Is It Legal to Have AI Answer Your Business Phone, and Do You Have to Tell Callers?
Yes. Having AI answer your inbound business phone is legal, and no federal rule currently requires you to announce that the voice is AI. The federal law contractors are afraid of, the TCPA, restricts calls a business makes. It does not restrict calls a business answers. The rules that do apply to you are call recording consent and a small number of state AI disclosure laws, and both turn on where your caller is sitting.
This is not legal advice. It explains how the rules are written and where the real exposure sits. Confirm it with your own attorney before you rely on it.
Why does everyone think this is a TCPA problem?
In February 2024 the FCC ruled that AI-generated voices count as "artificial" under the Telephone Consumer Protection Act. The ruling is real. What tends to get dropped in the retelling is what the ruling attaches to.
What did the FCC actually rule in February 2024?
On February 8, 2024, the FCC issued a unanimous Declaratory Ruling (FCC 24-17) confirming that an AI-generated voice is an "artificial" voice for TCPA purposes. The Commission described its own ruling as reaching AI technology that initiates an outbound telephone call using an artificial or prerecorded voice.
The word "outbound" is carrying the entire ruling.
What does the statute actually say?
The TCPA is 47 U.S.C. 227. The two provisions in question read like this:
- 227(b)(1)(A) makes it unlawful "to make any call" using an automatic telephone dialing system or an artificial or prerecorded voice to a cell phone, absent prior express consent.
- 227(b)(1)(B) makes it unlawful "to initiate any telephone call to any residential telephone line using an artificial or prerecorded voice to deliver a message," absent prior express consent.
Make. Initiate. Both verbs describe the party placing the call. Neither one reaches the party picking up.
How do the two categories compare?
| Homeowner dials your shop | Your system dials the homeowner | |
|---|---|---|
| Who placed the call | The homeowner | You |
| Statutory verb that applies | None. You did not make or initiate a call. | "Make" 227(b)(1)(A) or "initiate" 227(b)(1)(B) |
| Prior express consent required | No | Yes, for an artificial or prerecorded voice |
| Covered by the Feb 2024 AI ruling | No | Yes |
| Exposure | Recording consent and state AI disclosure rules only | $500 per call under 227(b)(3), up to $1,500 at the court's discretion for a willful or knowing violation |
The arithmetic in the right column is why the fear exists, and there the fear is legitimate. One AI voice campaign dialing 400 numbers without consent is 400 violations:
- 400 x $500 = $200,000
- Willful or knowing, trebled: 400 x $1,500 = $600,000
Now run the left column. Four hundred homeowners who dialed you: zero violations, because you never made a call.
When does a contractor actually cross the line?
Here is the version that matters on a live phone desk.
Outside the ruling. A homeowner sees your truck wrap, your Google ad, or your Business Profile at 9:40 PM, dials, and an AI voice answers, asks what is failing, and books tomorrow morning. The homeowner initiated. You answered. Nothing in 227(b)(1) is triggered. That is the ordinary AI receptionist setup for contractors, and the category holds whether the assistant takes a message or creates the job in Jobber or Housecall Pro during the call.
Inside the ruling. You export 400 unsold estimates from your CRM and have a system dial them with an AI voice asking if they are ready to move forward. You initiated. That sits squarely inside 227(b)(1) and inside the February 2024 ruling. You need prior express consent, and that is a conversation with an attorney, not with a vendor.
Identical technology. Two legal categories, decided entirely by who dialed.
Worth noting where the volume actually lives in the category that is not regulated by the TCPA. Cactus reports that a plumbing and HVAC company in Utah booked about $150,000 in confirmed revenue in four months, and that about 58 percent of those captured leads came in after hours. That is one company's reported result, not an average across contractors, and every one of those calls came in on the inbound side. After-hours calls in home service are the calls most likely to go unanswered, and answering them raises no TCPA question at all.
Do I have to tell callers they are talking to AI?
There is no federal rule requiring it on an inbound call today.
The FCC did propose one. On August 7, 2024 it adopted a Notice of Proposed Rulemaking that would require a caller using an AI-generated voice to disclose that at the start of the call. Two things about it: the proposal targets outbound calls and texts, and as of August 4, 2026 it remains a proposal rather than a final rule. Confirm its current status before you rely on that.
State law is where the real answer lives, and only a few states have acted. Here is what each one actually says.
California: AB 2905 does not reach an inbound call
AB 2905 is the state law most often waved at contractors. It was signed September 20, 2024 and took effect January 1, 2025, and it does require an artificial voice disclosure. The scope is the part that gets skipped.
AB 2905 amends Public Utilities Code Section 2874, which opens: "Whenever telephone calls are placed through the use of an automatic dialing-announcing device, the device may be operated only after an unrecorded, natural voice announcement has been made to the person called by the person calling." The added requirement is that the announcement must "inform the person called if the prerecorded message uses an artificial voice," with artificial voice defined as "a voice that is generated or significantly altered using artificial intelligence."
Two conditions have to be met before any of that binds you. Calls have to be placed, and they have to be placed through an automatic dialing-announcing device, which Section 2871 defines as equipment that stores telephone numbers to be called or generates numbers to be called, and that can deliver a prerecorded message to the number called. A system that sits on your main line and waits for a homeowner to dial in does not store numbers to be called, does not generate numbers to dial, and does not place calls. It falls outside the definition.
One correction while we are here: some vendor pages attach a specific per-call dollar penalty to AB 2905. The chaptered bill text contains no dollar penalty amount. It relies on existing Public Utilities Commission enforcement. Do not price your risk off that number.
Utah: a real duty, triggered by the caller asking
Utah is the state that actually reaches inbound. The Utah Artificial Intelligence Policy Act, as amended by SB 226 in 2025, requires a business to disclose that a person is interacting with generative AI when the person makes a clear and unambiguous request to know whether they are dealing with a human. Your assistant has to answer that question honestly when a homeowner asks it.
Utah also requires disclosure up front, without being asked, in interactions the statute treats as high risk, which centers on collecting health, financial, or biometric information or giving advice a consumer would rely on for financial, legal, or medical decisions. Booking a furnace repair is not that. If you hold a Utah trade license, have your attorney confirm where your work sits. The statute also provides a safe harbor: a business is treated as compliant when the AI itself discloses that it is not human at the start of and throughout the interaction. In other words, the line we recommend below satisfies Utah on its own.
Colorado: the law people cite is no longer the law
Colorado's 2024 AI Act (SB 24-205) would have required telling consumers they were interacting with an AI system regardless of what the interaction was about. It never took effect. A federal magistrate judge blocked enforcement on April 27, 2026, and a replacement, SB 26-189, was signed May 14, 2026. The replacement drops the blanket AI interaction disclosure and applies to automated decision-making used in consequential decisions such as employment, lending, housing, insurance, and health care, effective January 1, 2027. Scheduling a service call is not on that list. Pages still citing the 2024 version are out of date.
Those three states are the ones we have verified. State AI legislation is moving every session, so treat this as a snapshot dated August 4, 2026 and have counsel check your own state before you decide not to disclose.
The practical answer
Disclose anyway, in every state. It clears Utah outright, it clears the FCC proposal if that rule is ever finalized, and it removes the worst outcome on the call itself: a homeowner who works out mid-call that they were fooled is a homeowner who does not book.
What about recording the call?
This is the part that genuinely applies to you everywhere, and it has nothing to do with AI. Nearly every AI answering system records and transcribes, and recording law is state law that predates AI by decades. Twelve states are commonly treated as requiring every party to consent:
| All-party consent states | ||
|---|---|---|
| California | Illinois | New Hampshire |
| Connecticut | Maryland | Oregon |
| Delaware | Massachusetts | Pennsylvania |
| Florida | Montana | Washington |
Three edge cases worth knowing:
- Nevada is one-party in person but all-party for telephone calls. Treat Nevada calls as all-party.
- Michigan's eavesdropping statute reads as all-party, but Michigan courts have recognized a participant exception since Sullivan v. Gray (1982).
- Oregon is all-party in person and one-party for electronic communications, so the table above is the conservative read there.
Cross-state calls. If your shop sits in a one-party state and the homeowner sits in an all-party state, the standard conservative practice is to follow the stricter state's rule. That is common practice rather than a settled universal rule, so confirm it with counsel if you plan to rely on it. You usually will not know where the caller is until the call is over, which is the real argument for disclosing on every call instead of sorting by area code.
What should the top of my call actually say?
One line, delivered before the caller starts describing the problem, because consent has to come before the recording:
"Thanks for calling [Company]. This call is recorded and you're speaking with an automated assistant. I can get a technician scheduled or connect you with someone on the team. What's going on over there?"
That line does three jobs. It obtains recording consent in all-party states, where a caller who keeps talking after hearing it is generally treated as having consented. It identifies the AI up front, which satisfies Utah and would satisfy the FCC proposal if that rule lands. And it names the escape hatch in the same breath, because the fastest way to lose a homeowner is to let them believe there is no path to a person. That handoff is the same machinery that routes a real emergency to a live tech.
Frequently asked questions
Is it legal to have AI answer my business phone? Yes. The TCPA restricts calls a business makes or initiates, not calls it answers. When a homeowner dials your number, you have not made a call. Recording consent law and a small number of state AI disclosure laws still apply.
Do I have to tell callers they are speaking with an AI? No federal rule requires it on an inbound call as of August 4, 2026. The FCC proposed a disclosure rule in August 2024, aimed at outbound calls, and it has not been finalized. Utah requires you to answer honestly if a caller asks whether they are talking to a human. Disclose on every call regardless.
Does California AB 2905 apply to my AI receptionist? It is written for outbound calling. AB 2905 amends Public Utilities Code 2874, which is triggered when calls are placed through an automatic dialing-announcing device, defined in Section 2871 as equipment that stores or generates numbers to be called and delivers a prerecorded message. A system that only answers calls does not meet that definition.
Which states require me to announce that the call is recorded? Twelve states are commonly treated as all-party consent: California, Connecticut, Delaware, Florida, Illinois, Maryland, Massachusetts, Montana, New Hampshire, Oregon, Pennsylvania and Washington. Treat Nevada phone calls the same way. Since you rarely know where a caller is, announce on every call.
Does the FCC's February 2024 AI ruling apply to my answering system? Not to the answering side. The Commission scoped it to AI technology that initiates outbound calls using an artificial or prerecorded voice. An AI that answers a homeowner who dialed you is a different category.
Can I use an AI voice to call my customers back? That is outbound and inside the ruling. You need prior express consent, and exposure runs $500 per call under 227(b)(3), up to $1,500 for a willful or knowing violation. Talk to your attorney first.
Sources
All read on August 4, 2026.
- FCC Declaratory Ruling FCC 24-17, released February 8, 2024: docs.fcc.gov/public/attachments/FCC-24-17A1.pdf
- FCC Notice of Proposed Rulemaking on AI-generated robocalls and robotexts, adopted August 7, 2024: fcc.gov/document/fcc-proposes-first-ai-generated-robocall-robotext-rules-0
- 47 U.S.C. 227: law.cornell.edu/uscode/text/47/227
- California AB 2905 chaptered text: leginfo.legislature.ca.gov (bill_id 202320240AB2905)
- California Public Utilities Code Sections 2871 and 2874: california.public.law
- Utah SB 226 (2025), amending the Utah Artificial Intelligence Policy Act: le.utah.gov
- Colorado SB 24-205 enforcement order of April 27, 2026 and replacement SB 26-189 signed May 14, 2026: mcdermottlaw.com AI law analysis
The next step
The legal question has an answer. The operational question, whether an AI holds up on a 9:40 PM no-heat call with a crying kid in the background, is not something you settle by reading about it.