An AI voice calling a cell phone is a prerecorded call under federal law, and it carries statutory damages of $500 to $1,500 per call. The FCC decided this in a declaratory ruling issued in February 2024, numbered FCC 24-17, and it applies whether the voice reads a script or holds a genuine conversation. The ai cold calling laws for real estate operators run into are not new laws. They are the TCPA, applied to a technology that most vendors sell without mentioning it.
That gap matters because the people buying AI dialers are the people who get sued, not the people who built them.
The FCC did not write a new rule for AI. It said the old rule already covers it. A voice generated by software is an "artificial or prerecorded voice," which has been restricted since 1991.
What the ai cold calling laws for real estate actually say
On February 8, 2024, the Federal Communications Commission issued a declaratory ruling confirming that calls using AI-generated voices fall inside the Telephone Consumer Protection Act's restrictions on artificial and prerecorded voices.
Three parts of that are worth separating.
It covers conversational AI, not only recordings. A voice agent that listens, responds, and adapts is still an artificial voice. The fact that it sounds like a person and reacts like a person is the reason the rule reaches it, not a reason it escapes.
It covers cloned voices. An agent running on a clone of your acquisitions rep is an artificial voice under this ruling. The clone being of a real person who consented to being cloned changes nothing about the call.
It was a clarification, not a new restriction. There was no grace period and no phase-in, because in the Commission's framing nothing changed. The obligation existed the day before.
This is a flag for review, not legal advice. State rules stack on top of the federal ones and the penalties here are per call, which means a list of 4,000 numbers is not a small exposure. Run your setup past a telecom attorney before it dials anything.
Does this apply to how you are using AI?
Most investors use AI in ways the TCPA never touches. The test is who sends the message.
Outside the rule
AI writes a follow-up email, you read it and hit send
AI drafts a text, you send it from your own phone
AI summarizes a call that already happened
AI scores a lead so a human knows who to call first
AI researches a property before your rep calls
Inside the rule
AI places the call on a schedule you set
AI sends the text itself
AI dials a list and speaks
AI leaves a voicemail in a generated voice
A voice clone reads your script to a cell phone
The left column is a writing assistant. You are the sender, a human made the decision to contact that person, and no automated system placed the call. The right column is an automated system contacting someone on your behalf, which is the thing the statute regulates.
The line is not about how smart the tool is. It is about whether a person decided to make that specific contact.
What consent do you need?
Two levels, and which one applies depends on what the call is for.
- 1
Informational calls need prior express consent
The person gave you their number for this purpose. A seller who filled in your form asking for an offer has given it for a call about that offer.
- 2
Marketing calls need prior express written consent
A signed or digitally recorded agreement, specific to your company, that says they agree to receive automated or prerecorded marketing calls at that number.
The practical difference shows up on cold lists. A skip-traced number from a pre-foreclosure list has given you neither. Nobody on a purchased list has consented to anything, which means an AI voice calling that list is the exact scenario the statute was written for.
Consent does not transfer
This is the part that catches people who buy data. Consent given to one company does not pass to another. A list vendor cannot consent on your behalf, and a disclosure on their site saying partners may contact you is not consent to your company specifically.
If you did not collect it, you do not have it.
Consent has to be revocable
A consumer can revoke consent by any reasonable means. Not by a method you specify, not through a portal you built, by any reasonable means. Saying "stop calling me" on the call counts. A reply of STOP counts. An email counts.
Your system has to catch all of those, which is a harder engineering problem than it sounds when the thing receiving the revocation is a voice agent rather than a person.
A spoken opt-out on the call is honored immediately, on that call
STOP, CANCEL, END, QUIT and UNSUBSCRIBE are all handled on texts
An emailed revocation reaches the same suppression list
Suppression is checked before every dial, not nightly
The list is shared across every tool that can contact somebody
That last item is where most setups break. The CRM knows about the opt-out, the dialer does not, and the dialer is what calls them again on Thursday.
What does it cost to get wrong?
Statutory damages run $500 per violation, trebled to $1,500 for a willful or knowing violation. There is no aggregate cap in the statute.
Do that multiplication once and the economics of "we will sort the compliance out later" stop working. A single week of a misconfigured dialer on a cold list is not a fine, it is a business-ending number, and TCPA litigation is an established practice area with plaintiffs' firms that monitor for exactly this.
The reason this matters more in 2026 than in 2022 is volume. A human rep makes 80 dials a day. An AI agent makes 8,000. The same mistake costs a hundred times more.
What about texting?
Texting is covered by the same statute, and it has an extra layer the voice side does not: carrier registration.
Application-to-person messaging in the United States requires A2P 10DLC brand and campaign registration before carriers will deliver reliably. Unregistered traffic gets filtered, which shows up as "our texts stopped working" rather than as an error anyone can read.
Registration is not consent. It is permission from the carriers to send at all. You need both.
Quiet hours apply too, and they run on the recipient's clock, not yours. Nothing before 8am or after 9pm where they are. An automated system sending from Austin to a seller in California at 8:15am Central is sending at 6:15am Pacific, and the system is the one that has to know the difference.
What is still unsettled?
In August 2024 the FCC proposed additional AI-specific rules, including a separate consent standard for AI calls and a requirement to disclose in real time that the caller is AI. As of late 2026 those remain proposals.
That is a genuinely awkward position to plan around. The disclosure requirement is not federal law yet, several states already require some form of bot disclosure, and the proposal signals where the federal rule is heading.
The defensible answer is to disclose anyway. An agent that opens by saying it is an AI assistant calling on behalf of your company satisfies the state rules that exist today, satisfies the federal rule if it lands, and costs you the small number of people who were going to hang up on an AI agent in the first thirty seconds regardless.
Disclosure is also the honest answer. An AI agent that lets a distressed homeowner believe they are talking to a person has a consent problem and a trust problem, and the trust problem is the one that loses the deal.
What a compliant setup looks like
Working through it in the order that matters.
- 1
Decide what the agent is for
Inbound response to people who contacted you is a different consent question from outbound dialing on a cold list. Most of the value is inbound anyway.
- 2
Fix the form language first
The consent line on your lead form decides what you may do with every lead it produces. Store the exact wording, not a version number.
- 3
Register for A2P 10DLC before any texting
Weeks, not days. Start it before you need it.
- 4
Build one suppression list
Every tool that can contact a person reads from it, before every send.
- 5
Disclose that it is AI
Opening line, every call.
- 6
Keep the recording and the consent record together
When a claim arrives two years later, the record is the defense.
- 7
Have a lawyer read the whole thing
Per call damages make this the cheapest part of the project.
The strongest position is the boring one. Point AI at the leads who already raised their hand, where consent is clean and intent is high, and leave the cold outbound to channels that do not carry per-call statutory damages. Direct mail has no TCPA exposure. Paid search brings people who are already looking. Both feed an inbound pipeline the AI can work without this entire problem.
Speed to lead on an inbound lead is worth more than volume on a cold one anyway. The seller who filled in your form at 11pm and got a response at 11:01pm is the deal. The 4,000 people who did not ask to hear from you are a lawsuit with a low hit rate attached.
Frequently asked questions
Is AI cold calling illegal?
No. It is restricted, not banned. An AI voice calling a number where you hold the right level of consent is lawful. The same call to a number where you hold none is a violation, and the ai cold calling laws for real estate investors operate under treat a generated voice exactly like a prerecorded message.
Does the TCPA apply if the AI only sends texts?
Yes. The statute covers automated texts as well as calls. Texting adds A2P 10DLC carrier registration on top of consent, plus quiet hours in the recipient's timezone rather than yours.
Can I use AI to call numbers on the Do Not Call registry?
Not for marketing, without an established business relationship or express written consent. The registry applies to telemarketing whether a human or an AI places the call. Scrub against the national registry and your own internal list before every campaign.
Does an AI agent have to say that it is AI?
Federally, not yet. The FCC proposed a real-time disclosure requirement in August 2024 and it has not been finalized. Several states have their own bot disclosure laws, and disclosing is the only position that works under every version of the rule.
What if the AI drafts the message and I send it?
Then you are the sender and the automated-system rules are not in play, because a person decided to contact that person. This is the cleanest way to get most of the value from AI without taking on per-call exposure.
Does consent collected by a lead vendor cover my calls?
No. Consent is specific to the company that collected it. A vendor cannot grant it on your behalf, and a partner disclosure buried on their site does not name you.
How many calls does it take to become a real problem?
At $500 per call, a hundred wrong numbers is $50,000. An AI agent can place that many in an afternoon, which is the reason this matters more now than it did when a human rep made eighty dials a day.
Where this leaves you
The rule is not an argument against using AI. It is an argument for pointing it at the half of the funnel where consent is clean.
An AI agent answering inbound leads in under a minute, every time, at 2am, is the version of this that makes money and carries almost no regulatory risk. The same agent dialing a cold list is where investors get into trouble, and the vendors selling that version rarely mention the per-call math.
If you want a second pair of eyes on what you are running now, book a free strategy call. We go through the stack, the consent language on your forms, and which part to fix first. It is a working session, not a pitch.
*Sources: FCC Declaratory Ruling 24-17, February 8, 2024. FCC Notice of Proposed Rulemaking on AI-generated voice calls, August 2024. Telephone Consumer Protection Act, 47 U.S.C. 227. Last reviewed October 2026.*
