Can AI pre-qualify a borrower?

AI can draft a pre-qual based on stated income, credit, and asset estimates, but the loan originator must review, verify, and sign. ECOA / Reg B requires that adverse action be based on specific, verifiable reasons — 'the model said no' is not a compliant denial. Every AI output that could influence an adverse-action decision must trace back to documented, compliant criteria and be reviewed by a licensed NMLS-registered LO.

Does AI help with HMDA and fair-lending monitoring?

Yes. AI can build disparate-impact / disparate-treatment dashboards on the HMDA LAR, flag decline-rate gaps by race / ethnicity / sex / age, and surface pricing outliers. But the final fair-lending analysis, statistical model selection, and any remediation are owned by compliance. CFPB 2024-2025 guidance is explicit: algorithmic underwriting is not an excuse — lenders are responsible for model bias.

What about the LO Compensation Rule and AI?

Reg Z §1026.36(d) prohibits compensating LOs based on loan terms (other than amount). AI-generated 'lead scoring' or 'rate suggestions' that steer LOs toward higher-margin products for higher comp are a steering and LO-comp violation. Make sure your AI recommendations are consistent across the book and document why.

Can I use ChatGPT to draft a borrower email about their rate?

Drafting is fine; content is regulated. Any statement about rate, APR, payment, or terms must follow Reg Z §1026.24 advertising rules and MAP Rule (Regulation N) for mortgage ads. If you quote a rate, APR must accompany it. 'Free', 'guaranteed', 'government-endorsed' and similar phrases are restricted. And never paste NPI (SSN, income, bank data) into a consumer LLM — GLBA violation.

Is AI-scored underwriting allowed in mortgage?

Fannie Mae DU and Freddie Mac LPA already use ML; both investors have model risk governance. For broker/originator use, follow your wholesale investor's overlay, keep a human underwriter in the loop, and document that the AI output supported — not replaced — the credit decision. For state-regulated non-agency, state UDAAP and fair-lending rules apply.

How to Use AI for a Mortgage Brokerage (2026 Playbook)

May 4, 2026 · 13 min read

TL;DR

Mortgage brokers in 2026 win on two metrics: hours-to-clear-to-close and cost-per-funded-loan. AI collapses both by triaging leads, extracting income / asset docs, running pre-flight AUS scenarios, drafting conditions requests, generating investor-compliant marketing, and watching for HMDA / fair-lending drift. Live inside Encompass AI Copilot, LendingPad, Arive, BNTouch, Sales Boomerang — plus ChatGPT Enterprise / Claude for Work (GLBA-safe, no-training). Hard rules: SAFE Act licensing, TILA-RESPA TRID tolerances, Reg B/ECOA, RESPA §8 kickback prohibition, LO Comp Rule (Reg Z §1026.36(d)), and every AI output is a DRAFT — a licensed LO / UW signs.

Why brokers need an AI playbook now

The 2024-2026 rate cycle squeezed originators from both sides: volume down, competition up, cost-per-loan at multi-year highs (MBA Quarterly Performance Report). Wholesale channels added efficiency mandates. AI is the only way a 3-LO shop competes with a retail megalender on response time, conditions-clearing speed, and borrower experience.

Mortgage is also one of the most-regulated consumer-finance verticals. Every prompt below has a compliance frame. Where the AI touches NPI (SSN, income, credit, bank), it must be behind GLBA / state privacy law controls — consumer LLMs are out.

The compliance floor (read this first)

SAFE Act + NMLS: state-licensed MLOs must be NMLS-registered. AI cannot originate; LOs originate.
TILA-RESPA (TRID): LE within 3 business days of application, CD 3 business days before consummation, fee tolerances, changed-circumstance documentation.
Reg B / ECOA: prohibited basis factors, adverse action notice timing (30 days), notice content (specific principal reasons), consistent criteria.
Reg Z / Loan Originator Comp (§1026.36(d)): no comp based on terms; consistent steering rules; qualified mortgage; anti-steering safe harbors.
RESPA §8: no kickbacks, no fee-splitting for referrals; marketing services agreements tightly scoped.
HMDA: LAR fields; CFPB public dataset exposure; fair-lending analytics.
MAP Rule (Regulation N): mortgage advertising substantiation and prohibited practices.
GLBA + state (CA CCPA/CPRA, NY SHIELD, WA My Health My Data for health-income borrowers): NPI handling, breach notification, vendor DPAs.
Model Risk (SR 11-7 / OCC 2011-12): for depositories and larger nonbanks — model inventory, validation, monitoring.

The 2026 mortgage-broker AI stack

POS / LOS with AI: ICE Encompass Copilot, LendingPad, Arive, LendingQB, Blend, Maxwell AI.
Borrower intake: SimpleNexus (nCino), Blend, Floify AI with doc-OCR.
Income / asset verification: AccountChek, FormFree, Truework — with AI exception handling on variable income, self-employment, K-1s, bonuses.
Pricing / PPE: Optimal Blue, Polly AI, LoanNEX, with AI-driven lock recommendation models (review with compliance).
CRM / nurture: BNTouch, Surefire, Jungo, Sales Boomerang / TrustEngine for trigger alerts.
Compliance: ComplianceEase / LoanLogics / Ncontracts Compliance AI for TRID / fee / state-level checks.
Horizontal AI (internal, NOT NPI): ChatGPT Enterprise, Claude for Work, Microsoft Copilot with DPA + no-training.

10 copy-paste prompts for a mortgage broker

Run inside GLBA-compliant, DPA-covered tools. Never paste SSNs, full pay-stubs, or bank statements into consumer LLMs. Every output is a draft for a licensed MLO / UW to review.

1. Inbound lead triage (no NPI)

You are a pipeline coordinator. Here is today's inbound lead batch (de-identified fields only: est. FICO band, loan purpose, loan amount band, property type, state, lead source, self-reported timeline). Output: (1) rank by likely fund probability and size, (2) suggested LO assignment by specialization (purchase / refi / jumbo / VA / FHA / non-QM / investor DSCR), (3) contact-cadence template per lead (do not include rate quotes), (4) any leads that should be declined early (out-of-footprint state, loan type we don't offer). No adverse-action decisions here — those require ECOA-compliant process.

2. 1003 intake + completeness check

Review this URLA / 1003 draft and document package [paste inside NPI-safe system]. Output: (1) fields missing required for LE issuance within TRID 3-day clock, (2) red flags (income type needing 2-year history, gift letter required, large deposits, REO schedule gaps, non-occupant co-borrower), (3) asset and VOE routing needed, (4) suggested AUS path (DU vs LPA vs both), (5) file-ready checklist with responsibility tags (LO / processor / borrower). MLO reviews before LE goes out.

3. Self-employed income analysis draft

Draft a self-employment income analysis for [borrower, structure: sole prop / S-corp / partnership / LLC]. Inputs: 2 years personal 1040 with schedules, 2 years business returns (1120S / 1065), K-1s, YTD P&L, business bank statements. Output: Fannie Form 1084 / Freddie Form 91 style worksheet, with add-backs (depreciation, depletion, business-use-of-home, one-time items), distributions test, continuance analysis (ownership %, business liquidity), and qualifying monthly income figure. Flag any add-back that requires documentation the file does not yet have. UW signs; do NOT approve.

4. AUS pre-flight findings summary

Summarize this DU / LPA findings report for [loan]. Output: recommendation, eligibility, documentation waterfall (appraisal waiver, income rep/warrant, asset verification level), conditions list in human-readable language sorted by "borrower can provide today / needs outreach / UW decision", and any caution messages. Translate investor overlays into plain English. Do NOT change the recommendation. Processor executes, LO reviews, UW decides.

5. Conditions-clearing outreach

Draft a borrower email for conditions: [list from LOS]. Tone: clear, specific, no jargon. For each condition: what we need, why (1 sentence), how to upload, what happens next. If a condition touches sensitive context (bankruptcy explanation, medical collection letter, gap in employment), soften tone and offer a phone call. Do NOT quote rate or lock status in this email — use the lock-confirmation channel for those. LO signs.

6. TRID change-of-circumstance log

We had a change of circumstance on [loan]. Event: [describe — appraisal came in $X, rate locked, borrower changed loan amount, property tax reappraisal, etc.]. Draft the TRID change-of-circumstance memo: what changed, date event became known, which fees/tolerances move, whether a revised LE is required within 3 business days, CD timing impact on closing date, and a one-line borrower notice. Compliance reviews before revised LE is sent.

7. Fair-lending pulse (HMDA LAR)

Run a fair-lending disparate-impact pulse on our last 12 months HMDA LAR [inside secure env]. Compare decline rates, pricing (APR spread), and conditions frequency across protected classes. Output: tables + 3 largest gaps + plain-English narrative + suggested drill-down files. Flag any statistical anomalies that warrant formal BISG or comparative-file review. This is a management alert, NOT a fair-lending conclusion — compliance owns analysis and remediation.

8. MAP Rule / Reg Z compliant ad draft

Draft a social/email/paid ad for [product, e.g., 30-yr FHA purchase, 15-yr conventional refi]. Rules: if you quote a rate, APR must accompany it with assumption footnotes (loan amount, FICO, DTI, LTV, lock period); no "free", "guaranteed", "government-endorsed", "pre-approved" language; no misleading payment numbers; include NMLS ID; state licensing footer. Produce 3 variants + required disclosures. Compliance reviews. Do NOT publish without review.

9. Adverse-action notice (ECOA compliant)

Draft the adverse-action notice for [applicant] dated [date of decision]. Output: statement of action, specific principal reasons (no generic "did not meet our criteria"), ECOA §701 notice, FCRA §615 notice if credit-report based, reporting agency info, statement of applicant's rights, 30-day timing. Pull the actual decline drivers from the credit decision record. Do NOT invent reasons. UW signs.

10. Broker-principal weekly ops read

Write the broker owner's weekly operating summary. Inputs: new applications, locked, submitted, approved, CTC, funded, fallout, pipeline by stage, LO productivity, average days-to-CTC, top 3 investor mix, pricing competitiveness vs market, TRID error rate, compliance findings, net revenue and cost-per-funded-loan. Output: 1-page narrative + KPI table + 3 risks + 3 moves for next week. Written for the owner / principal.

Common mistakes we see

Pasting pay-stubs into consumer ChatGPT. GLBA NPI disclosure. Enterprise tier with DPA only.
AI-only adverse action. ECOA / Reg B requires specific, principal reasons — "model said no" is not compliant.
Rate quotes in AI-drafted borrower emails. Reg Z §1026.24 says if you advertise rate, APR must accompany with assumption footnotes.
AI-generated MSA / referral content. RESPA §8 landmines. Keep attorneys and compliance in the loop.
Lead-scoring that drives steering. LO Comp Rule §1026.36(d) prohibits steering to higher-margin products.
No TRID clock discipline with AI speed. Faster intake means you hit the 3-day LE / CD window sooner — document the application-complete trigger precisely.
Ignoring SR 11-7 / OCC 2011-12 model risk if you're depository-adjacent. Inventory models, validate, monitor.

A 60-day rollout for a 5-20 LO brokerage

Days 1-10: vendor DPA inventory, GLBA-compliant AI tier selection, update privacy notice, add AI section to WISP / ISP, assign compliance owner.
Days 11-20: deploy borrower-facing POS with AI OCR (Blend / SimpleNexus / Maxwell). Measure 1003-to-LE cycle time.
Days 21-30: pilot AUS pre-flight + conditions drafting inside LOS. Measure CTC days, conditions-per-loan.
Days 31-45: stand up monthly HMDA fair-lending pulse + TRID compliance review with AI assist. Document owner.
Days 46-60: MAP Rule / Reg Z-compliant content pipeline for LO marketing. Broker-principal weekly ops dashboard. Add AI policy to annual training.

Want the Happycapy Mortgage Broker toolkit?

Lead triage template, 1003 completeness check, SE income worksheet, AUS conditions translator, TRID change-of-circumstance log, and the broker weekly ops read — one pack.

See Happycapy Pro / Max →

Back to all Happycapy guides.