Seven questions to ask any dictation app about your voice data
Voice is biometric, contextual and often confidential. Before you trust a tool with it, here's what to ask, and what good answers sound like.
The Wisper Team
3 min read
Your voice is not like other data. It's biometric, because it can identify you. It's contextual, carrying emotion, health cues and background sound. And when you dictate, it's often confidential: unreleased plans, client details, half-formed ideas you'd never post publicly.
So before you trust any dictation tool, including ours, it's worth asking how it handles what you say. Here are seven questions, why each matters, and what a good answer sounds like.
1. When does it listen?
Why it matters: A tool that's always listening, even to detect a wake word, has a much larger privacy footprint than one that listens only when you ask it to.
A good answer: "Only while dictation is active, which you start deliberately and can always see." There should be a clear, visible indicator whenever audio is being captured.
2. Where is my audio processed?
Why it matters: Some tools process audio on your device. Others send it to the cloud, sometimes to third-party AI providers. Each approach has trade-offs in accuracy, speed and privacy, but you deserve to know which one you're getting.
A good answer: A plain-language explanation of where processing happens, which providers are involved, and what those providers are contractually allowed to do with your data. A published list of sub-processors is a strong signal.
3. How long are recordings kept?
Why it matters: Data that no longer exists can't be leaked, subpoenaed or misused. Retention is one of the most important privacy decisions a company makes, and one of the least discussed.
A good answer: Specific timeframes, not "as long as necessary". The best default is to delete audio as soon as your text is produced, with saving recordings as an opt-in you control.
4. Is my content used to train AI models?
Why it matters: Training on user recordings can improve a product, but it also means your words may influence a model indefinitely. Many people are comfortable contributing; many aren't. Either is a legitimate choice, but it should be your choice.
A good answer: "Not unless you opt in." Look for a clear setting you can turn off at any time, and a commitment that third-party AI providers can't train on your data either.
5. Can people at the company hear my recordings?
Why it matters: "Human review" of recordings, where staff or contractors listen to samples to check quality, has caught several large tech companies out over the years. It's not inherently wrong, but it should never be a surprise.
A good answer: No routine human access to your content, with narrow and clearly stated exceptions such as when you share a sample with support or a legal obligation applies. Access should be logged.
6. Can I see, export and delete my data?
Why it matters: Rights on paper are only useful if they're easy to use. Modern data protection laws, including India's Digital Personal Data Protection Act and the EU's GDPR, give you rights to access, correct and erase your personal data.
A good answer: Self-service controls in the app for deleting history and your account, a straightforward way to export your content, and a named contact for privacy requests that responds within a defined timeframe.
7. What happens if something goes wrong?
Why it matters: No company can promise it will never have a security incident. What separates trustworthy ones is how they prepare and how honestly they respond.
A good answer: Encryption in transit and at rest, strict access controls, a public way to report vulnerabilities, and a commitment to notify affected users promptly if a breach occurs.
How Wisper answers
We wrote these questions because we ask them of ourselves. In short:
- Wisper listens only when you activate dictation.
- We explain where processing happens, and AI providers we use are contractually barred from training on or retaining your content.
- Audio is deleted after processing by default. Saving recordings is off until you turn it on.
- We never train models on your content without opt-in.
- There's no routine human review, and any exceptional access is logged.
- You can delete history and your account at any time, and request a copy of your data.
- We encrypt data in transit and at rest, and run a responsible disclosure programme.
The details are in our Privacy Policy and on our Security page, both written to be read by people rather than lawyers.
The bottom line
Voice tools are genuinely useful, and the best ones can be both powerful and respectful of your privacy. You shouldn't have to choose. Ask the questions, expect specific answers, and pick tools that treat your words as yours.