OCR Real Estate Closing Documents Into One Record

To OCR real estate closing documents, gather the whole closing package, split it at document boundaries, run recognition on each part, and export a searchable PDF for the file plus Excel for the settlement figures. Check recorded instruments and legal descriptions against the page images before you rely on a word of the text.

A closing file is opened again long after the deal is done: at a refinance, a sale, a boundary question, a tax appeal or a claim on the title policy. Whoever opens it wants one answer from several hundred pages. The work below is aimed at that moment.

Open the deed first in the converted package. Search it for warranty deed or quitclaim deed and for the legal description you will follow through the chain of title. Then compare the title insurance commitment and the escrow totals with the images, including any exception that mentions marketable title. Copy book and page numbers off the scan.

What is in a closing package and why is it scanned?

A closing package is the full set of papers signed, delivered and recorded to complete a real estate transaction. It is scanned because closings still run on ink. Documents are signed at a table, notarized, sent for recording, and returned on paper weeks later with stamps that no electronic draft ever had.

The contents vary with the deal, but most packages hold some mix of these.

Even when every draft began as a Word file, the final versions that count are the signed ones. They arrive as a mix: a scan from the title company, a phone photo of a signature page, a fax from a payoff lender, and recorded copies returned by the recording office. Put together, they make one large image-only PDF or a folder of them. The post on why you cannot copy text from a PDF shows how to confirm which of your files are images.

Which closing documents are worth converting?

Convert the closing documents you will later search, quote or calculate from: the purchase agreement, the deed, the loan documents, the title policy with its exceptions, and the settlement statement. Drawings and large technical reports gain little. Deciding this first keeps page volume, and review time, in proportion to the value.

Two practical points follow from the table. First, check whether a document already has a text layer before you convert it. A report emailed by a consultant often does. Second, where the deal included leases, treat them as their own task. The post on converting a commercial lease PDF to Word deals with them.

Converting a closing binder to a searchable record

Convert a closing binder to a searchable record by working from the closing index: split the binder into one file per document, name each file by its tab, convert them, and file the searchable PDFs under the same index. One file per document means a search hit tells you which instrument you are in.

1. Start from the index. Most binders have a closing checklist or table of contents with numbered tabs. Use it as your file list. If there is none, write one as you go.
2. Split at document boundaries. A single scan of the whole binder is hard to use and may be too large to upload. Schema Law accepts PDF, PNG or JPEG files up to 20 MB, with up to 250 pages per document on the Solo plan and 1,000 on the Team plan. A binder scanned in color at high resolution can pass 20 MB quickly, so splitting by tab solves two problems at once.
3. Name files so they sort in binder order. For example 07_Deed_recorded.pdf, 12_Note.pdf, 21_Settlement_Statement.pdf. The tab number comes first, with a leading zero.
4. Prefer the recorded copy. Where you hold both the signed deed and the copy returned from recording, convert the recorded one. It carries the recording information you will be asked for later. Keep the other as well.
5. Upload and let each file be read. The homepage says there is nothing to prepare before you upload. You can choose several files at once, and they are read one after another. Offices that close many deals can compare the page allowance of each plan.
6. Export a searchable PDF of each file. The export is your original pages with a hidden layer of text added, so stamps, signatures and seals still look as they did. Export Excel as well for the documents that are mostly numbers.
7. Check the documents that matter most. The next two sections cover the recorded instruments and the figures.
8. File the set and note what was done. Record in the file which copy of each document was scanned, and that the text layer came from OCR.

One detail about the searchable PDF is worth knowing before you rely on it. In a Schema Law export, the hidden text for each page is laid down in reading order from the top of the page, not placed exactly beneath each printed word. A search takes you to the right page. The highlight may not sit on the word itself, so read the page once you land on it. The guide on how to make a scanned PDF searchable explains text layers in full.

If you are scanning the paper binder yourself, the scanner settings are covered in the guide to digitizing law firm paper files.

Deeds, legal descriptions and recording stamps

Deeds, legal descriptions and recording stamps are the hardest parts of a closing file to convert correctly and the costliest to get wrong. A legal description is a string of bearings, distances and references in which every character counts. Treat converted text of a recorded instrument as a draft to proofread, never as a copy.

Legal descriptions

A metes and bounds description is close to a worst case for recognition. A call such as N 89°14'32" E, 214.60 feet packs a compass letter, a degree sign, a minute mark, a second mark and a decimal distance into a few characters. Common failures include:

• the degree sign read as a zero, a letter o or a superscript, or dropped;
• minute and second marks read as apostrophes and quotation marks, or lost;
• the letters N, S, E and W confused with nearby characters, or a direction dropped;
• a 1 read as a 7 or a lower-case L, and a 5 read as an S;
• a decimal point lost, turning 214.60 into 21460.

Lot and block descriptions are shorter but carry their own risk in plat book and page numbers, lot numbers and subdivision names.

Proofread any description you plan to reuse with two people: one reads the recorded image aloud, character by character, while the other follows the converted text. For metes and bounds, a surveyor or a deed plotting tool can test whether the calls close back to the starting point, which catches errors that reading misses. Never paste a converted description into a new deed, mortgage or easement without that check.

Old deeds raise a further problem. Chain-of-title documents may be typewritten carbons, photostats with white text on black, or handwritten in cursive. Expect weak results and read them yourself. The post on OCR for handwritten wills and probate records covers old handwriting, and the limits it describes apply equally to old deed books.

Recording stamps

When an instrument is recorded, the recording office typically adds its own information: an instrument or document number, a book and page or similar reference, the date and time of recording, and fees or transfer tax paid. This may be a printed label, an ink stamp, a barcode strip or a cover sheet added to the front.

That information is what the next lawyer will ask for, and it is fragile under OCR. Stamps are often faint, tilted, or struck across the first lines of the document. Recognition may read the stamp and damage the text beneath, or the reverse. The post on whether OCR can read stamps over text explains what to expect.

Do not depend on search to find recording data. Type it into your closing index by hand from the image, one row per recorded instrument:

The entries are invented. With the table in the index, the recording data is searchable whatever happened to the stamp.

Notary acknowledgments need the same eye. The typed venue and date usually convert. The embossed or inked seal and the commission expiry inside it often do not. Confirm them from the image if they matter.

Settlement statements as tables

Settlement statements are tables of charges and credits, so export them to a spreadsheet and prove the conversion by re-adding the columns. If your totals match the totals printed on the statement, the rows were read correctly. A mismatch tells you a figure was misread or misplaced.

The form depends on the transaction. Where a closing disclosure was issued for a mortgage loan, the Consumer Financial Protection Bureau publishes a closing disclosure explainer built around a sample form of five pages. In that sample, the second page itemizes the closing costs, and the third covers the seller credit, adjustments and the cash to close. Other deals use a settlement statement prepared by the closing agent, often with buyer and seller columns side by side.

Both layouts are dense. Small type, narrow columns, section letters and line numbers, and amounts that sit in one of several columns depending on who paid. To convert one:

1. Upload the statement as its own file and export Excel. Excel export is included on every plan.
2. Open the workbook. Each detected table is on its own sheet, named with a number and the page it came from. Text outside the tables is on a sheet called Text.
3. Convert the amount columns to numbers. Cells arrive as text exactly as read, so strip dollar signs and commas first.
4. Check that each amount sits in the right column. A charge that slid from the seller's column to the buyer's will not change the grand total, only who paid.
5. Re-add each section and compare it with the printed subtotal.

A short worked check, with invented figures:

Debits less credits should equal the cash figure: 642,065.00 less 508,412.60 is 133,652.40. Run the same test on the seller's side. Then spot-check the lines a future reader will care about, such as prorations, payoff amounts and holdbacks, against the image.

How tables are detected, and how to repair shifted or merged columns, is covered in the guide to getting a scanned table into Excel.

Finding a term across the whole package years later

Finding a term across the whole package years later depends on two things you set up at closing: every document has a text layer, and an index says what each file is. With both, your PDF reader or your computer's file search can run one query across the folder and list each document that contains the term.

Questions a converted package can answer quickly:

• Is there an easement, and where? Search the title policy, the exception documents and the deed for "easement", "right of way" and the name of the utility.
• What survives closing? Search the purchase agreement for "survive" and "survival".
• Can the loan be prepaid? Search the note for "prepay".
• Who signed for the seller, and in what capacity? Search the authority documents for the signer's surname.
• What was the tax proration based on? Open the settlement workbook.

Search habits that make up for recognition errors:

• Try more than one form. "Right of way", "right-of-way" and "ROW" are three different searches.
• Search for a distinctive fragment. For a parcel number such as 12-34-567-008, search "567-008" as well, in case a digit elsewhere was misread.
• Search names by their least common word. An entity called Harlow Properties LLC is easier to find by "Harlow" than by the full name with its punctuation.
• Do not trust an empty result. A search that finds nothing means the text layer does not contain that string. The page still might. For a question that matters, read the document the index says should hold the answer.

Keep the index with the package, as a spreadsheet or a first page, listing tab, document, date, parties, recording data and file name. In ten years the index is what tells a stranger which of sixty files to open.

Closing files hold bank details, tax identification numbers and personal information about buyers and sellers, so decide where they may go before you upload them. The commitments on the Schema Law homepage state how uploads are treated, including that you can delete any document at any time. The post on whether OCR software stores your documents explains what to ask of any tool about retention.