How to Extract Data From a Closing Disclosure

To extract data from a Closing Disclosure, run the five-page CD (a PDF or scan) through software that reads it with OCR, identifies the borrower, loan terms, interest rate, the itemized closing costs, and the cash-to-close figures with AI, and outputs structured fields your loan or title system can use. A processor no longer has to read pages two and three line by line and re-key origination charges, title fees, prepaids, and escrow amounts into a spreadsheet or LOS. Upload a Closing Disclosure above and the tool pulls the loan amount, APR, monthly payment, the borrower-paid and seller-paid columns, the settlement charges, and the cash to close into clean rows you can review and export in minutes. This guide covers which fields to extract from a CD, how the extraction works across all five pages, how it handles the borrower and seller columns, and how to feed the data straight into your closing workflow.

Last updated June 2026.

What data should you extract from a Closing Disclosure?

Extract the fields your closing, audit, or post-close process actually uses: the loan and borrower header, the five loan terms, the projected payments, the full itemized closing costs, and the cash-to-close reconciliation. A Closing Disclosure follows the TRID format the CFPB standardized, so the same labeled boxes appear in the same places on every CD, which is exactly what makes it a good document to extract from automatically. The fields worth capturing are consistent across lenders.

A standard CD carries these fields:

• Page 1 header (closing date, disbursement date, settlement agent, file number, property address, sale price, loan term, purpose, product, loan type, loan ID, MIC number)
• Loan terms (loan amount, interest rate, monthly principal and interest, whether each can increase, prepayment penalty, balloon payment)
• Projected payments (principal and interest, mortgage insurance, estimated escrow, estimated total monthly payment, estimated taxes/insurance/assessments)
• Costs at closing (total closing costs, cash to close)
• Loan costs (page 2) (origination charges, services borrower did and did not shop for, with the borrower-paid, seller-paid, and paid-by-others columns)
• Other costs (page 2) (taxes and government fees, prepaids, initial escrow payment at closing, and other charges like the owner's title policy)
• Calculating cash to close (page 3) (the loan estimate vs final comparison, down payment, deposit, seller credits, adjustments)
• Summaries of transactions (page 3) (the borrower's and seller's debits and credits side by side)
• Loan disclosures and contacts (pages 4 to 5) (escrow, APR, total interest percentage, lender, mortgage broker, real estate brokers, and settlement agent contact details)

For a quick stip-check you may only need the loan amount, rate, cash to close, and closing date. For post-close QC, audit, or building a Uniform Closing Dataset (UCD) feed, you want every line so the file reconciles against the Loan Estimate.

How does AI extract data from a Closing Disclosure?

AI reads the CD the way a person does, then converts it to structured data. First OCR turns the scanned or PDF pages into machine-readable text while preserving the layout. Then a model trained on document structure finds each labeled field by its position and context, so it knows the number next to "Loan Amount" is the loan amount and the figure in the Borrower-Paid column on the origination line belongs to the borrower, not the seller. It captures the line-item tables on pages two and three including the multi-column layout, validates the math where it can (loan costs plus other costs should reconcile to total closing costs), and outputs the result as columns or JSON. Because the CD is a fixed federal form, accuracy on a clean digital PDF is high, and well-implemented extraction typically reaches 95 to 99 percent field accuracy on legible documents, with flagged low-confidence reads sent to a person to confirm.

How does it handle the borrower-paid and seller-paid columns?

It reads each cost line as a row and keeps the three columns (Borrower-Paid, Seller-Paid, and Paid by Others) as separate fields, and it tracks the At Closing versus Before Closing split within them. This matters because the same fee can appear in more than one column, and posting a seller-paid charge as borrower-paid throws off the cash to close and the net-to-seller. Good extraction returns each charge with its description, its dollar amount, and which party and timing it was assigned to, so your downstream system or auditor sees exactly who paid what without anyone re-reading the page.

What is a Closing Disclosure?

A Closing Disclosure is a five-page form that gives the final terms and costs of a mortgage loan. Under the TILA-RESPA Integrated Disclosure (TRID) rule, the lender must give it to the borrower at least three business days before closing so the borrower can compare the final numbers to the earlier Loan Estimate. It is one of several forms a lender assembles, and it helps to know what documents are in a mortgage loan file before you automate any of them. It lists the loan amount, interest rate, monthly payment, the full itemized closing costs, and the cash the borrower needs to bring, plus contact details for everyone involved in the transaction.

How many pages is a Closing Disclosure?

A Closing Disclosure is five pages. Page one summarizes the loan terms, projected payments, and costs at closing. Page two itemizes the loan costs and other costs in the borrower-paid, seller-paid, and paid-by-others columns. Page three calculates cash to close and shows the borrower and seller transaction summaries. Pages four and five carry the loan disclosures, the APR and total interest percentage, and the contact information. Extraction software reads all five and returns the fields from each.

What is the 3-day rule for a Closing Disclosure?

The three-day rule requires the lender to deliver the Closing Disclosure to the borrower at least three business days before the loan closes. Business days include Saturdays but exclude Sundays and federal holidays. The clock restarts if the APR changes beyond the allowed tolerance, a prepayment penalty is added, or the loan product changes, which can push the closing date. Fast, accurate extraction of the final CD helps lenders confirm the figures and the timing well inside that window.

What is the difference between a Closing Disclosure and a Loan Estimate?

The Loan Estimate is the early, good-faith estimate of the loan terms and costs the lender gives within three business days of application; the Closing Disclosure is the final version delivered three business days before closing. The two forms mirror each other on purpose so the borrower can compare them line by line. A core post-close QC task is exactly that comparison, and extracting both documents into the same structured fields lets you diff them automatically instead of eyeballing two PDFs.

How do you read the cash to close on a Closing Disclosure?

The cash to close is on page one and is detailed on page three in the Calculating Cash to Close table. It is the total amount the borrower must bring to closing, built from the total closing costs, the down payment, the loan amount, any deposit already paid, seller credits, and adjustments. Page three shows the Loan Estimate figure next to the final figure and whether it changed, so the borrower and the lender can see what moved. Extraction pulls the final cash to close and the components behind it so the number is auditable.

Can you extract data from a scanned or photographed Closing Disclosure?

Yes. The software runs OCR first, so it reads scans, faxed copies, and clear phone photos as well as native PDFs, though a sharp digital PDF gives the most accurate result. Skewed, low-resolution, or partially cut-off pages lower confidence, so the tool flags uncertain fields for a quick human check rather than guessing. For lenders and title companies handling files from many sources, this is what lets one workflow handle the messy scans alongside the clean exports.

How extraction fits a lender or title workflow

For a single file, upload the CD and review the fields. For volume, the work usually runs one of two ways: lenders and post-close QC teams batch CDs through an API into the LOS or a UCD feed, and title and settlement teams pull the final figures to reconcile the file and check them against the Loan Estimate and the settlement statement. Either way the pattern is the same. Extract, review only the flagged low-confidence fields, and push the structured data forward so nobody re-keys five pages of numbers. Lenders pairing CD extraction with broader loan underwriting document analysis software get a clean, structured read across the whole borrower file, not just the closing package. Title and escrow teams that also abstract the real estate documents behind the deal can lean on lease abstraction software for the leases that ride along with commercial closings. And when a transaction requires proof of coverage from the parties involved, tracking it with certificate of insurance tracking software keeps the closing file complete.

The bottom line

Extracting data from a Closing Disclosure turns five pages of fixed-format figures into a structured feed your closing and post-close process can run on. Capture the loan terms, the full itemized costs in their borrower and seller columns, and the cash-to-close reconciliation, review only the uncertain reads, and diff the result against the Loan Estimate so the file clears compliance. To handle CDs at volume, Closing Disclosure OCR software reads every page and exports clean fields, the broader mortgage document processing software handles the rest of your loan and title documents, and the OCR API wires extraction straight into your LOS. Upload a Closing Disclosure at the top of this page to see the fields it pulls.