General information, not legal advice. Rules change and depend on your state and your exact messages. Have a lawyer review anything you put into production.
AI makes it easy to text a lead back in seconds, call a customer to confirm a visit, or send a reminder at the right time. It also makes it easy to break a rule without noticing, because the tool will do whatever you set it to do.
This post sorts the rules into three piles: what's settled, what changed, and what's still pending. Everything is as of November 2026. We lean on the primary sources our longer guides already read, and we say where a point is unsettled. For the full texting rules, read our TCPA guide for contractors. For recording and AI disclosure, read AI phone agents and call recording laws. This post is the short, current version.
First, the idea behind the rules
The main federal law is the Telephone Consumer Protection Act (TCPA). In plain words: it limits calls and texts made with an autodialer or an artificial or prerecorded voice. The core rule is at 47 U.S.C. 227. Without the right consent, these calls and texts to a cell phone are unlawful, with exceptions such as emergencies.
Two things set how much consent you need. One is the technology you use. The other is what the message says. A message that only answers a customer's own request reads differently from one pushing a promotion, and the rules treat them differently. Marketing messages generally need written consent.
What's settled
AI voices count as "artificial." On February 8, 2024, the FCC announced a ruling that calls made with AI-generated voices are "artificial" under the TCPA. According to the FCC's announcement, this holds them to the same standards as other artificial or prerecorded voice calls. It's a clarification of an existing rule. It's not a ban on AI voices.
What that means in practice: if your AI voice calls a customer to sell something, that's a robocall that needs the consent the TCPA requires, which for marketing is prior express written consent. Our advice is simple. Don't have an AI voice place outbound calls to customers without clear consent, in writing, that covers that kind of call.
The rule applies to calls you place, not calls you answer. The FCC announcement frames the ruling around robocalls. An AI that answers a call a customer makes to you is a different situation. We didn't find a federal statement requiring a business to announce that an inbound caller is talking to AI. See our call recording and disclosure guide for the state-by-state picture we could confirm.
Opt-outs must be honored, fast. Under the current text of the FCC's rule at 47 CFR 64.1200, as summarized in our TCPA guide, a person can revoke consent by any reasonable method. Replying "stop" counts, and so can other plain words. You must honor the request within a reasonable time not to exceed ten business days. And you can't tell people they must use only one way to opt out. Set your AI to recognize "stop," "quit," "unsubscribe" and similar replies, and to stop sending.
What changed
The one-to-one consent rule is gone. In 2023 the FCC adopted a rule that would have required consent to go to one seller at a time. In January 2025, the Eleventh Circuit vacated it in a case called Insurance Marketing Coalition v. FCC. It never took effect. If a vendor or blog says it started in 2026, that's wrong. The court sent it back, so the story may not be over, but today it isn't in force.
Texting now needs registration. If your business texts customers through software, carriers expect the traffic to be registered (called A2P 10DLC). Unregistered traffic can be blocked. This isn't a law. It's the carriers' system, and it affects AI texting tools the same as any other. Our A2P 10DLC guide explains the two steps and what Twilio's documents say about timelines.
What's still pending
The "revoke-all" piece is delayed. One part of the opt-out rule says that when someone opts out of one type of message, it applies to all future robocalls and robotexts from you, including unrelated ones. The FCC's staff extended its waiver of that requirement until January 31, 2027, in an order adopted January 6, 2026. Check the status before you design opt-out logic around that date. The safest habit, either way: treat "stop" as "stop everything" for that customer.
State AI disclosure laws are thin and moving. The two we could confirm in enacted form are Utah and Maine, covered in our recording guide. Utah requires disclosure when a person clearly asks if they're dealing with AI, and offers a safe harbor if the AI says up front that it's AI. Maine bars using AI to mislead a reasonable consumer into thinking they're dealing with a human, unless clearly notified. Other states have bills or laws we haven't verified. If you serve customers in other states, ask counsel to check. We did not confirm any new Florida AI disclosure law, so don't assume one exists, and don't assume one doesn't. Verify at the source before you rely on either.
Court and agency follow-ups. The vacated rule is on remand, and the FCC has a pending rulemaking on the revocation rules. Expect more changes. This is a reason to build your setup so you can change a message or a consent flow quickly, not to wait.
Rules of thumb that hold up either way
These habits are safe under all the rules above. They're practices, not legal advice.
- Get clear consent and keep proof. Save when, where and how each person agreed, and what they agreed to. Our TCPA guide covers the difference between a reply to a customer's own request and a marketing message.
- Start every AI conversation honestly. Say it's an AI assistant. Never let it claim to be a person. If someone asks, it tells the truth.
- Always offer a person. "Reply HELP or say 'representative' to reach someone." Customers still want one. We talk about why in customers still want a person.
- Make opting out easy and instant. Test it yourself before launch, and every month after.
- No outbound AI voice calls without consent. Texts are simpler than calls. Use texts for reminders and confirmations and keep voices for inbound.
- Keep AI out of promises. A person approves prices, discounts and commitments.
- Test with your own phone. Send yourself the message. Reply "stop." See what happens.
Frequently asked questions
Can an AI voice call my customers?
Only with the consent the TCPA requires. The FCC ruled in February 2024 that AI-generated voices are "artificial," so calls using them are held to the rules for artificial or prerecorded voices. For marketing calls that means prior express written consent. Our TCPA guide explains the difference between marketing and non-marketing messages.
Do I have to tell customers they're talking to AI?
We found no federal rule requiring it for inbound calls. Utah requires disclosure when a person clearly asks, with a safe harbor for saying it up front, and Maine bars misleading people into thinking an AI is human. Other states may differ. Honest disclosure is a safe habit. See call recording and AI disclosure laws.
Did the FCC's one-to-one consent rule take effect?
No. A federal appeals court vacated it in January 2025, before it took effect. Any source saying it started in 2026 is wrong. It's on remand, so it could return in some form. Today it isn't in force. Other consent rules, like the opt-out requirements, still apply.
Do I need to register to send business texts?
If you text through software from a regular business number, carriers expect A2P 10DLC registration, and unregistered traffic can be blocked. It's a carrier requirement, not a law, and it doesn't make your messages legal on its own. Read our A2P 10DLC guide for the steps.
How Di-Hy fits
We set up AI texting and answering with the guardrails built in: a clear AI label, a handoff to a person, opt-outs that work, and consent records kept. Our Never Miss a Lead and Answer Customer Questions services cover that, and Keep It Running retests the flows when rules or tools change. We're not lawyers, and we'll tell you when a question belongs with yours. For the wider view see AI compliance for small service businesses. We work from Ybor City across Tampa Bay.
Next step: send yourself your own automated text today, reply "stop," and see whether it really stops. If you're not sure, call us.


