Is it legal for AI to call a business? (US)

Generally yes, with rules. Since 2024 the FCC treats AI voices as artificial voices under the TCPA, which means calls must identify who is calling and calls to cell phones need prior consent. Honest disclosure, a callback number, opt-outs and call caps are the practical safeguards. This is not legal advice.

The short answer

In the US, it is generally lawful for an AI assistant to call a business on your behalf about your own errand, as long as it follows the rules that apply to artificial-voice calls. Those rules come mostly from the Telephone Consumer Protection Act (TCPA) and FCC regulations, plus a patchwork of state laws. This page summarizes them as of September 2026. It is not legal advice.

The FCC's 2024 ruling: AI voices count as artificial voices

In February 2024 the FCC issued a declaratory ruling (FCC 24-17) saying AI-generated voices are "artificial or prerecorded voice" under the TCPA. It declined to carve out technology that claims to act like a live agent. So an AI voice assistant is treated the same as a prerecorded message for TCPA purposes, even when it holds a real conversation.

Three things follow from that.

1. Calls to cell phones need prior express consent

Under 47 USC 227(b)(1)(A)(iii), artificial-voice calls to wireless numbers need the called party's prior express consent unless it's an emergency. This applies to informational calls, not only marketing, and to business owners, not only consumers.

This matters because plenty of small businesses (a solo groomer, a contractor, a mobile mechanic) list a cell phone as their business line. For informational calls, the consent needed is non-written, and the FCC has long treated knowingly publishing a number as some evidence of consent to be called about related matters. Whether that reasoning holds for AI voices is unsettled.

2. Business landlines are mostly outside the residential rule

The TCPA's rule on artificial-voice calls to residential lines (227(b)(1)(B)) doesn't generally reach business landlines. Most calls to a restaurant's front desk or a store's main number fall here.

3. The call has to say who's calling

47 CFR 64.1200(b) requires artificial-voice calls to state, at the start, the identity of the entity responsible for initiating the call, and to provide a telephone number. Who counts as the "caller" when a platform dials at a user's request is not settled. The safe approach is to name both.

The FCC's proposed AI rules

In August 2024 the FCC proposed rules that would define AI-generated calls, require AI disclosure at the start of the call, and add a separate consent for AI calls, while protecting uses by people with disabilities. As of mid-2026 those rules were still not final.

State bot and disclosure laws

• California has the BOT Act (SB 1001, 2019), which covers bots used to sell things or influence votes, and AB 2905 (effective January 1, 2025), which covers artificial-voice disclosure on automated calls.
• Utah requires generative AI to say it's AI when asked, and upfront in regulated professions.
• Colorado, Illinois and Florida have their own variants.
• Texas SB 140 (2025) is often described by vendor blogs as requiring AI disclosure within 30 seconds, but the enrolled text reportedly doesn't contain that provision.

The practical rule that satisfies all of them: say it's an AI, and who it's calling for, in the first sentence of every call.

Recording and transcription consent

About a dozen states require all parties to consent to recording: California, Connecticut, Florida, Illinois, Maryland, Massachusetts, Montana, Nevada, New Hampshire, Pennsylvania and Washington, plus Delaware and Michigan under some readings. Some statutes treat a live transcript as a recording, and California has active litigation on this under CIPA. If you record, announce it. If you don't announce it, don't keep audio for calls into those states.

Speaking for someone else

When an assistant cancels, disputes or negotiates for you, it's acting as your representative. Banks, carriers and insurers often require the account holder to verify identity, and privacy laws like HIPAA and GLBA stop reps from sharing data without it. The clean fix is to have the assistant wait on hold and then patch you in. See hold and patch in.

Don't claim to be human, don't clone voices

Pretending an AI is a person, or a person is an AI, is where reputations go to die. Cloning a user's voice adds deepfake exposure under FCC and state rules.

How CallForMe handles this

• Disclosure in the first sentence. Every call opens with "Hi, I'm an automated assistant calling on behalf of" the name you give. If asked, it says it's an AI assistant. It never claims to be a person.
• A real callback number. Calls come from +1 (469) 966-0062, and you can pass your own callback_number for the business to reach you.
• Businesses only. No calls to individuals, no telemarketing, no harassment, no impersonation.
• Opt-outs honored for everyone. If a business says don't call, that number goes on an opt-out list that applies across all users.
• Per-destination caps. At most 3 calls to the same number per account per day, and 10 per number per day across all users.
• Blocked numbers. No 911, 988, 211, 311 or other N11 numbers, and no 900 or 976 premium numbers.
• No sensitive data over the phone. The assistant never gives card numbers, Social Security numbers or passwords.

More in safety and disclosure.

Questions