Voice Conversation
Voice Without Recording
Privacy by design in a spoken website conversation: no audio kept, the call stored as text, contact details typed, and no voiceprint of the visitor.
Updated 19 Nov 2026
A voice is personal. A website that invites visitors to talk should take less from them, not more. These are the limits that make a spoken conversation as safe as a typed one.
Keep the words, not the voice
What a visitor says is needed to answer them; the sound of their voice is not. The call is kept as text in the conversation, like a typed message, with a short written summary for the team. Privas AI does not store the audio, and the voice service that carries it is set not to record calls.
Type what must be exact
Speech recognition is good with sentences and poor with addresses. One misheard letter turns a real email address into a stranger's, and a follow-up email then goes to the wrong person. So an email address or a phone number is taken only when the visitor types it. If they say one aloud, the assistant asks them to type it on the screen.
No voiceprint
A voice can identify a person. A website assistant has no reason to: it does not need to know who is speaking, only what they asked. Nothing should be used to recognise the visitor by their voice.
Ask before the microphone
The visitor sees a short notice the first time, before the microphone is used, and nothing is heard until they press the call button. The details of how speech is processed belong in the privacy policy, where a visitor can read them in full.
This follows the same thinking as Privacy-First AI: collect what the conversation needs, and nothing else.