Key takeaways
- Utah requires a business using generative AI in a consumer transaction to disclose it when the person clearly asks, with a safe harbor for disclosing at the outset[1].
- The FCC ruled in February 2024 that AI-generated voices are artificial under the TCPA, so outbound AI calls follow the same consent rules as prerecorded calls[2].
- Recording consent is set state by state: one-party under federal law, all-party in states such as California[3].
- The safest national default is a short disclosure in every greeting plus an easy route to a person.
Inbound and outbound calls are different legal questions
Restaurants use AI on the phone in two ways. Inbound: the assistant answers when a guest calls to book, order or ask about hours. Outbound: the system calls or texts guests, for example to confirm a reservation or announce a holiday menu. Disclosure laws mostly bite on the first; telemarketing and consent rules mostly bite on the second. Recording rules apply to both.
This guide reflects the statutes and rulings we checked on 7 November 2026. State AI law is moving quickly, with new bills in most legislative sessions, so use it as a map and confirm the current position with counsel before you rely on it, especially if you operate in several states.
The rules at a glance
| Rule | What it covers | When it applies | What to do |
|---|---|---|---|
| Utah S.B. 226 (2025) | Generative AI interacting with people in consumer transactions | Disclose when the person clearly asks whether they are talking to AI; proactive disclosure for regulated occupations in high-risk interactions | Disclose at the start of every call to use the safe harbor |
| Colorado S.B. 24-205 | AI systems deployed to interact with consumers | Business doing business in Colorado that makes such a system available | Disclose to every caller; confirm the current effective date |
| FCC declaratory ruling (Feb 2024) | AI-generated voices in calls | Outbound calls using an AI voice are artificial-voice calls under the TCPA | Get the consent the TCPA requires before AI outbound calls |
| Federal wiretap law | Recording calls | One-party consent | Disclose recording anyway if any caller may be in an all-party state |
| All-party consent states (for example California) | Recording calls | Every party must consent | Say the call is recorded at the start and let callers opt out |
Utah: disclose when asked, or at the outset for safe harbor
Utah's 2025 amendments (S.B. 226) require a supplier that uses generative AI to interact with an individual in a consumer transaction to disclose that the person is interacting with generative AI and not a human if the person asks or otherwise prompts about it, and the request must be clear and unambiguous[1]. Stricter proactive disclosure applies to regulated occupations in high-risk interactions[1]; restaurants are not usually in that category.
The useful part for operators is the safe harbor: a business is not subject to enforcement under that section if its AI clearly and conspicuously discloses, at the outset and throughout the interaction, that it is generative AI, not human, or an AI assistant[1]. The division can impose an administrative fine of up to 2,500 dollars for each violation[1]. A greeting such as this restaurant's virtual assistant meets the spirit of the safe harbor; confirm the wording with counsel.
Colorado: disclosure as part of a wider AI law
Colorado's AI Act, S.B. 24-205, was signed on May 17, 2024[4]. Alongside duties for high-risk AI systems, it requires a business that deploys or makes available an AI system intended to interact with consumers to ensure disclosure to each consumer who interacts with it[4]. The legislature has since revisited when the act's requirements take effect, so check the current date and any amendments before you plan around it.
For a restaurant phone assistant the practical answer is the same as in Utah: tell every caller at the start. If you take calls from Colorado guests, even from a location over the state line, assume the rule may reach you.
Outbound calls: the FCC ruling on AI voices
On February 8, 2024, the FCC unanimously adopted a declaratory ruling that calls made with AI-generated voices are artificial under the Telephone Consumer Protection Act[2]. The FCC's release notes that the TCPA requires telemarketers to obtain prior express written consent before robocalling consumers, and that the ruling holds AI-generated voices to the same standards[2].
That matters if you plan AI calls to guests: holiday promotions, win-back calls, even some reminder calls. Consumers can sue under the TCPA for actual losses or 500 dollars per violation, and courts can treble damages for willful violations[5]. Telephone solicitations to residential subscribers are also barred before 8 a.m. or after 9 p.m. local time[6]. If your outbound use is anything more than a transactional message the guest asked for, get advice before launching.
Recording consent: the rule that already applies
Most AI phone tools record or transcribe calls. Federal wiretap law is one-party consent, but some states, such as California, require the consent of all parties[3]. Because callers can be anywhere, a recording notice in the greeting is the simplest national standard. Keep the notice short: long legal preambles cause hang-ups and do not add protection.
What clear disclosure looks like in practice
Utah's safe harbor asks for disclosure that is clear and conspicuous at the outset and throughout the interaction[1]. On a phone call, that means saying it in the first sentence, in plain words, before collecting any details. It also means the assistant must answer honestly every time a caller asks, even late in the call, and never claim to be a person or use a human name in a way that suggests it is one.
- Use plain words: virtual assistant, automated assistant or AI assistant. Avoid vague phrases such as digital concierge on their own.
- Place the disclosure before the first question, not after the booking is taken.
- Repeat it when transferring: say you are connecting the caller to a team member, so the change is obvious.
- Log the greeting version with each call, so you can show what was said.
Chat and text assistants count too
Utah's definition of generative AI covers systems that simulate human conversation through text, audio or visual communication[1], so the same thinking applies to website chat and two-way texting. Put a short line at the top of the chat window, such as You're chatting with our automated assistant, and show a clear option to reach staff. For two-way texting, identify the restaurant and the assistant in the first message of each new thread, and make sure replies such as STOP or HELP work.
Why disclosure is good practice even where it is not required
In Pew Research Center's survey, 76% of Americans said it is extremely or very important to be able to tell whether content was made by AI or by people[7], and 50% said they are more concerned than excited about AI in daily life[8]. A guest who discovers halfway through a call that they were talking to software feels misled; one who was told up front usually does not mind if the call goes well.
Source: Pew Research Center, 2025[7]
Show the data
| Item | Value |
|---|---|
| Say telling AI from people matters | 76% |
| More concerned than excited about AI | 50% |
Scripts that cover the main rules
Inbound greeting
Thanks for calling Harbor Grill. I'm the restaurant's virtual assistant, and this call is recorded so our team can follow up. I can take a reservation request, share our hours, or connect you with someone.
If the caller asks whether it is a person
No, I'm an automated assistant for Harbor Grill. Would you like me to connect you with a team member, or take a message for a callback?
If the caller objects to recording
No problem. I'll connect you with a team member now, or you can call back on our direct line during opening hours.
Outbound confirmation text the guest requested
Harbor Grill: your table request for 4 on Fri at 7:30 pm is confirmed. Reply C to cancel. Reply STOP to opt out of texts.
These scripts are a starting point. Names and times are an Example; keep the structure: identity, recording notice, a route to a person.
A compliance checklist for multi-state operators
- List every state where you have locations and the states your callers commonly come from.
- Put an AI and recording disclosure in every greeting, chat window and automated text, regardless of state.
- Keep a dated copy of each greeting version so you can show what callers heard on a given day.
- Separate transactional messages from marketing, and get written consent before any AI-voice outbound marketing call.
- Ask your vendor how it handles callers who object to recording or ask for a human, then test both paths yourself.
- Review the list each January and July; state legislatures often pass AI bills in spring sessions.
Disclosure is one part of a sound setup. Pair it with the vendor questions in AI and guest privacy, the business case in measuring ROI on AI phone and chat tools, and the same honesty standard for written content covered in AI drafts for review replies. This article is general information, not legal advice.
Frequently asked questions
Do I have to tell callers they are talking to AI?
In Utah you must disclose when a caller clearly asks, and disclosing at the outset gives you a safe harbor[1]. Colorado's AI Act also includes a disclosure duty for AI systems that interact with consumers[4]. Elsewhere it is good practice. A one-line disclosure in the greeting covers you in most cases.
Can I use an AI voice to call guests about specials?
Treat it as an artificial-voice call under the TCPA, following the FCC's 2024 ruling[2]. Marketing calls generally need prior express written consent, so get legal advice before launching.
Is a reservation confirmation text marketing?
A text the guest asked for that only confirms their booking is transactional. Adding promotions can change that, so keep confirmations clean and send marketing separately with consent.
Which states require all-party consent to record calls?
Several, including California[3]. Because callers travel, the practical approach is to announce recording on every call.
References
Utah State Legislature (2025). S.B. 226 Artificial Intelligence Consumer Protection Amendments (enrolled copy) (opens in a new tab) ↑
Federal Communications Commission (2024). FCC Makes AI-Generated Voices in Robocalls Illegal (news release) (opens in a new tab) ↑
Legal Information Institute (Cornell Law School), US Code (2026). 18 U.S. Code § 2511 - Interception and disclosure of wire, oral, or electronic communications prohibited (opens in a new tab) ↑
Colorado General Assembly (2024). SB24-205 Consumer Protections for Artificial Intelligence (opens in a new tab) ↑
Legal Information Institute (Cornell Law School), US Code (2026). 47 U.S. Code § 227 - Restrictions on use of telephone equipment (opens in a new tab) ↑
Legal Information Institute (Cornell Law School), eCFR text of FCC rule (2026). 47 CFR § 64.1200 - Delivery restrictions (opens in a new tab) ↑
Pew Research Center (2025). How Americans View AI and Its Impact on People and Society (opens in a new tab) ↑
Pew Research Center (2025). How Americans View AI and Its Impact on People and Society (opens in a new tab) ↑