I got 12 texts and an AI phone call from a brokerage I never contacted. Here's what it would cost them.

On July 2 a Massachusetts brokerage started texting me. I had never filled out a form on their site, never asked about a listing, never given anyone there my number for any purpose. Over the next two months I got at least 12 texts from the same 774 number and several phone calls, and the voice on the other end of those calls was not a person.

I know how their system saw me, because I eventually got a look at the CRM record. My source was tagged “referral.” Nobody referred me. The tag was the software’s way of filling a required field with something that sounded like consent.

This post is not about that brokerage. It is about the “digital assistant” product they bought, because there are hundreds of brokerages running the same category of tool, and most of them do not know what it is doing in their name.

What actually happened

The sequence, compressed:

Twelve texts between July 2 and September 4. Same number, same voice, a rotating set of “just checking in” and “are you still thinking about a move” openers.
Two of those texts arrived before 8 AM my time. The system was sending on Eastern time and I am on Pacific.
Multiple voice calls. When I asked the caller directly if it was a person, it said it was an automated assistant.
When I asked how they got my number, the assistant told me I had inquired about a property. I hadn’t. It invented a contact history on the spot to keep the conversation going.

Each of those four lines maps to a different rule.

The rules, one at a time

I am not a lawyer and this is not legal advice. What follows is what the statute and the FCC’s rules say, with the citations, so you can hand them to someone who is.

Artificial or prerecorded voice calls. 47 U.S.C. § 227(b)(1)(A)(iii) prohibits calls to a cell phone using an artificial or prerecorded voice without the prior express consent of the called party. In February 2024 the FCC issued a declaratory ruling confirming that AI-generated voices count as “artificial” under that section. So an AI voice agent dialing someone who never consented is the plainest violation on the list, and it does not depend on any of the autodialer arguments below.

Time of day. 47 CFR § 64.1200(c)(1) prohibits telephone solicitations before 8 AM or after 9 PM in the called party’s local time. Not the sender’s time. The called party’s. Two of the texts I received violate this regardless of anything else, because a solicitation sent at 7:15 AM Pacific is a violation even if I had signed a consent form. The fix is trivial (look at the area code) and the product did not do it.

Do Not Call. 47 CFR § 64.1200(c)(2) bars telephone solicitations to numbers on the National Do Not Call Registry absent an established business relationship or written consent. Separately, 64.1200(d) requires the caller to maintain an internal do-not-call list and honor requests to stop. If a system keeps texting after “stop,” or texts registered numbers it never scrubbed, it fails both.

Automated texts. Texts are “calls” for TCPA purposes; that has been settled since the FCC’s 2003 order and courts have followed it. Whether a given texting platform is an “automatic telephone dialing system” is the unsettled part. The Supreme Court narrowed the definition in Facebook v. Duguid (2021) to systems that store or produce numbers using a random or sequential number generator, and most CRM-driven texting does not do that. So a plaintiff’s strongest claims on the texts themselves usually run through the DNC and time-of-day rules rather than the autodialer prong. Some state laws define autodialers more broadly than the federal statute now does, and a few states have their own texting statutes with separate damages. Talk to an attorney about your states.

The fabricated referral tag. This one is not a rule violation on its own. It is evidence. The TCPA allows treble damages for violations that are “willful or knowing,” and a consent field populated with a source that did not exist is a bad fact for a defendant trying to argue good faith. The same goes for an assistant that tells a consumer they inquired when they did not.

The math

Statutory damages under 47 U.S.C. § 227(b)(3) and § 227(c)(5) are $500 per violation, and up to $1,500 per violation if the court finds it willful or knowing. Consumers can bring these claims themselves, in small claims court if they want.

Take my case and count conservatively. Twelve texts and the calls, which I did not log carefully. Counting only the texts: 12 contacts.

At $500 each: $6,000.
If a court agrees the conduct was willful: $18,000.

That is one consumer. The assistant was not built for me specifically. Whatever list it was working, I was one row. If the brokerage uploaded 2,000 dormant contacts and the tool ran the same sequence on all of them, the exposure is not $6,000. It is $6,000 times however many of those people never consented, and a class action attorney only needs to find a few of them.

For scale: the TCPA settlements you have read about in real estate over the last few years (the large brokerage portal cases, the “do you still want to sell” text campaigns) have run into the tens of millions. Those did not start with a sophisticated scheme. They started with a tool that texted people who had not asked to be texted.

Why this is worth writing about if you sell an automated texting product

Klosed sends texts on behalf of real estate agents. Automated ones. So the honest version of this post has to answer the obvious question: what stops us from being the 774 number?

The difference is not that we are careful and they were not. The difference is a set of specific design decisions, each of which corresponds to one of the failures above.

We only text people who already talked to the agent. Klosed reactivates an agent’s own dormant leads: people who inquired, toured, or worked with that agent and then went quiet. The source of every contact is the agent’s CRM, and the source field has to say where the person actually came from. There is no “referral” default.

Send windows are set by the recipient’s area code, not the office clock. A 661 number gets Pacific hours. A 774 number gets Eastern hours. This is a lookup table. It is not hard, and any product that skips it is telling you how much thought went into the rest.

The assistant says it is an assistant. On the first direct ask, the reply is yes. It does not claim to be the agent, and it does not claim the person inquired about something they did not. If the system does not know why someone is in the database, it says so and offers to have the agent follow up.

Every list gets scrubbed before the first message. National DNC, reassigned numbers, litigator lists, prior opt-outs. The numbers that fail the scrub do not get billed and do not get texted.

Stop means stop, and it means stop across the whole account. One opt-out suppresses the number everywhere, not just in one campaign.

None of this makes automated outreach risk-free, and none of it substitutes for a conversation with a TCPA attorney about your own consent records. It does mean that when a consumer asks “how did you get my number,” the answer is a real one.

If you already run one of these tools

Three questions to ask your vendor this week:

Where does the consent source for each contact come from, and can it ever be populated by the system rather than by a human or a form?
What time zone governs the send window, and is it the recipient’s?
What does the assistant say when someone asks whether it is a person, and when someone asks how you got their number?

If the answers are “we set it to referral by default,” “the account’s time zone,” and “it says it’s you,” you are running the product that texted me. I am not going to sue anyone over 12 texts. The next person might.

This post describes the author’s own experience. Statutory and regulatory citations are provided for reference. Consult an attorney licensed in your state about how the TCPA and state telemarketing laws apply to your business.