raw audio is none of those things. It is the most identifiable, most sensitive artifact in your entire workflow, and it is the one piece you almost never have a clinical reason to keep once the note is written.

So the question is not “how do I protect the audio?” but “why am I still holding it?” Once you’ve extracted what you needed — a faithful draft of what was said — the recording has done its job. Keeping it past that point converts a tool into a stockpile.

Why auto delete session audio should be the default

A setting that deletes audio after you finalize the note should be the default, not a buried option you have to discover. Defaults carry moral weight: they are the choice made for the busy clinician who never opens the settings screen. If the default is “retain indefinitely,” you are quietly accumulating risk on behalf of every client who walks through your door.

A sound default looks like this:

  • Audio is captured locally and used to produce a transcript and a draft note.
  • You review, edit, and sign the note — the clinical record you actually keep.
  • The audio is deleted automatically when you finalize, with no extra step to remember.

The note is the record. The audio was always just scaffolding. CouchNotes is built around exactly this: transcription runs entirely on your Mac, and audio auto-deletes after finalize according to your setting — so the careful choice is the one you get without having to remember it. (You can find a fuller treatment of retention windows in how long to keep session audio.)

The safest recording is the one that has already been turned into a note and erased.

What you give up — and why it’s usually fine

Deleting audio is not free. It’s worth naming the trade honestly, because “delete everything” with no caveats is its own kind of overstatement.

You loseWhy it’s usually fine
Re-running transcription laterYou reviewed and corrected the draft at the time — when memory was freshest
Audio as evidence in a disputeRecordings are rarely the record boards or courts ask for; your signed note is
A “verbatim” archiveA verbatim archive is a liability most practices have no policy to govern

The recurring theme: the situations where you’d want the audio back are far rarer than the situations where holding it creates exposure. If your jurisdiction or payer genuinely requires you to retain recordings for a defined period — some do, in specific contexts — then set your retention accordingly. Retention and recording rules vary by state, licensing board, and payer, and this is not legal advice, so confirm your own obligations with your board or an attorney before you settle on a policy. Minimization means keeping what you must and nothing more; it does not mean deleting what the law tells you to keep.

There’s also a quieter benefit. When you tell a client “the recording is deleted as soon as I’ve drafted my note, and it never leaves this computer,” you are describing something concrete and verifiable, not a privacy policy they’ll never read. That clarity is hard to come by any other way.

Why on-device makes deletion trustworthy

Auto-delete only means something if you can be confident the audio wasn’t copied somewhere first. A cloud transcription service that “deletes after processing” still received your recording, held it on infrastructure you don’t control, and asks you to trust its deletion. Local-only processing closes that gap: if the audio is transcribed on your machine and never uploaded, then deleting the local file is genuinely the end of it. There is no second copy to wonder about.

This is the difference between deletion as a promise and deletion as a fact. When processing happens entirely on the device, there is no transmission, no vendor-side retention, and no server log to reconcile. (For how on-device processing differs from “cloud with a privacy policy,” see what on-device AI really means.)

It also reframes the compliance conversation. Compliance language describes safeguards around data you are still holding and possibly transmitting — necessary, but it is not the whole story. On-device processing with prompt deletion changes the question from “are our safeguards adequate?” to “what sensitive data do we even have?” The strongest answer is: as little as possible, for as short a time as possible.

None of this removes you from the loop. The draft note is still a draft; you remain the author of record, reviewing and signing every word. What deletion changes is what’s left behind afterward — and the right amount to leave behind, almost always, is the note and nothing else. Treat the audio as a means to a draft, give it a short and predictable life, and you spend less of your attention guarding something you were never going to need again.

Dario Valles

Building CouchNotes — on-device AI session notes for therapists on macOS and Windows. Sessions never leave your computer; that's the whole point.

Get the free beta