Automated Calls Have Rules. Design Consent In.
A platform that can place a thousand calls an hour will not tell you whether you were allowed to place any of them. That question sits with you.
The short version
6 things that decide this
- 01The Telephone Consumer Protection Act governs calls placed with an autodialer or a prerecorded or artificial voice. An AI voice agent placing outbound calls falls under that definition.
- 02Liability for an unlawful automated call sits with the business that placed it, not the voice AI platform it ran on. A vendor's terms of service do not move that liability.
- 03What consent a call needs depends on its purpose. Informational or transactional calls need a lower bar than anything that markets or sells.
- 04The FCC requires that a called party can revoke consent through any reasonable method. A voice agent needs a working way to hear and log a stop request mid-call, not just at signup.
- 05Caller identification rules under the Truth in Calling Act still apply to an AI-placed call. The system has to present accurate caller ID, not a masked or spoofed number.
- 06None of this is solved by a disclaimer in the system prompt. It has to be built into call flow, consent records and the number the call comes from.
Vendors sell the capability, not the liability
A voice AI platform's pitch is usually about what the agent can do. Pick up on the first ring. Handle names and dates. Transfer to a human when it should. Almost none of them mention the Telephone Consumer Protection Act, the federal law that governs automated calls in the United States.
That silence is not an oversight. The TCPA puts liability on the business that placed the call, not on the software that dialed it. A platform can sell you the ability to call ten thousand numbers a day and carry none of the risk from doing it wrong. The contract you sign usually says so directly.
This is not a reason to avoid voice AI. It is a reason to treat consent and identification as part of the system, the same way you would treat authentication or logging. They are requirements, not settings a vendor configures for you.
The call's purpose decides the consent bar
The TCPA and the FCC's rules under it do not treat every automated call the same way. A purely informational call sits under a lighter consent standard than a call that markets or sells something. An appointment reminder to a number given for that purpose is one example.
The same line that governs texts governs calls. A confirmation call is one thing. Add an upsell offer at the end, and the call moves into a category the FCC treats as advertising, which needs stronger consent. Our answer page on service-text consent walks that classification line, because a voice agent has to respect the same one.
What this means for a build: the system needs to know what kind of call it is placing, and hold the matching consent record. A single blanket "we have consent" flag is not enough once a business runs both service calls and outreach through one agent.
- 01Store what a contact consented to, not just that they consented.
- 02Classify each outbound call against that record before it dials.
- 03Treat a marketing line added to a service call as a change in what consent it needs.
- Check consentMatched to this call's purpose, before dialing.
- Present real caller IDNo spoofed or masked number.
- Identify on connectState who is calling and why, early in the call.
- Listen for opt-outAny reasonable phrasing, not one keyword.
- Log the revocationWrite to the same suppression list every channel reads.
- Suppress the numberBefore the next scheduled call, not after a complaint.
The fourth and fifth stations are where most builds fail. An agent trained to sound natural still has to recognize "stop calling me" as an instruction, not conversation.
Opt-out has to work inside a live conversation
Text-based opt-out is a known problem: recognize STOP, update a list, confirm once. Voice makes it harder. The request can arrive in any words, at any point in the call, and the agent has to notice it while still doing its job.
The FCC's guidance on revocation does not require a magic word. It requires that any reasonable method expressing a wish to stop be honored. "Take me off your list" and "don't call again" both count. A voice agent that only responds to a scripted phrase is not meeting that bar.
The engineering answer is the same one that applies to text. Revocation has to reach one suppression list, read by every calling channel before it dials, not a note buried in a call transcript. A call center's do-not-call list and a voice agent's calling queue have to be the same list. Otherwise the second one keeps calling a number the first one already stopped.
Questions this raises
01Is an AI receptionist legal under the TCPA?
An AI receptionist answering inbound calls a customer placed is not the situation the TCPA restricts. The law targets automated outbound calls made with an autodialer or a prerecorded or artificial voice. Legal exposure appears when that same system starts placing outbound calls, where consent and identification rules apply based on the call's purpose.
02Do TCPA rules for automated calls work the same as the rules for texts?
They run on the same framework. What consent a message or call needs depends on whether it is transactional or promotional, and revocation must be honored through any reasonable method. Our answer page on service-text consent covers the classification line and the ten-business-day revocation window in detail.
03Who is liable if an AI voice platform places a call that breaks TCPA rules?
The business that placed the call, or had it placed on its behalf, carries the liability. A voice AI platform's terms of service routinely disclaim responsibility for how a customer uses the dialer. That is why consent and opt-out have to be built into the calling system, not assumed to be the vendor's problem.
04What counts as valid consent revocation on a call?
Any reasonable method that clearly expresses a wish to stop receiving calls, spoken in the caller's own words rather than a required phrase. A system that only recognizes one scripted response is not meeting the standard the FCC has described for texts and calls alike.
Related
- TCPA rules for automated service texts →The consent tiers and revocation window this post's rules are built on.
- AI agent development →Where consent, identification and opt-out get built into the call flow.
- Should an AI receptionist handle emergency calls →The other line an automated phone system has to know not to cross.
