AI voice and the rules: what US contractors should know
Why inbound answering is the safe side of the line, where TCPA exposure actually begins, and the disclosure habits worth adopting regardless.
This is a plain-English orientation, not legal advice. If you are making decisions with real money attached, run them past a lawyer who knows telecom rules in your state.
That said, the shape of the rules is simple enough that most contractors can understand it in five minutes, and understanding it removes the main reason people hesitate.
The line that matters: inbound versus outbound
Nearly all of the regulatory risk around AI voice lives on the outbound side — you calling them. Almost none of it lives on the inbound side — them calling you.
The reason is the Telephone Consumer Protection Act. Its restrictions attach to calls placed to consumers using artificial or prerecorded voices. In 2024 the FCC confirmed that AI-generated voices count as “artificial” for this purpose, which put automated outbound AI calling squarely inside a rule with statutory damages of $500 to $1,500 per call. Per call. Multiplied across a list, that arithmetic ends companies.
When a customer dials your number and your system answers, none of that applies. They initiated the call. There is no dialing, no list, no consent question.
This is why we only do inbound. It is not caution for its own sake — it is that the outbound version of this product is a liability we are not willing to sell to a contractor.
What that means in practice
Safe, and what we build:
- Answering your inbound line, day or night
- Booking appointments for people who called you
- Sending an SMS confirmation for an appointment the caller just made
- Warm-transferring to your team
Not safe, and we will decline to build it:
- AI cold calling a purchased lead list
- Automated “just checking in” campaigns to past customers
- Robocalled promotions of any kind
The middle ground — calling back a customer who explicitly asked for a callback — is defensible but fact-specific. Get advice before automating it.
Disclosure: do it even where it is not required
Federal law does not currently require an inbound AI agent to announce itself unprompted. Several states are moving in that direction, and the direction of travel is obvious.
More importantly, disclosure is simply better business. A customer who discovers mid-call that they were misled will not book, and may leave a review saying so. A customer who is told plainly, and gets their appointment in ninety seconds, does not care at all.
The habits worth having:
- Never claim to be human. If asked directly, answer directly: “I’m an AI assistant for Summit Air — I can book you in right now, or put you through to the team.”
- Introduce by business name. The caller should know who they reached in the first sentence.
- Offer the human exit. Any caller who wants a person should get one, or a callback commitment with a time on it.
- Never take card details. Payment information should not pass through an automated voice system, full stop.
- Never guess a repair price. Quote your published dispatch fee and nothing more.
Every one of those is baked into our default script, and none of them are configurable downward.
Call recording
Recording rules are state law, not federal, and they split into two camps:
- One-party consent — the majority of states. You can record because you are a party to the call.
- Two-party (all-party) consent — including California, Florida, Illinois, Pennsylvania, Washington and several others. Every party must be informed.
The workable answer is to just announce it everywhere: “This call may be recorded for quality.” One sentence, no state-by-state logic, no exposure. If you operate near a state line, this is not optional housekeeping.
The UK, briefly
If you serve UK customers, the relevant regime is PECR alongside UK GDPR, enforced by the ICO. The maximum penalty rose sharply in early 2026, so the stakes there are now comparable to the US. The same principle applies: inbound answering is straightforward, automated outbound marketing is where the enforcement lives.
The short version
Answering your own phone with AI is a well-lit, low-risk area. Calling strangers with it is not. Every rule you would want to follow anyway — say who you are, tell the truth, do not take card numbers, do not invent prices — happens to also be the compliant path.
If a vendor offers you AI cold calling and waves off the question, that is the whole review you need of that vendor.
Stop losing the calls behind this problem.
Forward your line tonight and see what a 24/7 answer rate does to next week’s schedule. Fourteen days free, no contract.
