Call Recording and Consent: What California Businesses Need to Get Right

Recording calls is a standard feature of every cloud phone system, enabled with a toggle. The legal obligations attached to that toggle vary by state, and California is among the strictest.
This is a practical overview of how the rules shape a phone system's configuration. It is not legal advice — if recording is material to your operation, have counsel review your specific setup.
The core rule
California's Invasion of Privacy Act requires the consent of all parties to record a confidential communication. Most states require only one party to consent, which — since you are a party to your own calls — effectively means you may record without telling anyone.
Two consequences follow for a business phone system.
Within California, you need consent from everyone on the call. A notification at the start of the call, with the caller continuing, is the standard mechanism.
Across state lines, assume the stricter rule applies. A call between a California business and a one-party state is a genuinely contested area, and the conservative, widely-adopted practice is to treat any call touching an all-party state as requiring all-party consent. This is the practical position most multi-state businesses adopt, and it is the one your phone system should implement by default.
What "confidential communication" means
The statute covers communications where a party has a reasonable expectation that it is not being overheard or recorded. A business call where a recording announcement has played is not confidential in that sense — which is exactly why the announcement works.
It also means some calls are lower risk than others. A call into a published support line where callers hear a recording notice is in a very different position from a recorded outbound call to someone who was told nothing.
Implementing it properly
Announce before recording starts
The announcement must play before the recording begins, not alongside it. On a well-built system the call flow is explicit: answer, play notice, then begin recording. If your platform starts recording at answer and plays a notice afterwards, you have recorded a portion without consent.
Check this rather than assume it. It is a real difference between platforms and between call flows within the same platform.
Cover outbound as well as inbound
Inbound recording announcements are routine. Outbound is forgotten constantly, because the caller is your employee and no IVR plays. If outbound calls are recorded, the agent must say so — scripted, at the start, before anything substantive.
Configure outbound recording as an explicit, separate setting. Teams that rely on agents remembering will have gaps.
Make it possible to decline
If someone objects, there must be a way to continue the call without recording. In practice that means a per-call stop-recording control the agent can use, and a policy that says what to do when it is used.
A system where recording cannot be stopped mid-call forces agents to choose between the customer's objection and your configuration.
Handle sensitive data within recordings
A recording of a payment conversation contains card details. A recording of a medical appointment contains health information. Recording brings the call inside the scope of whatever regime covers that data — PCI DSS, HIPAA and so on — with requirements for storage, access and retention.
Two standard controls:
- Pause-and-resume for payment capture, so card details never enter the recording. Some platforms automate this around a payment step.
- Redaction in transcripts and recordings, where the platform supports it.
Set a retention period and enforce it
Recordings kept indefinitely are a liability that grows. They are discoverable, they are a breach target, and under California's consumer privacy rules they may be subject to deletion requests.
Decide a retention period based on why you record — quality assurance needs weeks, dispute resolution may need longer, a regulatory requirement may dictate it — write it down, and configure automatic deletion. A policy with no enforcement is not a retention period.
Control who can listen
Access to recordings should be role-based and logged. "Anyone who can log into the phone system" is not an access policy. Supervisors reviewing their own team is a reasonable default; access to everything should be rare and audited.
Interaction with consumer privacy law
Call recordings are personal information under California's consumer privacy framework, and transcripts and AI-derived analysis — sentiment scores, summaries — are too. That brings obligations around disclosure in your privacy notice, responses to access and deletion requests, and limits on sharing with third parties.
If you use AI features on call content, confirm what the provider does with it: whether recordings or transcripts leave your tenant, whether they are used to train models, and what their retention is. That belongs in your vendor assessment, not in a feature comparison.
A practical configuration checklist
- Recording announcement plays before recording starts, on every inbound flow that records
- Outbound recording is a deliberate setting with a scripted agent disclosure
- Agents can stop recording mid-call, with a written policy for when to use it
- Payment steps pause recording, or card data is redacted
- Retention period is defined, configured and automatic
- Access to recordings is role-based and logged
- Privacy notice describes recording, transcription and AI analysis
- Vendor's data handling for recordings and transcripts is documented
- The configuration is reviewed when you add a location in another state
Why this is worth the attention
Recording is genuinely useful — for training, for dispute resolution, for understanding what customers actually ask. None of that requires getting the consent mechanics wrong, and the cost of getting them wrong is statutory damages per call in a state that provides for them.
The controls above are all standard features of a competent cloud phone platform. The work is deciding the policy and confirming the configuration implements it — not building anything.


