Mortgage brokers have quietly become heavy users of general-purpose AI. In AD Mortgage's 2026 survey of 250 brokers, 55% said they use AI tools daily or regularly, and more than 75% use tools like ChatGPT — yet fewer than one in three have adopted any mortgage-specific AI. The gap says a lot about where AI actually helps: not underwriting or pricing, but the writing and communication work that fills the hours between files.
That matches what brokers describe in practice: follow-up emails that chase documents, plain-English explanations of loan estimates, weekly notes to realtor partners, and social content that stays on the right side of advertising rules. Fannie Mae's lender research points the same direction — 73% of lenders cite operational efficiency as the main reason to adopt AI, while very few trust it anywhere near a lending decision.
The non-negotiable backdrop: mortgage origination is one of the most regulated consumer finance activities in the country. TILA and Regulation Z govern what you can say about rates, RESPA Section 8 governs what you can exchange with referral partners, and GLBA's Safeguards Rule governs every piece of borrower data you hold. AI drafts fast, but every number, disclosure, and data-handling decision stays on your license.
In AD Mortgage's 2026 survey of 250 mortgage brokers, 35% use AI tools daily and 20% regularly, and 72% expect AI to become significantly more important to their business within three years.Source ↗
More than 75% of brokers use general-purpose AI tools like ChatGPT, but fewer than one-third use mortgage-specific AI tools — and 57% say they need more training to use what they have.Source ↗
79.5% of mortgage brokers report using generative AI platforms such as ChatGPT, Claude, and Gemini, but only 34% use AI chatbots to help navigate mortgage guidelines and just 26% have deployed AI-backed underwriting or income-verification products.Source ↗
73% of mortgage lenders cite improving operational efficiency as the primary motivation for adopting AI, per Fannie Mae's Mortgage Lender Sentiment Survey — up from 42% when the question was first asked.Source ↗
A borrower is torn between a 30-year fixed and a 7/6 ARM, or between conventional and FHA, and the deciding conversation happens over email. You've already priced the scenarios in your pricing engine — the slow part is turning three columns of numbers into a comparison a first-time buyer can actually follow. AI is good at the translation; it is terrible at the pricing, so the numbers must always be yours.
Prompt
You are a communication assistant for a licensed mortgage broker. Your job is to turn loan scenarios I have already priced into a clear, neutral comparison for a borrower. You do NOT price loans, estimate rates, or calculate payments — use only the exact figures I provide below.
Loan scenarios from my pricing engine (rates, APRs, payments, closing costs): {{loan_scenarios}}
Borrower situation: {{borrower_situation}}
What the borrower says matters most to them: {{borrower_priorities}}
Write an email that:
1. Presents the options in a simple side-by-side summary (plain text, not a giant table).
2. Explains the practical tradeoff between them in 2-3 short paragraphs, tied to the borrower's stated priorities.
3. Ends with 2-3 questions that would help the borrower decide.
Hard rules:
- Use ONLY the numbers I provided. If a figure you need is missing, write [CONFIRM] instead of estimating it.
- Do not recommend one option as "best" — lay out tradeoffs and let the borrower decide with me.
- No guarantee language ("you'll save," "you're approved") — use "based on today's pricing" framing.
- Include a closing line noting these figures are illustrative, not a rate lock or commitment to lend.
- Eighth-grade reading level, under 300 words.
Fill in your details and the prompt updates live — then copy.
Every purchase file generates the same anxious questions: what are points, why is the APR higher than the rate, what is escrow, why did my cash-to-close change. You've answered each a thousand times, but writing a patient, personalized reply at 8 p.m. is where service quality slips. AI drafts the explanation; you make sure any numbers in it are real.
Prompt
You are an assistant to a licensed mortgage broker, drafting an educational reply to a borrower's question. You explain concepts; you never give advice on which loan product to choose, and you never state current rates, program limits, or loan amounts from memory.
The borrower asked: {{borrower_question}}
Concept or document involved: {{concept_or_document}}
Relevant context from their file, with figures I have verified: {{loan_context}}
Write a reply that:
- Answers the actual question first, in one or two sentences, at an eighth-grade reading level.
- Uses one concrete worked example ONLY if I supplied the figures for it; otherwise explain the mechanics without numbers.
- Anticipates the most likely follow-up question and answers it briefly.
- Closes by inviting them to call or reply with questions.
Hard rules:
- Any number not in my provided context must appear as [CHECK] — never fill in a rate, loan limit, mortgage insurance factor, or seasoning requirement from memory, because these change and your training data may be stale.
- No product recommendations and no "you should" financial advice — educational framing only.
- Warm but not chatty. Under 250 words. No exclamation points.
Fill in your details and the prompt updates live — then copy.
Files don't die from rate shocks; they stall waiting on a bank statement, a letter of explanation, or an updated pay stub. Chasing documents politely — for the third time, near a rate-lock deadline — is exactly the repetitive writing AI handles well. Personalized AI-drafted email sequences are one of the highest-ROI uses in the industry, with mortgage CRM vendors reporting 2-3x reply rates over stock templates.
Prompt
You are an assistant to a mortgage broker, writing a document follow-up sequence for a borrower whose file is waiting on items. Tone: helpful teammate working toward THEIR closing date, never a debt collector.
Borrower first name: {{borrower_first_name}}
Outstanding items and the reason each is needed: {{outstanding_documents}}
Deadline context: {{deadline_context}}
Write a 3-touch sequence:
1. Day 1 email — friendly, lists each item with a one-line plain-English reason it's needed, and one clear instruction: upload through the secure portal link [PORTAL LINK].
2. Day 3 text — under 300 characters, references the closing timeline, offers help if they're stuck on any item.
3. Day 5 email — short, explains specifically what happens to the timeline if items arrive this week vs. next, and offers a 10-minute call to knock them out together.
Hard rules:
- Use only the reasons I provided for each document; if I gave no reason, say "the underwriter needs this to finalize your approval" — do not invent one.
- Every message must say to use the secure portal and NEVER to email documents containing Social Security numbers or account numbers.
- No pressure phrases like "final notice" or "urgent action required." One ask per message.
Fill in your details and the prompt updates live — then copy.
Agent relationships drive purchase business, and the brokers who win them show up every week with something useful — not "thinking of you!" filler. Mortgage CRM vendors now build entire workflows around AI-drafted realtor outreach. The catch is RESPA: what you send agents can be valuable content, but the relationship can never become paying for referrals.
Prompt
You are a marketing assistant for a licensed mortgage broker writing a weekly email to real estate agent partners. The goal is to be genuinely useful to agents in 60 seconds of reading, so they think of me when their buyer needs financing.
This week's market notes, with figures I have verified and their sources: {{market_notes}}
Audience: {{audience}}
This week's call to action: {{call_to_action}}
Write a 150-200 word email:
- Subject line under 7 words, specific, no clickbait.
- Open with the single most useful fact for agents this week, then 2-3 sentences of what it means for their buyers in practical terms.
- One short "deal note" section: a generic scenario lesson (no borrower details) agents can learn from.
- Close with the call to action, low-pressure.
Hard rules:
- Use only the figures and facts I provided; do not add market statistics from memory. Mark anything missing as [DATA].
- No predictions ("rates will fall") — describe what happened, not what will happen.
- Do not state specific rates or payment amounts; describe direction and context instead, since this is a mass communication.
- No promises of approval speed or terms. Professional, plain, zero hype.
Fill in your details and the prompt updates live — then copy.
Consistent social posting is how loan officers build a purchase pipeline without buying leads, and AI removes the blank-page problem for posts and short video scripts. But mortgage content is regulated advertising: one casual "rates as low as..." post is a Reg Z violation, and regulators have said social media's format limits are no excuse for missing disclosures.
Prompt
You are a social media assistant for a licensed mortgage broker. Write educational content that builds trust with future homebuyers — strictly educational, never an advertisement of specific loan terms.
Topic: {{topic}}
Platform and format: {{platform}}
My local market context: {{local_market}}
Produce three options: two text posts (one hook-first, one story-first) and one 45-second video script with a spoken hook in the first five seconds and a simple closing call to action ("follow for more" or "questions? DM me" — nothing transactional).
Hard rules:
- NO interest rates, APRs, payment amounts, down payment amounts, or loan terms — these are Reg Z trigger terms that would require full disclosures. Teach the concept instead.
- Never use "guaranteed approval," "lowest rates," "no closing costs," "everyone qualifies," or any superlative claim.
- Do not invent statistics or program rules. If a stat would strengthen the post, insert [CITE: what's needed] and I'll supply or cut it.
- Include placeholder lines for my NMLS ID and Equal Housing Opportunity notice.
- Plain conversational language; no hashtag walls (max 3); write like a person, not a brand.
Fill in your details and the prompt updates live — then copy.
A conditional approval lands with a list written for processors, not people: "LOX for NSF occurrences," "VVOE within 10 days of close," "source large deposit." Translating that into a to-do list a borrower can execute — without panicking them — takes an evening you don't have when three files hit at once. AI is a strong translator here, as long as the condition list you paste is scrubbed of personal data.
Prompt
You are an assistant to a mortgage broker. Convert this underwriting condition list into a clear action plan for the borrower. The list below is sanitized — refer to the borrower only as {{borrower_first_name}}.
Conditions from the underwriter: {{condition_list}}
Scheduled closing date: {{closing_date}}
Produce:
1. A short reassuring opener — conditional approval is normal and these are the last steps.
2. Three groups: "You handle" (borrower actions, each with plain-English description, exactly what to provide, and a suggested done-by date working back from closing), "I handle" (broker/lender items, so they see the full picture), and "Do NOT do" — standard warnings: no new credit or large purchases, no moving money between accounts, no changing jobs without calling me first.
3. A closing line offering a call to walk through anything confusing.
Hard rules:
- Translate only what each condition says; if the underwriter's reason is not stated, do not speculate about why they asked. Mark genuinely ambiguous items [ASK LO] for me to clarify.
- Never advise the borrower to pay off debts, close accounts, or move funds — those decisions can create new conditions and must go through me.
- Plain language, no underwriting jargon left untranslated, under 350 words.
Fill in your details and the prompt updates live — then copy.
Common questions from mortgage brokers
Am I allowed to use ChatGPT or Claude as a mortgage broker?
Yes — no rule bans drafting emails or content with AI. But as a financial institution under GLBA, you're responsible for safeguarding borrower data, so use a business-tier AI plan whose terms exclude training on your inputs, and keep all nonpublic personal information (names tied to finances, SSNs, income docs, account numbers) out of prompts entirely. Check your wholesale lender agreements and company policy too.
Can AI quote rates or pre-qualify borrowers for me?
No. Generative AI fabricates plausible-looking numbers and its training data lags current pricing and guidelines, and discussing loan terms with a consumer is licensed MLO activity. Use AI to explain and format numbers you produced in your pricing engine — never to produce them.
Do TILA and RESPA really apply to an Instagram post I wrote with AI?
Yes. AI-drafted content is advertising like any other: stating a rate, payment, or loan term triggers Reg Z disclosure requirements, superlative claims like "lowest rates" invite UDAAP trouble, and anything of value flowing to referral partners has to survive RESPA Section 8. Run AI drafts through the same compliance review as human-written ads, NMLS ID included.
Will AI replace mortgage brokers?
The evidence says it's replacing tasks, not brokers — surveys show adoption concentrated in writing, follow-up, and marketing, while under a third of brokers use AI anywhere near underwriting. Complex files, licensing, and agent relationships stay human; the practical risk is losing business to brokers who respond faster because AI handles their drafting.