AI phone assistants for brokerages: how consent and opt-outs should work

A brokerage that adds an AI voice assistant is not just buying software. It is adding a system that places and answers calls on the brokerage's behalf, and calling is one of the most heavily regulated things a business does.

This is not a reason to avoid the category. It is a reason to ask a specific set of questions before you switch one on, and to know which answers should end the conversation. Most of them are questions a vendor with a well-built product will answer immediately and a vendor without one will talk around.

This is an operations post, not legal advice. Telephone consumer rules are federal, state and sometimes local, they change, and your brokerage's obligations depend on facts about your business. Have your counsel review your calling practices before you scale anything.

First, the distinction that governs everything

Inbound and outbound are different products wearing the same interface.

When someone calls you, they initiated contact. An assistant answering that call is doing something closer to what a receptionist does. The obligations are real but light: be honest about what the caller is talking to, handle the information properly, and get them to a human when they want one.

When the system calls or contacts someone who did not initiate it, an entirely different regime applies — consent requirements, do-not-call obligations, time-of-day restrictions, identification requirements, and rules that are specific to calls placed using automated technology and artificial or prerecorded voices. The exposure here is substantial and it is per call.

So the first question for any vendor is: which of these am I buying, and can I have one without the other? A brokerage that wants its phone answered should be able to switch that on without also acquiring an outbound dialer. If the product does not separate them, you are taking on the harder compliance posture in order to solve the easier problem.

The questions to ask, and what a good answer looks like

"What does the assistant say it is?"

The assistant should identify itself as an automated assistant, clearly, near the start of the interaction, without the caller having to ask. Disclosure requirements for AI in calls vary by state and have been moving; several states have added specific rules in recent years.

Beyond the legal floor, there is a practical argument: a caller who works out mid-conversation that they have been talking to software feels misled, and they attribute that to your brokerage, not to your vendor.

A bad answer: "It sounds so natural that people can't tell." That is being sold as a feature. It is a liability and a reputation problem.

"Where does consent live, and can I see it?"

For anything outbound, you need a record of the consent you are relying on: who gave it, when, through what, and what they actually agreed to. Not a checkbox in a settings page — a per-contact record you can produce.

A good answer shows you a contact and the consent attached to it. A bad answer describes consent as something you are responsible for tracking elsewhere, in a product that then places calls with no reference to it.

"How does someone opt out, and what happens in the next sixty seconds?"

This is the highest-value question in the list, and the one that most reliably separates products.

A person should be able to opt out in the channel they are already in — saying "stop calling me" on a call must work, without them having to email anyone or visit a website. When they do:

• The opt-out is recorded immediately, against that person.
• It takes effect before the next call, not after a nightly sync.
• It is honoured across the whole brokerage, not just for the agent whose campaign made the call. A consumer who tells your office to stop expects your office to stop.
• It survives an import. Re-uploading a list must not resurrect people who opted out.

That last one is worth pressing hard on, because it is the most common way an opt-out is silently undone, and a consumer who opted out and then gets called again is the profile of a complaint.

A bad answer: any version of "the agent marks them as do-not-call in the CRM." Manual, and it will be forgotten on exactly the call where it matters.

"Do you check the national and state registries, and my own list?"

There are three layers: the national registry, state registries where applicable, and your brokerage's own internal list. All three have to be checked, and the internal one is the most often missed even though it is the one you control completely.

Ask specifically: is the list checked at the moment of the call, or at the moment the campaign was built? A list checked at build time and dialed three days later is a list that is three days stale.

"What happens to the recordings and the transcripts?"

Call recording consent rules vary by state, and several require all parties to consent. If the product records, you need to know what it does about that, where recordings are stored, how long they are kept, who can access them, and what happens when a consumer asks for their data to be deleted.

A brokerage handling consumer data has obligations under privacy law independent of anything to do with calling.

"How does a caller reach a human?"

Every assistant needs a clean escalation path, it needs to work on the first ask, and it needs to work when the caller is frustrated rather than only when they say a magic phrase. Test this yourself before you buy, and test it rudely.

"Can you prove any of this later?"

Complaints arrive months after the fact. You will need to reconstruct one specific call: what was said, what consent was relied on, whether the person had opted out. If the product cannot produce that, the fact that it behaved correctly will be very hard to demonstrate.

What a brokerage should do regardless of vendor

Four things, none of which depend on which product you buy.

Write down your calling policy. One page. Who may be called, on what basis, at what hours, and what happens when someone says stop. Most brokerages have never written this down, which means every agent has their own version.

Make the opt-out path a single path. Whatever route a consumer uses — a call, a reply, an email to the office — it lands in one list that everything checks. Two lists means one of them is wrong.

Sample your own calls. Listen to a handful every month. This is how you find out that the disclosure stopped playing, or that the assistant started saying something you would not have approved. Nothing will alert you; the calls just get worse.

Train agents that the tool does not transfer responsibility. If a call goes out under your brokerage's name, the brokerage answers for it. An agent who uploads a purchased list into a well-built system has created a problem the system cannot detect, because everything about that list looks legitimate to the software.

The short version

Ask a vendor those seven questions. A serious product answers all of them in a single conversation, and usually volunteers a couple of them before you ask.

If consent and opt-outs come up as an afterthought — or as your problem rather than a feature of the system — that tells you what the product was designed to optimise for, and it is worth knowing before your brokerage's name is on the caller ID.