Recording consent

What the platform records

Recording is on by default for every call. A recording holds the full call audio, the transcript, call metadata (timestamps, caller ID, and similar), and any post-call analysis fields you have defined. See post-call-analysis for how those fields are produced and retention for how long the data lives.

You can turn recording off in two places, both in the dashboard. There is no API for either:

Turning recording off is not retroactive. Calls already recorded stay in storage until their retention window expires.

Where disclosure is mandatory

In two-party-consent jurisdictions, every party on the call must consent to being recorded. In the United States that includes California, Florida, Illinois, Massachusetts, Maryland, Montana, New Hampshire, Pennsylvania, and Washington, plus a few others. The rule follows the person, not your office. If your agent calls into one of these states, or takes inbound calls from one, you need a disclosure even if your business sits elsewhere.

This is not legal advice. The list of two-party states shifts, and other countries have their own rules. Confirm with a lawyer who works in telecom before you launch into a new jurisdiction. See compliance for the wider regional picture.

Adding the disclosure

Two options. Pick one, not both, or the caller hears the disclosure twice.

Option 1 — put it in the welcome message. Edit the agent's welcome message so the first thing the caller hears includes the disclosure. A working line:

"This call may be recorded for quality purposes."

This gives you control over wording and placement, and it works with agents-multilingual if you write the disclosure into each language variant.

Option 2 — enable disclosure injection. Settings → Compliance → Recording Disclosure. The platform prepends the disclosure to the call. This is a dashboard toggle. There is no endpoint for it.

Two failure modes to watch for. First, a disclosure buried three sentences into a long welcome message is weak evidence of consent, because the caller may have started talking over it. Put it in the opening seconds. Second, if you use a multilingual agent and only write the disclosure in English, callers routed to another language get no disclosure at all.

Opt-out

Recording consent and call consent are separate. Disclosure covers recording. It does not cover whether you were allowed to call the person at all.

For call opt-outs, define an opt_out post-call field (yes/no). When the agent detects an opt-out request during a call, the field is marked yes. Enable Settings → Compliance → Auto-Suppress on Opt-Out and any number with opt_out = yes is added to your global suppression list. Suppression is global and immediate: it applies to every audience and every deployment.

What this does not do:

• It does not delete existing recordings. Suppression stops future calls. Deletion is a separate action under retention.
• It does not fire if you never defined the opt_out field. The auto-suppress toggle has nothing to act on, and the call ends with no record of the request.
• It does not cover a caller who objects to recording but still wants the call. There is no mid-call recording kill switch. If a caller refuses recording, your options are to route to a human via tools-transfer or end the call.

Under GDPR, a caller can ask you to delete their data, and you have 30 days to comply. Right-to-delete is available in the dashboard and via API. See retention and data-residency.

Redaction

PII redaction is opt-in under Settings → Compliance → PII Redaction. It strips SSNs, full credit card numbers, and similar from transcripts.

The audio is untouched. A redacted transcript sitting beside an unredacted recording of the same call is a common misreading of this feature. If the audio itself must not contain the card number, delete or trim that recording by hand. There is no automated audio redaction.

Disclosing that the agent is AI

Not the same obligation as recording disclosure, but it lands in the same conversation. Using an AI voice agent is legal in most jurisdictions. Actively deceiving someone who directly asks whether they are talking to a human is not.

Add a guardrail to the prompt so the agent can always answer honestly:

"If the caller asks if you're a real person, tell them honestly that you're an AI assistant."

See agents-prompting for where guardrails go in the prompt.

Checklist before launch

1. Decide whether you need recording at all. If you do not, turn it off per agent or workspace-wide and skip the rest.
2. Identify every jurisdiction your agent will call into or receive calls from.
3. Add the disclosure by welcome message or by injection, in every language the agent speaks.
4. Define the opt_out post-call field and enable auto-suppress.
5. Place a test call and listen to the recording. Confirm the disclosure is audible in the first seconds.