What 'your words aren't training data' really means — and how to check
Last updated June 2026
Every app on earth says it respects your privacy. The claim that matters — 'your words aren't training data' — is checkable. Here are the specific questions to ask, and what a good answer actually looks like.
"We take your privacy seriously" tells you nothing. It's the universal disclaimer, printed by apps that protect you and apps that quietly monetize you alike.
Why "private" is meaningless on its own
Privacy isn't one thing. An app can encrypt your data in transit and still train models on it. It can promise not to sell data and still store every recording indefinitely. "Private" with no specifics is a feeling, not a guarantee.
The questions to ask
- Where is my audio processed? On my device, or uploaded to a server?
- Is anything used to train AI models? Mine or anyone else's.
- What's retained, and for how long? Audio, transcripts, both, neither.
- Who can access it? Employees, subprocessors, advertisers.
"Don't ask whether an app is private. Ask where your words go, and what's done with them there."
What a good answer looks like
Strong answers are concrete and verifiable: processing happens on your device; nothing is used as training data; little or nothing is retained on servers; access is minimal by design. Weak answers lean on adjectives — "secure," "trusted," "enterprise-grade" — without saying where your data is or what touches it.
How Kalpa answers
Kalpa transcribes on your iPhone, your words aren't turned into training data, and your second memory lives on your device rather than a company's servers. That's the whole reason it's safe to capture freely — the value of a second memory only shows up when you stop self-censoring, and you only stop self-censoring when the privacy answer is specific.
Frequently asked
What does 'not training data' mean?
Your captured words are never fed into the training of AI models — not the app's, not a third party's. Your content is used to serve you, not to improve a model.
How do I verify an app keeps my data private?
Ask four concrete questions: where audio is processed, whether anything trains models, what's retained and for how long, and who can access it. Vague answers are a red flag.
Is on-device processing enough?
It's the strongest signal, because your audio and transcript never leave your phone. Pair it with no-training and minimal retention for a complete picture.