AI for Mortgage Brokers in 2026: Stips, Scenarios and Closing Dates

AI and the automation around it can already chase missing stips, read paystubs and bank statements, answer guideline questions across agency and non-QM programs with citations, and show realtor partners where a file stands. It can't quote a rate your pricing engine didn't produce, decide who gets credit, or sign anything with your NMLS number on it. Use it for the paperwork around the loan, not the loan decision.

A brokered file rarely stalls on the hard part. It stalls on the second bank statement page that never got uploaded, the condition the lender added on Thursday, the realtor texting to ask whether the closing date still holds, and the scenario you need to shop across four wholesale lenders before the borrower goes cold. This guide is written for US brokers and loan officers, where the rules below apply. Brokers in the UK, Canada and Australia work under different regulators, so check your own.

Where a broker's week goes, file by file

The federal government's Occupational Outlook Handbook entry for loan officers lists the core of the job: contacting people who may need loans, explaining loan types and terms, collecting and verifying income and credit information, and making sure loan agreements meet federal and state regulations. It also notes that mortgage loan originators need a license, with prelicensing courses, a national exam, background and credit checks, and annual renewal.

For a broker shop, those duties break into a loop that runs for every file in the pipeline:

• Intake and pre-approval: the application, the credit pull, a pre-approval letter the buyer's agent can attach to an offer.
• Scenario shopping: matching an unusual borrower (self-employed, 1099, DSCR, recent credit event) to a lender whose guidelines and overlays fit.
• Pricing and locks: running the scenario through the pricing engine, locking, and watching the lock expiration against the contract closing date.
• Document collection: the initial needs list, then every condition the underwriter adds after submission.
• Disclosures: the Loan Estimate on time, then the Closing Disclosure before consummation.
• Status updates: the borrower, the buyer's agent, the listing agent and sometimes the title company, all asking the same question in different channels.
• Past clients: rate-drop refinance checks and the referral partners who send the next purchase.

That workload is why adoption is already broad. In AD Mortgage's 2026 survey of more than 250 brokers, reported by HousingWire, 55% said they use AI daily or regularly, yet satisfaction with training averaged 6.49 out of 10 and 57% said they need more support to use their tools fully. AD Mortgage is itself a lender, so read the numbers as one lender's survey of its market.

Mortgage broker tasks AI can take on now

For a broker, the AI that matters sits inside four kinds of software: the loan origination system (LOS) and pricing engine, the borrower portal, guideline research tools, and the lender's own underwriting tech. A general chatbot can help draft a needs-list explanation in plain English, but it should never be the source of a rate or a guideline answer.

Stip chasing and the needs list

Chasing conditions is mostly a schedule, and a good borrower portal already runs it. Floify's email overview describes a Deadline Notice sent to the borrower based on the deadline date you set in their loan flow, which you can leave automatic or switch off. That covers the borrower who forgot the 2024 W-2. It doesn't cover the condition that needs a letter of explanation: AI is useful there for drafting the LOE request in words a first-time buyer understands, which you then check against what the underwriter actually asked for.

Guideline lookups and lender scenario shopping

Scenario questions ("Will anyone do a 12-month bank statement loan on a 640 score with a recent job change?") used to mean calling three account executives. Guideline assistants now answer them first. Zeitro's Strata is pitched to loan officers and brokers for comparing programs across lenders; Zeitro says each answer includes citations so you can read the exact rule, and that the tool is meant for checking program fit before submission, not for approving loans.

Overlays are where these answers go wrong. Agency guidelines are public, but a lender's overlays change with its appetite, so confirm the cited rule against the lender's current matrix or with your AE before you tell a borrower they qualify.

Pricing, locks and the rate-lock calendar

Rates come from your pricing engine, never from a chatbot. ARIVE, an LOS built for independent brokers, combines the LOS, a borrower portal, a product and pricing engine for exact-rate quotes and fee worksheets, and a lender marketplace for submitting loans and getting status back. ARIVE's homepage doesn't sell that pricing engine as AI, and it doesn't need to: it's the right tool for the number.

Where AI helps is the calendar around the lock: a reminder that a 30-day lock expires two days before the contract closing date, or a draft email to the borrower explaining what an extension costs. Both are drafts you check against the lock confirmation.

Income, assets and the underwriting package

Document AI for income is real but mostly sold to lenders, so you'll usually meet it on the wholesale side. Ocrolus' income page describes turning bank statements, pay stubs and tax forms into income data for W-2, 1099, gig, self-employed and rental income, with an underwriter dashboard for review. For a broker, the practical effect is faster condition turnaround from lenders that use tools like it, and a reason to send clean, complete documents the first time.

Status updates for borrowers and realtor partners

The buyer's agent cares about one thing: will the loan close on the contract date. Floify's note on partner features says realtor partners can see how many documents the borrower still owes and which milestones are checked off, without ever seeing the borrower's documents, and can create their own pre-approval letters from the partner portal. That design is the right instinct for AI updates too: tell the agent where the file stands, never what's in the borrower's bank statements.

Past-client and refinance outreach

AI phone outreach has arrived in wholesale. HousingWire's May 2025 report on UWM's tools describes Mia, a virtual loan officer assistant for UWM broker partners that makes calls, sends messages, schedules appointments, sends payment reminders 20 days after a loan closes, and checks in with past borrowers when rates drop. Before you switch on any AI voice outreach, read the consent rules below.

The rules that limit AI in mortgage brokering

US mortgage rules don't mention AI by name, but they decide what an AI tool may do with a borrower's file. These are the ones a broker hits first.

• TRID timing (Regulation Z). The Loan Estimate goes out no later than three business days after the application, and a broker that takes the application may provide it, though the creditor stays responsible (12 CFR 1026.19). The borrower must receive the Closing Disclosure at least three business days before consummation. An AI reminder can protect those dates; it can't extend them.
• ECOA and Regulation B. A person who regularly refers applicants to creditors counts as a creditor for the rules against discrimination and against discouraging applicants (12 CFR 1002.2). If an AI intake chatbot or AI-written ad discourages prospects on a prohibited basis such as race, sex, age or marital status, that's on you. When credit is denied, the notice must give specific principal reasons within 30 days of a completed application, and on applications sent to several creditors it can go through a third party (12 CFR 1002.9). "The model said no" is not a reason.
• Anti-steering. A loan originator may not steer a borrower to a loan because it pays the originator more, unless that loan is in the borrower's interest. The safe harbor rests on showing options that include the lowest rate, the lowest rate without risky features, and the lowest total points and fees (12 CFR 1026.36). If an AI tool ranks lenders for you, know what it ranks on.
• NMLS IDs. The same section requires the loan originator organization's and the individual originator's names and NMLS IDs on the credit application, the Loan Estimate and Closing Disclosure, the note and the security instrument. Template those; don't let a drafting tool retype them.
• Borrower data (GLBA Safeguards Rule). The FTC's Safeguards Rule guide names mortgage brokers as covered. You must choose service providers that can protect customer data, put security expectations in their contracts and monitor them, and report a breach involving 500 or more consumers' unencrypted information to the FTC within 30 days of discovery. Every AI tool that touches a loan file is a service provider.
• AI voice calls (TCPA). In February 2024 the FCC confirmed that AI-generated voices count as "artificial" voices under the Telephone Consumer Protection Act, as Cooley's summary explains, which brings in the prior-express-consent rules for those calls. Check consent before an AI dials a past client.

What AI shouldn't do on a loan file

AI shouldn't make a credit decision, explain one it can't trace, or quote numbers it didn't get from your systems. A few lines to hold:

• No rates from memory. A chatbot's "today's rate" for a 30-year fixed is a guess. Quote from your pricing engine and lock confirmation only.
• No eligibility promises from a summary. A guideline answer without the cited section and the lender's current overlay is a lead, not an approval.
• No raw borrower files in consumer chatbots. Tax returns, bank statements and Social Security numbers stay in tools you've vetted under the Safeguards Rule.
• No auto-sent disclosures. TRID documents come out of your LOS with a person checking them, not from a drafting tool.

Where Merlin fits for mortgage brokers

Merlin handles the part of brokering that lives in Gmail and Google Calendar: the listing agent who emailed on Monday and hasn't heard back, the borrower call where you promised to confirm the appraisal date, the closing on Friday you need to prepare for. Merlin runs on macOS, keeps a record of each borrower, agent and AE you deal with, and drafts the emails you request.

Merlin is not an LOS, a pricing engine or a borrower portal. It doesn't collect documents, run AUS, produce disclosures, price loans or answer the phone, and it doesn't place calls or send texts. Keep those in ARIVE, Floify or whatever stack you already run.

• Unanswered agents and borrowers. The People panel in Merlin has a Needs you list: a buyer's agent or borrower lands there when an email they wrote to you (not a newsletter or automated notice) has sat unanswered for 2 to 30 days. Only Gmail counts, so a realtor's text message won't show up.
• Closing-date updates you sign off on. Type "draft an update to the buyer's agent: we're clear to close, closing stays Friday" and the email appears as an editable card, sent only after you approve it. Match the dates to your LOS first.
• Condition emails sorted for you. Set up a routine that fires when mail arrives from a lender's domain, or with "conditions" in the subject line, and posts a summary of the new conditions in chat. Once a routine is turned on it runs without you, so don't give it permission to send mail.
• Ready for the closing table or the realtor lunch. Type "prep me" before a meeting, or wait for the Meeting Brief that arrives roughly ten minutes before a calendar event, gathers earlier threads and outstanding promises with everyone on the invite.
• Recorded borrower calls. Borrower calls you take on the Mac get recorded by Merlin when you choose, with speaker-labeled transcripts on paid plans. Tell the borrower first and follow your state's recording consent rules.

If borrower documents arrive by email, connecting that Gmail account gives Merlin access to them, so treat it as a service provider under your Safeguards Rule program before you connect it.

Five first moves for a broker shop this week

Begin with whichever task causes the most rework per file, use the automation already inside your LOS and portal, and keep borrower data inside vetted tools.

1. Turn on portal reminders. Set deadline reminders for outstanding document requests on your next three files, and check what borrowers receive.
2. Test a guideline assistant on closed files. Ask it three scenarios you already know the answer to, and check every citation against the lender's matrix.
3. Add your realtor partners as partners, not CCs. Give agents milestone updates through the portal so borrower documents never sit in their inbox.
4. Write down your AI vendors. List each tool that touches a loan file, its data terms and contract, and add it to your Safeguards Rule program.
5. Check consent before any AI calling. Confirm prior express consent for every past client an AI voice tool would call.

For splitting lender, borrower and partner mail before any AI touches it, see how to use Gmail labels, and for recording borrower calls compare the options in this roundup of AI note takers.