What Utah law makes your AI say when it picks up

Utah passed the Artificial Intelligence Policy Act in 2024 and it originally read much broader. SB 226, in the 2025 session, narrowed it and moved the consumer-facing disclosure rules into their own chapter: Title 13, Chapter 77, Generative Artificial Intelligence — Consumer Disclosures and Enforcement, effective May 7, 2025. If you have heard that Utah requires you to announce AI on every call, you heard the 2024 version.

Read the two nouns before you decide this is about somebody else. Both are borrowed from the Utah Consumer Sales Practices Act. A 'supplier' is a person who regularly engages in consumer transactions. A 'consumer transaction' is a sale of goods or services to a person for primarily personal, family, or household purposes. A homeowner buying a furnace repair is the textbook case. You are the supplier. This chapter is about your phone.

Subsection (1)(b) then adds the limit that does most of the work: the caller's question has to be a clear and unambiguous request to determine whether the interaction is with a human or with artificial intelligence. 'Wait, are you a real person?' clears that bar. A homeowner sighing 'is this a machine' at a hold menu is a harder call, and you do not want to be the shop litigating where the line sits.

13-77-103(2) is the part most vendors skip, because it does not depend on the customer asking anything. An individual providing services in a regulated occupation has to prominently disclose that the person receiving services is interacting with generative AI — but only if that use is a 'high-risk artificial intelligence interaction.' And subsection (3) sets the timing: verbally at the start of a verbal interaction, and in writing before the start of a written one.

So the real question is whether an intake call is high risk. 13-77-101(5) defines it three ways, and you should read them against what your phone actually does rather than against what you intend it to do.

Here is the honest read. A plain booking call — name, address, what's broken, when can we come — probably is not a high-risk interaction under any of those clauses. The instant that call collects a card number, the financial-data clause is squarely in scope, and the disclosure becomes proactive and mandatory rather than on-request. We know of no rule or enforcement action that has tested where a trade intake call falls, and anyone telling you otherwise is guessing with more confidence than the record supports.

Which is the argument for not litigating it in your own head. Take the safe harbor. Have the agent say what it is at the outset, keep saying it if asked, and the distinction between 13-77-103(1) and 13-77-103(2) stops mattering to your business.

The failure mode is not omission. It is a legalistic paragraph that burns the ten seconds you had. Nobody standing in water needs to hear about Title 13. They need to hear who picked up, that it can actually help, and that it is software — in that order, fast.

On recording: Utah Code 77-23a-4 makes it lawful to record a conversation you are a party to, without telling anyone else. Your business is a party to a call placed to your own number, so recording an inbound call is legal in Utah with no announcement at all. That is not the whole picture. Roughly a dozen states require every party to consent, and a caller reaching your Utah shop from a phone in California or Washington can pull their state's rule into the conversation. The cheap answer is to announce it every time, in one clause, and stop tracking which caller is where.

This is the distinction that decides whether an AI phone product is a nothing-burger or a compliance project, and almost nobody selling one draws it clearly. Answering a ringing phone is not regulated by the TCPA. Dialing a homeowner with a synthetic voice is.

• Inbound — the customer dialed you. Utah Chapter 77 applies to what you say. The TCPA's artificial-voice restrictions do not apply, because you did not initiate the call. This is the entire missed-call use case, and it is the clean one.
• Outbound callbacks with a synthetic voice — you initiated. The FCC's February 8, 2024 declaratory ruling put AI-generated voices inside the TCPA's 'artificial or prerecorded voice' restriction, which means prior express consent, identification of who is calling, and an opt-out on anything that counts as telemarketing.
• Outbound texts — a different rulebook again, with A2P 10DLC registration on top. If a vendor offers to text your customers on day one without asking about carrier registration, that is a tell.
• Review requests — the FTC's rule on fake and misleading reviews governs what you may ask for and what you may offer in exchange. We wrote that one up separately; the short version is that the gating is on incentives and selective solicitation, not on automation.

The practical consequence for a Utah shop: an AI that answers your line is a narrow legal question with a two-second answer. An AI that calls your list back is a project. Sequence them in that order, and be suspicious of anyone who quotes you one price for both.

We are not going to pretend that population growth is a call-volume statistic. Nobody publishes what share of Utah trade calls go unanswered, including the vendors who quote a number for it. What the growth data does establish is direction: the same crew, more rooftops, and no more hours in the day than you had in 2023.

One Utah-specific overlay worth naming while you are writing the greeting anyway. Salt Lake County is 20.6% Hispanic or Latino, and 160,965 county residents speak Spanish at home. Most shops here already have Spanish on the crew and none on the phone. If you are rewriting your greeting to satisfy 13-77-104, you are already in the file — add the Spanish clause in the same edit.

1. Read Chapter 77 yourself. It is four short sections and you will finish it in ten minutes. Everything a vendor tells you about Utah AI rules should be checkable against that page, and if it isn't, that tells you something about the vendor.
2. Write one disclosure sentence and put it first. Not after the shop name and the hours and the menu — first, inside the greeting, in the same breath. If it takes longer than two seconds to say, it is too long and it will get cut.
3. Find every place the agent could touch a card, a deposit, or a health detail, and re-disclose there. That is the financial-data trigger, and it is the one place where the on-request standard is not enough.
4. Decide the handoff before you need it. "I'm an AI assistant" is only reassuring if the next sentence is a real option. Name who the caller gets transferred to and how fast, and confirm what happens at 2am when that person is asleep.
5. Call your own published number and ask the question. Not the office line — the one on your Google Business Profile and your truck. Say the words: am I talking to a real person? Then call again at 9pm and do it a second time. If the answer is anything other than an immediate, plain yes-it's-AI, you have found the gap.
6. Write down what you changed and when. One paragraph in a file. The safe harbor turns on what the AI actually says at the outset and throughout — being able to show when that wording went live is worth more than remembering that it did.

Do the test calls before you buy anything, from us or anyone else. If your line already forwards to a human who answers at 9pm, this whole article is a filing exercise for you and you should spend the hour somewhere else. If it went to voicemail — that is the actual problem, and the disclosure sentence is the easy part of fixing it. Either way, the statute is free to read and your own phone is free to call.