Mortgage AI Compliance: What Loan Officers Need First

Mortgage is the industry where an AI appointment setter pays for itself fastest and where it can create the most trouble. Rate-table leads go to five lenders at once, so response speed decides who gets the application. At the same time, almost everything a borrower wants to ask about is regulated: rates, payments, down payments, approval odds, all of it.

Mortgage AI compliance comes down to where the setter stops talking. It can qualify, explain process and book. Once it states a rate, payment, down payment or term, Regulation Z advertising rules attach and the conversation is in licensed loan originator territory. Build it to hand off before that line.

This is not legal advice. Mortgage rules vary by state and by product, and your compliance officer and counsel decide what your agent says. What follows is the map of which rules apply where, so you can have a useful conversation with them.

Where the line sits

1. Regulation Z triggering terms

This is the rule most people have not thought about, because they think of Reg Z as applying to ads rather than to conversations. It applies to advertisements for credit, and a scripted message sent to a list of prospects is an advertisement.

Under 12 CFR 1026.24(d), four things are triggering terms in a closed-end credit advertisement:

• the amount or percentage of any down payment
• the number of payments or period of repayment
• the amount of any payment
• the amount of any finance charge

State any one of those and the advertisement must also state the down payment amount or percentage, the full repayment terms including any balloon payment, and the annual percentage rate using that term, noting if the rate may increase. The CFPB's commentary makes clear that implied terms count too: "80 percent financing available" implies a 20% down payment and triggers the rule just as a stated figure would.

For dwelling-secured credit, 1026.24(f) adds more. If an ad states any payment amount, it has to disclose each payment over the loan term and its time period, and for first-lien loans, state that payments exclude taxes and insurance and that the real obligation will be greater.

What this means for the agent: an SMS saying "rates from 5.99%" or "payments around $1,800" is an advertisement with a triggering term in it, and the required disclosures do not fit comfortably into a text message. The practical design is that the agent never states a number. It says a loan officer will go through the numbers, and it books that call.

Also in scope: the Mortgage Acts and Practices rule, Regulation N, 12 CFR Part 1014, which prohibits material misrepresentations in commercial communications about mortgage credit, including about rates, payments, the amount of cash to be received, and whether a consumer has been approved. Note the breadth: "commercial communication" covers a text message, not just a billboard.

2. TCPA consent, and what AI changed

On February 8, 2024, the FCC issued a Declaratory Ruling (FCC 24-17) confirming that AI-generated voices are "artificial" voices under the Telephone Consumer Protection Act. Calls using them need the same prior express consent as other artificial or prerecorded voice calls, and for telemarketing that means prior express written consent.

For a loan officer, the practical questions are about where your leads come from:

• Your own forms. Straightforward, if the disclosure is there. The language has to say the person agrees to receive calls and texts, including from automated systems and AI, from your named company at the number given, and that consent is not a condition of getting a loan.
• Rate-table and aggregator leads. This is where it goes wrong. Consent collected by an aggregator "and its partners" is not reliably consent for you. The FTC takes the position that a written agreement for prerecorded marketing calls identifies a specific seller, per its Telemarketing Sales Rule guidance and 16 CFR Part 310. Ask every vendor for the exact consent text the consumer saw and which companies it named, and keep that record.
• Trigger leads. Credit-bureau trigger leads are a separate conversation with your counsel entirely, and AI voice on them is not a shortcut around it.
• Purchased lists. We do not run AI outreach to purchased lists, and neither should you.

Then there are the mechanics: honor opt-outs across both channels immediately, scrub against the Do Not Call registry and your internal list, and set calling hours to the strictest rule that applies, since several states have mini-TCPA laws tighter than the federal ones. The FCC's revocation rules have been amended and delayed more than once, so have your compliance officer confirm the current requirement rather than relying on anything written a year ago.

3. Licensing: who is allowed to say what

Under the SAFE Act, taking a residential mortgage loan application or offering or negotiating terms requires a licensed or registered mortgage loan originator. The licence belongs to a person, and your AI is not a person.

That sounds obvious, and the problem is that the boundary is narrower than it looks. "Offering or negotiating terms" covers more than quoting a rate. An agent that tells a borrower what they would qualify for, or suggests a product is better for them, is doing something close to that work.

The safe construction keeps the agent on factual intake:

• It collects what the lead volunteers: purchase or refinance, property type, timeline, the credit range they report, best time to talk.
• It never evaluates. No "that should be fine," no "you'd qualify," no "that program is better for you."
• It books the licensed MLO and says that is who will go through the options.

Your NMLS ID and company ID belong in the material the agent sends, matching whatever your state and your compliance team require for advertising. The NMLS site is the authority on registration itself, and state-by-state advertising rules sit with your regulator.

One more licensing wrinkle worth naming: if your agent contacts borrowers in states where you are not licensed, you have a problem that has nothing to do with AI. Set geographic rules on which leads the agent is allowed to touch.

4. Disclosure that it is AI

Several states now require disclosure when a consumer is interacting with AI, and more are adding it. Whatever your state says today, disclosing is the right default in mortgage, because this is a transaction built on trust and finding out later that the "person" was software is a bad surprise.

Ours open with something like: "Hi Marcus, this is Ava, the AI assistant for Dana at Lakeside Lending. You asked about a refinance. Can I ask two quick questions and get you on Dana's calendar?"

If a lead asks whether they are talking to a person, the agent says it is an AI assistant and offers to connect them. It never claims to be human.

5. Call recording consent

Federal law permits recording a call with one party's consent (18 U.S.C. 2511). A number of states require all parties to consent, and those laws follow the number you are calling, not where your office is.

Since mortgage leads routinely cross state lines, the simplest workable rule is to announce recording on every call. It costs you three seconds of script and removes the question. Your retention period and who can access recordings are decisions for your compliance team.

6. What the agent must never do

Write these into the knowledge base as hard rules, not as guidance.

• State a rate, APR, payment, down payment or term. Triggering terms, as above.
• Say or imply approval. Not "you're approved," not "you're pre-approved," not "that should be no problem."
• Promise savings. "You could save $400 a month" is a misrepresentation risk under Regulation N and it is the single most common thing a sloppy mortgage script says.
• Advise on product choice. FHA versus conventional is a licensed conversation.
• Ask for a Social Security number, full date of birth or account numbers over text.
• Discuss someone else's loan with whoever answered the phone.
• Guess. When a question falls outside the knowledge base, it hands off. Agents we build on StrideOps offer a person rather than improvising, which is the behavior a regulated product needs.

A compliant opening sequence

What a compliant, useful sequence actually looks like for an opted-in rate-table lead:

Text, within seconds: Hi Marcus, this is Ava, the AI assistant for Dana at Lakeside Lending (NMLS #123456). You asked about refinancing on our site. Dana can walk you through your options. Are you free today at 2:15 or tomorrow at 10:00? Reply STOP to opt out.

Lead: what rate can you do

Agent: Dana handles rates, and they depend on your situation, so I do not want to guess at a number. She can go through it on the call. Does 2:15 today work?

Lead: is this a real person

Agent: I am an AI assistant for Dana's team. Happy to get her on the phone with you instead, if you would prefer that now.

The rate question is the whole test. Notice what the agent does: acknowledges it, declines to answer, gives an honest reason, and returns to the booking. It does not dodge, and it does not invent a range.

More patterns are in our AI appointment setter script guide, and the follow-up cadence for mortgage specifically is in mortgage lead follow-up.

A pre-launch checklist

• Compliance officer and counsel have reviewed every script and the knowledge base
• No triggering terms anywhere in any message the agent can send
• Consent language reviewed on every lead form you own
• Written consent text and named sellers obtained from every lead vendor
• Consent records stored with timestamp, form version and source
• Opt-out keywords tested across both SMS and voice, synced to the CRM
• Calling hours set to the strictest applicable state rule
• Geographic rules set so the agent only contacts states you are licensed in
• NMLS ID included where your regulator requires it
• AI disclosure in the opening message
• Call recording notice configured
• Hand-off to a licensed MLO tested, including after hours
• Every transcript readable and exportable by your team
• Simulated calls run, including a lead who pushes hard for a rate

FAQ

Can loan officers use AI to call leads?

Yes, with prior express written consent for telemarketing calls, AI disclosure, proper opt-out handling and an agent that never states rates, payments or terms. The FCC confirmed in FCC 24-17 that AI voices count as artificial voices under the TCPA.

Can an AI quote mortgage rates?

It should not. Stating a rate, payment amount, down payment or repayment period is a triggering term under Regulation Z 1026.24(d), which pulls in a set of disclosures that do not fit in a text, and quoting terms moves into licensed loan originator activity.

Do I need to tell borrowers they are talking to AI?

Several states require it, and it is the right default everywhere in mortgage. A good agent discloses in its opening message and confirms whenever asked.

Does an AI setter need an NMLS license?

The AI is not a licensed person and cannot take an application or offer or negotiate terms. Keep it on factual intake and booking, and make sure a licensed MLO handles everything else. Follow your regulator's rules on including your NMLS ID in advertising.

What about leads bought from a rate table?

Get the exact consent language the consumer saw and the list of sellers it named, in writing, before any AI call or text. Consent given to an aggregator "and its partners" is not reliably consent for you.

Build it with the guardrails from day one

We build AI follow-up for mortgage teams with consent, disclosure, triggering-term rules and hand-off to a licensed MLO configured before launch. Your compliance team reviews every script. Book a Stride Scaling Call, or see how the full setup works for mortgage loan officers.