Skip to main content
Privacy in voice AI is not one toggle. You shape it through several controls that affect what callers hear, what you store, what you publish, and what your team can later review or export. This guide shows how those controls work together.

​
The Four Main Privacy Layers

LayerWhat it controlsMain doc
NoticeWhat the caller hears before or during the conversationAnnouncements
StorageWhat the platform stores and for how longData Retention Settings
Caller choiceWhether the caller can stop recording mid-callRecording Opt-Out
Public transparencyWhat callers and visitors can read about how your agents handle their dataTrust Center

​
Why This Matters

Most privacy failures happen at the boundaries:
  • a call is recorded but callers were not prepared for it
  • retention settings do not match the public policy
  • a widget is live, but the privacy links are missing or unclear
  • a team exports or forwards data without checking what is actually stored
The safest pattern is to make notice, storage, and public disclosure align.

​
Start With The Policy You Actually Want

Before changing the UI, decide:
  • do you want to store recordings, transcripts, both, or neither?
  • do some agents need stricter rules than the account default?
  • do callers need an explicit way to opt out of recording?
  • do public web visitors need a visible Trust Center before they interact?
Once that policy is clear, configure the product around it.

​
Common Privacy Patterns

Standard customer-service workflow

Typical setup:
  • pre-call notice enabled
  • transcripts and recordings stored with defined retention
  • public Trust Center enabled for widget traffic
  • recording opt-out enabled when required by policy or jurisdiction
Typical setup:
  • short notice with clear disclosure
  • reduced retention windows
  • some data types set to Do not store
  • public Trust Center aligned to the stricter policy
Typical setup:
  • explicit announcement before the agent starts listening
  • tight retention rules or no recording storage
  • recording opt-out enabled when recordings exist
  • clear internal review process for exports and webhook consumers

​
How The Controls Fit Together

​
1. Announcements set the caller expectation

Use Announcements when callers should hear a notice before the agent begins listening. This is the right place to explain:
  • that the conversation may be recorded
  • why it is being recorded
  • what the caller can do if they do not want recording

​
2. Retention decides what stays in the system

Use Data Retention Settings to control:
  • whether transcripts are stored
  • whether recordings are stored
  • how long analytics artifacts remain available
  • whether expiry leads to deletion or anonymization
The important rule: your retention settings should match what your team promises publicly.

​
3. Recording Opt-Out handles live caller objections

Use Recording Opt-Out when recordings are enabled and callers may reasonably ask for recording to stop. If recordings are already set to Do not store, this control becomes less relevant because the recording is not kept in the first place.

​
4. Trust Center explains your public-facing posture

Use the Trust Center to publish:
  • privacy policy links
  • subprocessors information
  • retention posture
  • transparency material for website visitors
This matters most for web widgets, public demos, and externally shared assistants.

​
Widgets Need Extra Care

For widget-based conversations, align these three things:
  1. the widget consent and privacy copy
  2. the public Trust Center page
  3. the actual retention and recording behavior behind the widget
If those three are inconsistent, visitors lose trust quickly. See Web Widget Deployment and Web Widget Implementation for the deployment side.

​
Privacy Review Checklist Before Launch

  1. Confirm whether transcripts and recordings are stored.
  2. Confirm whether the agent-level retention overrides match the account default you intended.
  3. If recordings are enabled, decide whether callers should hear a notice and whether they can opt out.
  4. If a widget is public, confirm the Trust Center and policy links are live.
  5. Check which downstream systems receive conversation data through exports or webhooks.
  6. Run a real test conversation and confirm the behavior matches the policy you plan to publish.

​
Common Mistakes

The strongest public privacy copy does not help if the actual settings still store more than your policy says.
Webhooks, exports, and internal review workflows can extend the privacy footprint beyond the voice call itself. Review the full flow, not only the agent tab.

​
Next Steps

Announcements

Configure the notice callers hear before the conversation begins

Data Retention Settings

Set agent-level storage and expiry rules

Recording Opt-Out

Let callers stop recording during a live conversation

Trust Center

Align your public privacy posture with the actual system behavior