Scanned Contracts: What You Can Do With a Signed PDF

Scanned contracts are photographs of signed pages, so you cannot search, quote, compare or edit them until OCR has added text. Start by converting one agreement in full, exhibits included. Then pick the output that fits the job in front of you, and read the figures, names and negatives against the signed image.

When the signed PDF is the only copy you hold and the Word file it was printed from cannot be found, the scan is your starting material. This post walks through it part by part, with pointers to the posts that cover each job in depth.

What can you do with scanned contracts once they have text?

Once scanned contracts have text, you can search them, copy clauses out of them, compare one against another, chart their terms and draft from their language. None of that works on a bare scan, because a scan stores the look of each page and no letters.

The gains fall into five groups.

Finding. A text layer lets Ctrl+F land on "indemnif" in section 11 of a sixty-page agreement. Across a folder of agreements, your desktop search or document system can index the words and return every contract that mentions a named subsidiary.

Quoting. You can copy a clause into an advice email or a demand letter without retyping it. Retyping a limitation of liability clause by hand is slow and introduces its own mistakes.

Comparing. Two contracts with text can be put through a comparison tool. Two pictures cannot.

Charting. With text you can pull the parties, dates and amounts into a table, one row per agreement.

Drafting. The language of a signed deal becomes the starting point for its renewal, its amendment or the next deal with the same counterparty.

What text does not give you is a new original. The scan of the signed pages stays in the file exactly as received, and everything made from it is a tool for reading it. Whether that scan can stand in for the paper it was made from is a separate question, taken up in the post on whether a scanned copy of a signed contract counts as an original.

Read the signed pages as an express contract. On a bilateral contract you will usually see two signature blocks, so check that both names and both dates are on the image. If the preamble states consideration, copy the amount or the description with the page number. If only one block is signed, write that down. Whether you are looking at a unilateral contract is for the lawyer.

How does OCR treat each part of a signed agreement?

OCR treats a signed agreement unevenly. Numbered clauses in body type usually convert well. The cover page, the signature block, the schedules and anything written by hand convert worse, because they break the pattern of plain lines of print that recognition software handles best.

Here is the agreement from front to back.

The preamble and the definitions

The preamble is short and carries a high share of the facts people later rely on: who the parties are, what kind of entity each one is, and the date. A blank there may have been filled in by pen. Printed text beside handwriting is a hard case for recognition, so read the preamble against the image every time.

Definitions suffer in a quieter way. A contract marks a defined term with capitals and quotation marks, and OCR can return curly quotes as straight ones, as commas, or not at all. The words survive but the signal that a term is defined can vanish. If you plan to reuse the text, check that "Services" and "services" still differ where the signed page says they do.

The signature block

Recognition reads what is printed in a signature block: "By:", the typed name, the title, the entity name above the line. The signature itself is a drawing. It may come back as nothing, or as a few random letters where the pen crossed the printed line. Dates written by hand beside the signature are unreliable. The post on extracting signatures and dates from scanned documents covers how to record who signed and when.

Exhibits and schedules

The back of the agreement looks least like the front. A fee schedule is a table, a site plan is a drawing, and an attached form of guaranty is a second document with its own numbering. Expect to spend more checking time per page here than anywhere else. The post on OCR for contract exhibits deals with the attachments that are not tables.

Taking one scanned contract from signed PDF to usable text

To take one scanned contract from signed PDF to usable text, confirm the file is complete, convert every page in one pass, export the format your task needs, and then check the result against the image. Budget more time for checking than for converting.

A fictional example runs through the steps. Your client, Alder Creek Bakery Co., sends a 22-page equipment lease with Tidewater Leasing Corp., signed in 2022. It arrives as one PDF from an office copier.

1. Confirm it is a scan. Try to select a word on page 5. If the whole page highlights as one block, there is no text. Some copiers add a rough text layer of their own, so also paste a sentence into a blank document and see whether it reads correctly.
2. Count the pages. The footer says "Page 1 of 22". The PDF has 21 pages. Stop and find the gap before you convert. The post on missing pages in a scanned contract lists the checks.
3. Save the received file untouched. Put it where it will not be overwritten, under a name that says what it is: 2022-06-14 Equipment Lease - Tidewater - SIGNED SCAN.pdf.
4. Convert the whole file. Do not trim the exhibits to save pages. The payment schedule in Exhibit B is the part you will be asked about.
5. Read the result page by page on screen before exporting, with the original beside it. You are looking for pages that came out empty, sideways or scrambled, not yet for single characters.
6. Export what the task needs. The table in the next section maps tasks to formats.
7. Check the high-risk characters against the image: amounts, percentages, dates, section references, names and negatives.
8. File the outputs beside the scan with names that show they were derived from it.

In Schema Law, step 4 means uploading the PDF in the browser. The file can be up to 20 MB, and the pages are read into text and tables that you view page by page before you export. A 22-page lease is over the three-page limit of the free plan, so use the free plan to try three pages and see the pricing page for the plans that take longer documents.

If the copier split the agreement into several files, or a counterpart signature page arrived separately, put the pages in order before converting. The post on counterpart signature pages covers assembling one executed copy.

Which contract task needs which kind of output?

Each contract task wants a different output. Searching wants a searchable PDF, drafting wants Word, a key-dates chart wants a spreadsheet, and comparison wants plain text. Convert once, then export as many formats as the work calls for.

Two notes on choosing.

First, the searchable PDF is the only output that still shows the signed pages. Schema Law's searchable PDF is your original file with a hidden text layer, so the signatures, initials and stamps remain visible while search works underneath. Every other format is text lifted off the page. The list of export formats says what each one holds.

Second, you can export more than one. A renewal can use three: the searchable PDF to look things up, Word to draft from, and plain text to run a comparison against the other side's draft when it comes back. For the Word copy in particular, the post on converting a signed agreement to Word explains what to do with the signature block and how to mark the file so it is never mistaken for the executed agreement.

Misread characters that change what a contract clause says

A handful of misreads change what a contract clause says while leaving it perfectly readable. A wrong digit in an amount, a lost decimal point in a percentage, a defined term that loses its capital, and a dropped "not" all produce a sentence that looks right and is wrong.

Ordinary OCR errors announce themselves. "Tbe Lessee sball" is obviously damaged, and you fix it or ignore it. The dangerous errors are the ones that yield another valid word or number.

Why these, in a contract

Contracts pack meaning into small marks. A 3 and an 8 differ by a gap of a few pixels in a faxed or photocopied page. A decimal point is a single dot, and a speck of toner can add one or a light scan can remove one. "Lessor" and "Lessee" differ by two letters at the end of a word, and pairs like that run through every agreement: licensor and licensee, assignor and assignee, mortgagor and mortgagee. How digits and decimal points get misread in general is covered in the post on why OCR gets numbers wrong.

Negatives are the worst case. "Not" is three letters, and it can fall at the end of a line or beside a hole punch. A line that is cut off at the margin of a skewed scan can lose it without leaving a trace.

A reading routine

Use the contract's own redundancy. Where an amount or a period is written in both words and numerals, compare the two in the OCR text. If they disagree, one was misread, or the signed page itself is inconsistent, and the image will tell you which.

Then read these against the image, in this order, in every converted agreement:

• Every currency amount and every percentage.
• Every period of days, months or years.
• Every date.
• Every section reference inside a clause you intend to rely on.
• Party names in the preamble and the signature block.
• The full sentence around every "not", "no", "except", "unless" and "other than" in the clauses that matter to your question.

Do not count on a converted file to show which characters the software was unsure of. Schema Law marks no passage as uncertain, so the reading routine is the control. Marks made by pen are a further category. Where a figure has been struck out and another written above it, the printed text and the deal differ, and the post on handwritten changes to a contract explains how to find and record those edits.

Why the OCR text of a contract is never the contract

The OCR text of a contract is a reading of the contract made by software. The parties signed the pages shown in the image, and where the text and the image differ, the image shows what they signed. Treat the text as a finding aid and a drafting aid, never as the agreement.

That status has practical consequences for how you work.

Quote from the text, cite from the image. Copy the clause out of the converted file to save typing, then read it once against the signed page before it goes into a letter, a memo or a pleading. Give the page of the scan as the reference.

Never replace the scan. A Word file treated as the contract invites argument from language nobody signed. Keep the scan as received, and store converted files next to it under names that say what they are.

Send the scan when someone asks for the contract. A client, an auditor or the other side who asks for a copy of the agreement should get the signed pages. The searchable PDF qualifies, since it shows them. A Word or text export does not.

Treat a mismatch as a finding about the text. If your chart says the term is five years and the image says three, the chart is wrong. Correct it, and then ask what else came from the same page.

Say so on the working copy. Put a line at the top of any converted file that names the scan it came from and the date it was converted. A reader who meets the file a year later should see at a glance that it is a derivative.

One question sits outside all of this: whether the scan itself is good enough when the paper original is gone. That depends on rules of evidence and on the kind of document, and it varies by jurisdiction. The post on the scanned copy of a signed contract sets out what to check.