Call recording

Recording is set per organization, not per person. An administrator picks one of three modes:

Always

Every answered call records automatically, inbound and outbound.

The agent’s choice

Agents with the right permission start and stop recording on a call themselves. An administrator can separately decide whether an agent is allowed to stop one that is already running.

Never

Recording is off and the control is hidden.

Recordings are dual-channel: each side of the conversation on its own track. The audio is held by Twilio; we store a reference to it and stream it on demand. Permission to play a recording back is separate from permission to start one, so an agent who can record is not automatically an agent who can listen to everyone else’s calls.

An organization that has never configured a recording policy is in the agent’s-choice mode. Any agent with the recording permission can record any call, at their discretion, with no announcement.

This is deliberate: it is what every account had before the policy existed, and silently switching them off would have stopped recordings people were relying on. But it means “we never set it up” is not the same as “we do not record”. If you want recording off, set it to never.

Each organization can store an announcement to be spoken to the other party on a recorded call. When one is stored it is spoken on the call. When none is stored nothing is said, and recording still happens. The field is optional and we do not write one for you.

When it is spoken depends on the mode. In always mode the caller hears it before they are connected, which is the last moment anything can be said off the recording: the conference they join is recorded from its first second. In the agent’s-choice mode nothing is said at the start, because at that point nobody knows whether there will be a recording at all. The announcement is played into the call when an agent starts recording, which puts it just after the recording begins rather than just before. Everyone on the call hears it, your own people included.

If the announcement cannot be resolved or played, the call still connects and any recording still runs. We log it, but the other party heard no notice. A stored announcement is therefore a notice we speak, not proof that a particular person heard one.

So: if you are recording in a jurisdiction that requires notice, store an announcement; use always rather than the agent’s choice if the notice has to precede every recorded second; and brief your agents to say it out loud as well. A person saying “this call is being recorded” is the one notice that does not depend on us.

Yours. We provide the mechanism. Whether a particular recording is lawful depends on where each party to the call is sitting, what the call is about, and what you told them (none of which we can see, and none of which we determine).

Broadly, and this is not legal advice:

Canada

A party to a conversation may record it without the other’s permission under the Criminal Code. Privacy law is stricter than that: you should tell the person that the call is being recorded, why, and give them a way to decline, which usually means an alternative to continuing on a recorded line.

Quebec

The same, plus Law 25’s requirement that consent be clear, free, informed and given for a specific purpose. A recording made for quality assurance is not thereby consented to for anything else.

United States: federal, and most states

One party’s consent is enough, and the recording party counts.

California and other all-party states

Every party has to consent. California attaches statutory damages per violation and a private right of action, and the rule extends to any call involving a mobile or cordless phone whether or not the conversation was confidential. Assume a caller may be on a mobile, because they usually are.

If you take calls across borders (and a phone number does not stop anyone dialling it), the safe operating assumption is the strictest rule that could apply.

The audio is stored by Twilio rather than by us. Playback streams through our servers so we can check the listener’s permission first, which is why a recording URL is not something that can be forwarded to someone who should not have it.

Recordings are deleted on a schedule. Each organization sets how long recordings and voicemail are kept, and an organization that sets nothing keeps recordings for 30 days and voicemail for 90. A job runs nightly and deletes what has passed its window. Shortening a window therefore deletes audio on the next run.

A single recording can also be deleted before its window ends. That is a permission of its own, separate from the permission to play a recording back, and an erasure request under the privacy policy reaches the audio the same way.

In each of those cases the audio is erased at Twilio first, and our own record of the recording is marked only once Twilio confirms. There is no archive and no second copy, so a deleted recording is gone for us as well as for you. What stays behind is a note that the recording existed and when it was erased, which is what lets anyone check afterwards that an erasure really happened. Voicemail is transcribed automatically and the transcript is stored as text alongside it.

Ringfully does not have supervisor listen-in, coaching or barge. A supervisor cannot join a call silently; joining a call means being a participant on it, visible to everyone else on the roster. If you have seen those features named on a competitor’s comparison table alongside ours, they are not ours.

Write to [email protected]. Clause 7 of the terms of service carries the contractual version of this.