P
AI for Mortgage Loan Processor

Custom GPT: Build a "UW Condition Resolver" for Your Pipeline

For Mortgage Loan Processors

Tools: ChatGPT Plus | Time to build: 45-60 minutes | Difficulty: Intermediate-Advanced Prerequisites: Comfortable using ChatGPT for condition letter drafting — see Level 3 guide: "Build a Condition Letter Prompt Library"

What This Builds

A Custom GPT named "UW Condition Resolver" that you open each day like a specialized coworker. You paste any underwriting condition text, and it immediately identifies the condition type, tells you exactly what documentation to collect, drafts the condition letter to the borrower, and suggests a cover note to the underwriter — all without you explaining the context from scratch. Over time, it becomes the fastest condition-handling tool in your workflow.

Prerequisites

  • ChatGPT Plus subscription ($20/month) — Custom GPTs require a paid account
  • Comfortable using ChatGPT for basic condition letter drafting (see Level 3 guide above)
  • A list of your 8-12 most common condition types (spend 10 min writing these down)
  • Cost: $20/month (ChatGPT Plus)

The Concept

A Custom GPT is like a specialized employee you've trained once. You give it a set of instructions — who it is, what it knows, how to respond — and every conversation starts from that training without you re-explaining it. The regular ChatGPT knows about everything generally; your Custom GPT knows about mortgage conditions specifically and responds in exactly the format you need.

Think of it this way: regular ChatGPT is a smart generalist assistant who can help with anything. Your Custom GPT "UW Condition Resolver" is a mortgage conditions expert who works your way and already knows your lender's preferences — on call 24/7, no training required after setup.

Build It Step by Step

Part 1: Access the GPT Builder

  1. Open chatgpt.com and make sure you're logged into your Plus account
  2. Click "Explore GPTs" in the left sidebar (or go to chatgpt.com/gpts)
  3. Click "Create" in the top right corner
  4. You'll see a "Create" tab (natural language builder) and a "Configure" tab (manual setup). Click "Configure" — this gives you direct control.

What you should see: A split screen with the Configure settings on the left and a preview chat on the right.

Part 2: Configure Your Custom GPT

Name: UW Condition Resolver — [Your Lender Name]

Description: "Mortgage loan processing assistant. Identifies UW condition types, specifies required documentation, drafts condition letters to borrowers, and prepares condition response notes."

Instructions (paste this entire block, customizing the [BRACKETS]):

Copy and paste this
You are a mortgage underwriting condition resolution assistant working for [YOUR NAME], a Loan Processor at [LENDER NAME].

YOUR ROLE:
When given an underwriting condition (text pasted from the condition report or email), you:
1. Identify the condition type (from the list below or use your best judgment for unlisted types)
2. State clearly what documentation is needed to satisfy the condition
3. Draft a condition letter to the borrower explaining what's needed and why
4. Draft a brief underwriter cover note summarizing what is being submitted

CONDITION TYPE LIBRARY:
- Large deposit LOE: Need written LOE from borrower explaining source of any deposit >50% of monthly income. Acceptable sources: payroll, transfer from verified account, gift (with gift letter), sale of personal property (with bill of sale), tax refund, insurance payout.
- Employment gap LOE: Need signed LOE explaining any gap in employment >30 days. Include start/end dates and what borrower was doing during the gap.
- Credit inquiry LOE: Need signed LOE for each credit inquiry within 120 days of application. Explain purpose of inquiry and whether new account was opened.
- Rental income documentation: Need executed lease, 2 years Schedule E, 12 months bank statements showing rent deposits.
- Gift letter: Need signed gift letter from donor stating amount, relationship to borrower, that no repayment is required. Bank statement showing gift transfer.
- Self-employment income: Need 2 years personal tax returns (all schedules), 2 years business tax returns if corporation, YTD P&L, 3 months business bank statements.
- Title exception explanation: Need title company clarification letter or legal document addressing the exception.
- Address discrepancy: Need LOE if borrower's current address differs from what's on credit report or application.
- Bankruptcy explanation: Need LOE from borrower explaining the circumstances of the bankruptcy and what has changed since.

LETTER FORMAT:
- Professional business letter, not email format
- Opening: "Dear [Borrower Names],"
- Body: Explain the condition clearly, what documentation is needed, why it's required, and by what date
- Closing: "Please don't hesitate to contact me with any questions. [YOUR NAME], Loan Processor, [LENDER NAME]"
- Maximum 250 words unless complexity requires more

TONE: Professional and warm. Borrowers are often stressed. Reassure them that conditions are normal and routine.

ALWAYS produce both: (1) the borrower letter and (2) the UW cover note.

Conversation Starters (add 3-4 to help you remember how to use it):

  • "Condition text: [paste condition here]"
  • "Draft a follow-up for missing [document] from [name]"
  • "What documentation satisfies a large deposit condition?"

Click Save (top right). Choose "Only me" for visibility (keep it private).

Part 3: Test and Refine

Test 1: Standard condition In the preview chat (or open your new GPT from "My GPTs"), paste: "Condition: 12 months cancelled checks or bank statements evidencing rental payment history for borrower's current residence at 88 Pine Street, Springfield."

You should receive: (1) the condition type identified as "rental payment history," (2) clear documentation needed, (3) a formatted condition letter, (4) a UW cover note.

Test 2: Unusual condition Try a condition type not in your library: "Condition: Need explanation for derogatory trade line — collection account from 2021 for medical debt ($850) that appears on credit report."

The GPT should handle this gracefully even though it's not in your explicit library.

Refine based on what you see:

  • If the letters are too long: add "Maximum 200 words per letter" to your instructions
  • If the format is inconsistent: add a more specific format example to your instructions
  • If a common condition type is missing: add it to your Condition Type Library

Real Example: Running a Full Condition Batch

Setup: Your Custom GPT "UW Condition Resolver" is built and ready.

Input: You open the GPT each morning with your UW condition report. For a file on the Ramirez family, the underwriter issued 3 conditions.

You paste: "Run these 3 conditions for the Ramirez family (Carlos and Ana):

  1. Need LOE for $4,200 deposit on 01/15/2024 in Chase account
  2. Need LOE for 45-day employment gap July-August 2022
  3. Need gift letter for $12,000 gift from parents used for down payment. Closing target: April 30."

What you get: Three complete condition letters — one for each condition — formatted and ready to review. Plus a three-item UW cover note listing what's being submitted for each condition. You review, personalize with any file-specific details, and your Ramirez file condition package is ready in 8 minutes instead of 45.

Time saved: 3 conditions × ~15 min each = 45 min manual → ~8 min total with Custom GPT.

What to Do When It Breaks

  • GPT gives generic output that ignores your library → Open your GPT settings, check that Instructions saved correctly. Copy-paste the Instructions block again if needed.
  • Letters don't match your lender's preferred format → Add a sample letter to your GPT's Knowledge (upload a text file with a sample well-formatted letter).
  • GPT makes up documentation requirements that aren't right → Add "Only specify documentation from the Condition Type Library above. For unlisted condition types, ask me for clarification before specifying requirements."
  • GPT loses context mid-conversation → Start a new conversation — Custom GPTs reset between sessions, which is actually helpful for keeping each file batch clean.

Variations

  • Simpler version: Skip the Custom GPT and just use the Level 3 prompt library — same output, more manual effort each time
  • Extended version: Add a "Knowledge" file to your Custom GPT with your lender's full condition type glossary and required documentation list — makes the output even more lender-specific

What to Do Next

  • This week: Build the Custom GPT and test on 5 real conditions from your current pipeline
  • This month: Refine the Instructions based on what you've noticed — add condition types you encounter frequently
  • Advanced: Add your lender's common turn-times and milestone language to the instructions so the GPT can also draft status update emails with accurate timing

Advanced guide for Mortgage Loan Processor professionals. Requires ChatGPT Plus ($20/month). Custom GPT interfaces may change — if the "Create" flow has changed, look for "My GPTs" in the ChatGPT menu.