Recording customer calls, and what changes when an AI answers

If software answers your phone, it is almost certainly recording and transcribing the call. That is useful — and it puts two separate legal obligations on your business, not on your vendor. This is what they are and how to tell whether a vendor is actually handling them.

This is not legal advice. Consent law varies by state and by circumstance, and the penalties are not trivial. Talk to a lawyer about your own situation.

Is it legal to record a customer call?

In the United States, it depends on which state’s law applies — and on a call between two states, potentially both.

One-party consent — the majority of states. If you are a party to the call, you can record it. Your own consent is enough.

All-party consent — roughly a dozen states, including California, Florida, Illinois, Pennsylvania, Washington, Massachusetts and Maryland. Every party must consent before recording.

Two practical consequences owners routinely miss:

1. You do not control which state your caller is in. A shop in a one-party state taking a call from a customer in California has a real problem if it assumed its own state’s rule applied.
2. Notice and consent are not the same thing. “This call is recorded” tells someone what is happening. It does not obtain their agreement. In all-party-consent jurisdictions that distinction can be the whole case.

Because of point one, the workable approach for most service businesses is to hold every call to the strictest standard rather than trying to detect the caller’s state mid-ring.

Do I have to tell customers the call is recorded?

If you want the recording to be usable and defensible, yes — and the phrasing matters more than most vendors admit.

A statement — “this call is being recorded” — is one-party-consent framing. It informs. It does not ask.

A question — “this call may be recorded for quality, is that okay?” — invites an affirmative answer, which is what all-party-consent states are actually looking for.

The difference costs you one word and covers considerably more ground. It also has to be spoken before the caller starts talking, which sounds obvious and is the part that most often goes wrong in practice: a disclosure delivered after a caller has already explained their problem has not protected the part of the call that mattered.

Do I have to disclose that an AI is answering?

Increasingly, yes. California and a growing list of jurisdictions require that a person be told they are interacting with an automated system in commercial calls.

Two things worth separating:

• Disclosure up front is the requirement. The caller should know at the start.
• Dwelling on it is not. Once disclosed, a well-built agent should get on with helping rather than repeatedly announcing itself.

That is two obligations on one greeting — recording consent and bot disclosure — and they have to coexist with a greeting that still sounds like your business rather than a terms-of-service reading.

How can I tell if a vendor actually enforces this?

This is the question worth asking, and very few buyers do.

Most AI voice products let the greeting be configured — in a vendor portal, per assistant. That is convenient and it is a compliance hole: a custom assistant, a well-meaning edit, or a per-tenant override can silently drop the disclosure, and nobody finds out until it matters.

Three questions to put to any vendor:

1. Is the disclosure spoken from code, or from an editable setting? If it is a text field in a portal, it can be deleted.
2. Can the AI model itself skip or paraphrase it? If the greeting is model-generated rather than spoken verbatim, then on some percentage of calls it will come out differently.
3. Is there an audit that checks every live assistant still says it? Including custom ones.

For transparency, here is how Ansdale answers those three, because we think it is a fair standard to be held to:

The assistant’s greeting is set to a single placeholder token. The actual greeting line is composed server-side, per call, and includes both the recording question and the AI disclosure. Because it is spoken verbatim rather than generated, the model cannot rephrase it away. And an audit script checks that every live assistant — the shared trade assistants and every tenant’s custom override — still greets that way, so a custom configuration cannot quietly opt out.

We are describing this in detail because “we’re compliant” is not a checkable claim and “here is the mechanism” is.

What happens to the recordings afterwards?

The second half of the obligation, and the half that gets ignored.

A recorded, transcribed call is a store of your customers’ personal information — names, addresses, sometimes payment discussion, occasionally health or household detail people volunteer without thinking. Holding it forever is a growing liability rather than an asset.

Ask any vendor: what is the retention period, is it configurable, and what exactly is deleted when it expires — the audio only, or the transcript too?

Ansdale runs scheduled retention sweeps that prune call breadcrumbs and clear both recording links and transcript content past a configurable window. Details of what we hold and for how long are in the privacy policy.

Does this apply to text messages too?

Different law, same principle. Text messaging to customers is governed by the TCPA and, for business messaging in the US, by A2P 10DLC registration requirements — which means documented opt-in, a working STOP keyword, and records showing when and how consent was given.

Our SMS consent policy sets out how consent is collected, what it covers and how to opt out.

What is the practical checklist?

For a service business whose phone is answered by software:

• Hold every call to all-party-consent standard rather than guessing where the caller is
• Phrase the recording disclosure as a question, not a statement
• Disclose the AI at the start of the call
• Confirm the disclosure is enforced in code, not in an editable portal field
• Confirm an audit covers custom assistants
• Know your retention period and what is actually deleted
• Keep documented SMS opt-in separate from call consent — they are different permissions
• Have a lawyer review it against your states

What should I read before deciding?

• Answering service vs virtual receptionist vs AI — what each model does with the call
• How it works — what happens between the ring and the booked job
• Privacy policy — what we hold, and for how long
• SMS consent policy — how text consent is collected and revoked