Key takeaways
- In the US, an AI-voiced call counts as an artificial voice under the TCPA, so true cold calling with AI to a purchased list is off the table.
- Sales calls with an AI voice need prior express written consent, and the agent must identify your business and offer an opt-out.
- The safest place to start is calling back people who just asked to hear from you, with consent captured on the form.
A sales leader asks a reasonable question on a Monday call: “We’ve got 8,000 contacts from a data vendor. Can we just point an AI voice agent at them and let it book meetings?” The demo looked great. The cost per minute looked even better.
The short answer is no, not in the US. Since February 2024, the FCC treats AI-generated voices as “artificial” voices under the Telephone Consumer Protection Act (TCPA). That means AI calls need the called party’s prior express consent, and sales calls need it in writing. AI cold calling, in the sense of dialing people who never asked to hear from you, is the exact thing the rule is built to stop.
That doesn’t make AI calling a dead end. It means you aim it at the right calls. Here’s how the rules work in plain English, and how we set up voice agents so they stay on the right side of them.
This is general information, not legal advice. Rules differ by country and state, so check with counsel for your markets.
What the FCC’s 2024 ruling changed
The TCPA has restricted “artificial or prerecorded voice” calls since 1991. Before 2024, some vendors argued a conversational AI wasn’t a recording, so the old rules didn’t fit. The FCC closed that argument in Declaratory Ruling FCC 24-17: voice cloning and other AI-generated voices count as artificial voices.
So every rule that applies to a robocall applies to your AI agent. The statute bars artificial-voice calls to mobile numbers and to residential lines without prior express consent, with narrow exceptions. The FCC’s rules add that telemarketing calls with an artificial voice need prior express written consent.
The penalties are real. Under the statute, the person called can sue for actual losses or $500 per violation, whichever is greater, and up to three times that for willful or knowing violations. On a dialer making thousands of calls, that’s not a rounding error.
Which consent you need for which call
Not every AI call carries the same requirement. The purpose of the call decides it.
| Call type | Example | Consent you need |
|---|---|---|
| Sales or marketing | Pitching a demo to a new contact | Prior express written consent |
| Informational | Confirming a booked meeting | Prior express consent |
| Callback the person requested | “Call me about pricing” on a form | Consent captured on that form, written if the call sells |
| Cold list, no relationship | Purchased contacts | Don’t call with an AI voice |
Written consent means a clear, signed agreement (the FCC’s rules allow electronic signatures) that the person agrees to receive calls from you using an artificial voice. Your counsel should write the wording. Your job is to make sure it’s captured, stored with a timestamp and tied to the phone number the agent dials.
Here’s what we see go wrong most: the consent language exists on the demo form, but not on the webinar form, the chat widget or the trade show scanner. The AI agent then calls everyone in the CRM the same way. Map every source of leads before the agent dials anyone.
Not sure your consent capture covers AI calls?
Book a free 30-minute consultation. We'll review your lead sources and calling use cases, and show you how a compliant AI calling setup would work for your team.
What the agent has to say on every call
Consent gets you permission to dial. The FCC’s rules also govern what happens once someone picks up. For artificial-voice calls, the agent has to:
- Identify the business clearly at the start of the call
- Give a callback number for your business during or after the message
- Offer an automated opt-out on telemarketing calls, by voice or keypress, within two seconds of that identification
- Honour the opt-out by adding the number to your internal do-not-call list straight away
On top of the legal minimum, we always have the agent say it’s an AI and that the call is recorded. Some states require all-party consent to recording, and buyers trust a call more when it doesn’t pretend. Always offer a human on request.
Do-not-call lists, calling hours and B2B
Three more rules trip teams up.
Do-not-call. Sales calls to numbers on the National Do Not Call Registry are restricted, and the FCC’s rules extend that protection to wireless numbers. Scrub against the national list and your internal list before every outbound dial, not once a quarter.
Calling hours. Telephone solicitations to residential subscribers are barred before 8 am or after 9 pm in the called person’s local time. Set the window by the prospect’s time zone, not your office’s. Some states are stricter.
B2B isn’t a free pass. Plenty of teams assume the TCPA only protects consumers. But the artificial-voice rule covers any call to a mobile number, and your buyers answer their mobiles. If your CRM can’t tell a direct office line from a personal mobile, treat every number as a mobile.
A compliant AI calling setup, step by step
Here’s the order we work in when we set up an outbound or callback agent.
List every form, chat, event and import that creates a contact the agent might call.
Add counsel-approved AI calling consent to each source and store it with a timestamp.
The agent only dials records with valid consent, a clean DNC check and an open calling window.
Business name, AI and recording notice, callback number and opt-out, every time.
Consent source, call recording, transcript and opt-outs written back to the CRM record.
The gate in step three is the part that matters most. Compliance shouldn’t depend on someone remembering to filter a list. It should be a hard rule in the system: no consent field, no call.
Before your AI agent makes its first outbound call
- Consent wording approved by counsel on every lead source
- Consent timestamp and source stored on the contact record
- National and internal DNC scrub before each dial
- Calling window set by the prospect's local time
- Business identification and AI disclosure in the opening line
- Working opt-out that updates the internal DNC list instantly
- Escalation to a human for pricing, legal and complaints
Where AI calling makes sense today
If cold lists are out, where’s the value? In the calls people asked for. A demo request at 9 pm that gets a callback in under a minute is welcome, consented and far more likely to turn into a meeting. Our post on speed to lead explains why the first few minutes matter so much.
The other strong uses are inbound calls, where the person called you, and follow-ups with contacts who’ve already given consent: meeting confirmations, event follow-ups and reactivation of past leads whose consent is on file. We cover the inbound side in our AI inbound call agent setup guide, and the wider picture in our guide to AI voice agents for sales.
Our view: if your plan only works by calling people who never opted in, the plan is the problem, not the technology.
The bottom line
AI cold calling to a purchased list isn’t legal in the US without consent, and sales calls need that consent in writing. Build the consent capture, DNC checks, calling windows and disclosures into the system first, then point the agent at callbacks and consented follow-ups. That’s where AI calling earns its keep, and where it stays out of trouble.
Sources
- FCC, Declaratory Ruling FCC 24-17 on AI-generated voices (Feb 2024)
- 47 U.S.C. § 227, Telephone Consumer Protection Act (Cornell LII)
- 47 CFR § 64.1200, FCC telemarketing and artificial voice rules (Cornell LII)
Want this working for your team?
Book a free 30-minute consultation. We’ll review your current setup and use cases, show you how it would work for you, and leave you with a recommended plan.