AI Phone Calls Can Cost You $1,500 Each If You Do Them Wrong

You’ve probably seen the ads by now. AI that answers your business phone. AI that calls your customers to follow up. AI that cold-calls leads and books appointments while you sleep. The technology is real, and it’s getting very good very fast.

But here’s what those AI tool vendors aren’t rushing to tell you: the FCC ruled in February 2024 that AI-generated voices are “artificial” under the Telephone Consumer Protection Act. That single ruling changed everything. And most small business owners using these tools have no idea they’re exposed.

The $1,500-Per-Call Problem

The TCPA — the federal law governing phone calls — has been around since 1991. It was written to stop robocalls. But the FCC’s 2024 ruling (FCC 24-17) explicitly brought AI voice technology under that same umbrella. If you use an AI to call someone’s cell phone without proper consent, each call is a violation. The penalty is $500 per call. If a court decides you did it knowingly, that triples to $1,500 per call.

There is no cap on damages. None.

Let me put that in real numbers. Say you set up an AI tool to call 200 past customers about a spring special. Sounds reasonable. Good business. But if you didn’t get the right kind of written consent from those 200 people, you’re looking at $100,000 to $300,000 in potential liability. For one afternoon of calls.

$500 to $1,500 per unauthorized AI call. No cap. TCPA class action lawsuits surged 95% in 2025. Lawyers are actively hunting for these cases.

The Biggest Misconception: “They’re My Customers”

This is the one that catches people. You fixed someone’s water heater last year. They’re your customer. Of course you can call them, right?

Wrong. At least, not with AI.

There’s a legal concept called the “established business relationship” (EBR). It used to be a safe harbor — if someone did business with you in the last 18 months, you could call them. But the EBR exemption does not apply to calls made with artificial or prerecorded voices to cell phones. Since the FCC now classifies AI voices as “artificial,” your existing customer relationship doesn’t protect you.

Let me say that again, because this is the part that trips up every business owner I talk to: having done work for someone does not give you permission to have an AI call their cell phone.

The only thing that protects you is Prior Express Written Consent — PEWC. That’s a specific legal standard. The customer has to sign (or check a box on) a disclosure that clearly says you might contact them using automated or AI-generated calls. And the checkbox can’t be pre-checked. They have to actively opt in.

What About Answering Inbound Calls?

Here’s the good news. Using AI to answer calls that your customers make to you is a completely different legal situation. The customer initiated the call. They called your number. Having an AI agent answer that call and help them — that’s fine. The TCPA regulates outbound calls, not how you answer your own phone.

This is why AI answering services (like the one I wrote about in my article on missed calls) are on solid legal ground. The customer called you. You answered. The fact that an AI answered instead of a human doesn’t trigger TCPA liability.

The risk is entirely on the outbound side. And that’s exactly where most of these AI voice tools are marketing hardest.

The AI Disclosure Requirement

Even when you do have proper consent to make an outbound AI call, there’s another requirement many people miss: you must disclose that the caller is AI. At the start of the call. Every time.

This isn’t just a best practice — it’s becoming law. The FCC requires it at the federal level, and states like California (SB 1001), New York, and Washington have their own AI disclosure laws with additional penalties. Your AI agent can’t pretend to be a person named “Sarah from Smith Plumbing.” It needs to say something like: “Hi, this is an AI assistant calling on behalf of Smith Plumbing.”

Done right, this doesn’t hurt response rates. Customers appreciate the honesty. Done wrong — or not done at all — it’s another violation on top of everything else.

What You Can Do Safely (And What You Can’t)

Let me be specific, because the line between legal and illegal is sharper than most people realize.

Safe (with proper setup)

• AI answering your inbound calls. Customer called you. You’re answering. No TCPA issue.
• AI calling customers who gave written consent. They checked the box on your booking form, your estimate form, or your website. The consent is documented. You’re covered.
• AI calling to confirm or remind about existing appointments. These are “transactional” calls, not marketing. Lower consent threshold. Still need to disclose AI.
• Sending texts to customers who opted in. Same PEWC rules apply to texts, but SMS opt-in is well-understood and common.

Dangerous

• AI calling past customers without written consent. Even if you installed their water heater last year. EBR doesn’t cover AI calls to cell phones.
• AI calling people from a purchased lead list. This is cold calling with AI. Full TCPA exposure. Do not do this.
• AI calling neighbors after a job (“We just did work on your street”). These are people you have no relationship with. This is the highest-risk activity possible.
• AI leaving voicemails without disclosure. A voicemail from an AI voice that doesn’t identify itself as AI is a violation.
• Using any tool that doesn’t check the Do Not Call list. The national DNC list is a federal requirement. Your AI tool needs to scrub against it before every campaign.

The line is simple: if the customer called you, AI is fine. If you’re calling them, you need written consent — and your AI needs to say it’s AI.

The Vendor Problem

Here’s what frustrates me about this space. The AI voice vendors — and there are dozens of them now — are selling the dream of automated outbound calling. “Your AI calls 1,000 leads while you sleep.” “Reactivate dormant customers automatically.” “Never miss a follow-up again.”

The technology delivers on those promises. The compliance infrastructure usually doesn’t. Most of these tools don’t check the DNC list. They don’t track consent status per contact. They don’t verify that your customer list actually opted in. They make it easy to do the thing, and leave the legal risk entirely on you.

One vendor in this space — Air.ai — was hit with a $19 million FTC settlement in 2025 for exactly this kind of behavior. But the businesses that used their tool? They’re still liable for the calls that were made from their numbers.

The Right Way to Use AI Voice

AI voice technology is genuinely powerful for small businesses. I build these systems. But I build them with guardrails, and those guardrails aren’t optional — they’re the foundation.

Here’s what a compliant AI voice setup looks like:

• Consent collection built into every intake point. Your booking form, your estimate form, your website contact form — all include an unchecked opt-in checkbox with clear disclosure language. This takes five minutes to add and eliminates your biggest risk.
• A consent database. Every customer has a consent status: opted in, opted out, or no consent on file. The AI checks this before every outbound call. No consent = no call. Period.
• DNC list checking. Before any campaign, the system scrubs your call list against the national Do Not Call registry. This is a legal requirement, not a nice-to-have.
• AI disclosure at call start. Every outbound call begins with: “Hi, this is an AI assistant calling on behalf of [your business].” Clear, honest, and legally required.
• Time-of-day enforcement. No calls before 8 AM or after 9 PM in the customer’s time zone. The system enforces this automatically.
• Opt-out on every call. “If you’d like us not to call again, just say so.” And when they do, the system honors it immediately and permanently.

With these guardrails in place, AI voice becomes one of the most effective tools a service business can use. Post-job feedback calls that customers actually appreciate. Maintenance reminders that generate repeat business. Appointment confirmations that reduce no-shows. All automated, all compliant, all documented.

Without the guardrails, you’re one class-action lawsuit away from a very bad year.