Does OCR Change the Original PDF? What to Keep and Prove
By Schema LawPublished Updated
Whether OCR changes the original PDF depends on the tool. Some programs write the recognized text into the file you opened and save over it. Others hand back a new file and leave yours alone. Either way, run OCR on a copy, keep the original PDF untouched, and compare pages so you can show the two match.
That answer has two halves, and they need different proof. One half is about the file: is the thing you received still sitting on your drive exactly as it arrived? The other is about the pages: does the searchable copy show the same pictures as the file it was made from? A folder habit and a hash value settle the first. Only looking at pages settles the second.
When you compare pages, search the copy for the plaintiff and the defendant in the caption, then for the docket number on that sheet. Open each hit and match it to the line on the picture. Do the same with a Bates number farther in. A complaint title that the search places on a different page means the pictures moved.
Does OCR change the original PDF or make a new file?
OCR changes the original PDF only when the program saves its result over the file you opened. A tool that takes an upload and gives you a download makes a new file and cannot reach the one on your drive. You can tell which kind you have from the file's modified date and size after the run.
Two ways a tool hands back the result
Recognition inside an open document. A desktop PDF editor opens your file, recognizes the pages, and holds the result in memory. Nothing on disk has changed at that moment. The original is replaced when you press Save, or when the program saves for you on closing. Save As under a new name leaves the first file alone.
Recognition that produces a separate output. A browser service, a batch converter with an output folder, and a scanner utility that writes a second file all work this way. The input is read and a new file is written somewhere else.
Schema Law is of the second kind. You upload a PDF, PNG or JPEG from your browser and download whichever export you choose. The export formats section of the homepage describes the Searchable PDF export as "Your original file with a hidden text layer, so search finds the page". That download arrives as its own file, and the PDF you uploaded stays on your computer as it was.
A test on a file that does not matter
If you are unsure how your own program behaves, find out with a throwaway document before a client file is involved.
- Copy any scanned PDF into an empty folder and give it an obvious name, such as
test_before.pdf. - Write down its size in bytes and its modified date and time from the file properties.
- Run recognition the way you normally would, and close the program, accepting whatever it asks on the way out.
- Look at the folder again.
| What the folder shows afterwards | What happened |
|---|---|
| One file, same name, new modified time, different size | The program wrote its result over the input |
| Two files, the first with its old time and size | The program wrote a separate output |
| One file, old time and size, and no second file anywhere | Nothing was saved; the result was discarded on closing |
| One file with a new time but the same size to the byte | The file was rewritten; open it and check whether text was added at all |
The third row deserves a second look, because it looks the next morning as if OCR "did not work". The post on a PDF that still will not search after OCR ran deals with that case.
Whatever the test shows, the safe habit does not change: make the copy first, and point the program at the copy.
What OCR can alter in the page picture: resolution, compression and rotation
Reading a page does not require changing it, yet a conversion can alter the page picture when the program is set to shrink, clean or straighten the scan while it works. Resolution can drop, compression can get harsher, and pages can be rotated, straightened or cropped. Each change gives you a page that differs from the one you received.
| Change | What the program did | What you see on the page | Why it matters in a legal file |
|---|---|---|---|
| Lower resolution | Resampled the picture to fewer dots per inch | Small type and fine lines turn soft when you zoom in | Footnotes, fax headers and faint stamps become harder to read |
| Harsher compression | Re-saved the picture with a method that discards detail | Blocky halos around letters, or gray tones forced to pure black and white | Pencil notes, light signatures and highlighter marks can vanish |
| Color removed | Converted color pages to gray or black and white | A blue signature and a red stamp turn the same black as the print | You can no longer tell an ink original from a copy of a copy |
| Rotation | Turned a sideways page upright | The page reads correctly on screen | Harmless for reading, but the page no longer matches how it was produced |
| Straightening | Tilted the picture a degree or two to level the lines | Text sits level; edges show thin white or black wedges | Content near the margin can be pushed off the page |
| Cropping or border removal | Trimmed dark scanner edges | A cleaner page | Punch holes, edge stamps and page numbers near the border can be cut |
| Rebuilt page | Replaced the scanned letters with fresh type | Crisp text that was never on the paper | A misread character now shows on the visible page |
None of these is wrong in itself. A reading copy for your own desk may be better for being straightened and smaller. The point is to know which of them your program applied, because a file note that says "same pages, text added" is false if the pages were recompressed.
The post on conversion options explains where these controls sit and what each one trades away, under image downsampling and compression choices during conversion. If you want the short rule for leaving the picture alone, the guide to making a scanned PDF searchable has it under keeping the page image identical to the original.
A quick sign that the pictures were touched
Compare the two file sizes. Added text alone makes a copy a little larger, so a copy that is half the size of the scan had something done to its pictures. The post on searchable PDF file size explains what can make a copy smaller.
Size is a hint and nothing more. A copy can be the same size and still have been rotated or cropped, so treat a surprising size as a reason to look at pages.
How do you prove the searchable copy shows the same pages as the original?
You prove the searchable copy shows the same pages by comparing it with the original page by page and writing down what you compared. No property of the file does this for you. Confirm the page count, then the page sizes, then put chosen pages side by side at high zoom, and record the pages you looked at.
The comparison, in order
- Count. Open both files and read the page count of each. A difference ends the exercise: find the missing or extra page before anything else.
- Check first, last and one middle page. Confirm that the same content is on the same page number in both files. This catches a dropped blank page that has shifted everything after it.
- Check page size and direction. Look at the page dimensions your reader reports for two or three pages in each file. A letter page that has become slightly smaller was cropped. A page that is now upright was rotated.
- Put the hard pages side by side. Open the original and the copy in two windows at 200% or more. Choose pages with the detail most likely to suffer: a signature in colored ink, a stamp, a handwritten note, a faint fax line, small print at the foot of the page.
- Look at the edges. Scroll to all four margins of two pages. Edge stamps, punch holes and page numbers are where cropping shows.
- Write it down. Record the files, the pages compared, what you saw, the date and your initials.
How many pages to compare
That is a judgment about how the copy will be used, and it is yours to make for the matter. A reading copy for your own notes may need the count and a handful of pages. A copy that will be shown to a witness, filed or handed to the other side may justify every page. If one compared page shows a difference, assume the setting that caused it applied to the whole file and rerun the conversion with that setting off.
A worked record
Suppose a firm receives a 212-page scanned ledger in the fictional matter Estate of Calloway and makes a searchable copy for the team. The note in the matter file might read:
| Field | Entry |
|---|---|
| Original | calloway_ledger_as_received.pdf, 212 pages |
| Searchable copy | calloway_ledger_ocr.pdf, 212 pages |
| Pages compared at 200% | 1, 2, 57, 106, 107, 180, 212 |
| Why those pages | 57 has a blue signature; 106 and 107 have pencil totals; 180 is a faint fax page |
| Result | Pictures match on all seven pages; margins intact; no rotation |
| Text checked | Not proofread; searched "Calloway" and "4,210.00" and found both |
| Compared by, date | R. Imai, 14 March |
A note like that is short, and it answers the question a colleague or an opponent is likely to ask a year later: who looked, at what, and what did they see?
Hash values for the original scan: what they show and what they cannot
A hash value shows that a file is byte for byte the file it was when the value was recorded. It cannot show that a searchable copy matches the original, because adding text changes the bytes. Use the hash to protect the original, and use your page comparison to vouch for the copy.
A hash is a short string computed from the entire contents of a file. Microsoft's documentation for the Get-FileHash command in PowerShell puts the two properties that matter this way: "changing even a single character in the contents of a file changes the hash value of the file", and "If the hash values of two files are identical, the contents of the files are also identical." The same page says the name of a file can change without changing its hash value.
On a Windows computer the command is one line:
Get-FileHash .\calloway_ledger_as_received.pdfThe documentation states that the command uses the SHA256 algorithm by default. Record the algorithm with the value, since a value means nothing without it.
What the hash lets you say
- The original has not changed since the day you recorded it. Compute the value again at any later date. The same value means the same file.
- Two files are the same file under different names. A copy in the archive and a copy on a colleague's laptop with matching values are identical.
- The file you hold is the one the sender described. If a delivery comes with a list of hash values, matching yours against the list confirms nothing was damaged or swapped on the way.
What the hash cannot say
- That the searchable copy shows the same pages. The copy contains text the original lacks, so its hash differs by design. A mismatch between original and copy tells you nothing about the pictures.
- How big the difference is. Two values are either equal or not. One added word and a wholesale recompression look the same: different.
- That the original is authentic. A hash fixes the file as you received it. It says nothing about what happened to the document before it reached you.
- Anything, if nobody wrote it down at the time. A value computed today proves the state of the file today.
You can also record a hash for the searchable copy on the day you compare it. That ties your comparison note to one exact file, so a later rerun of OCR cannot be confused with the copy you checked.
For a production from an opposing party, the routine for the day the files arrive is set out in the post on making a production searchable, under preserving the production as received. Whether hash values carry any evidentiary weight in your matter is a question for the lawyers on it and the rules of your court.
Where to keep the original PDF once a searchable copy exists
Keep the original PDF in a folder of its own that nobody works in, under the name it arrived with, and keep the searchable copy beside it in a working folder under a name that marks it as a copy. The original should be reachable in one step from the copy, and impossible to mistake for it.
A layout that works for a single matter:
| Folder | Holds | Who writes to it |
|---|---|---|
01_as_received | Files exactly as delivered, with the email or letter that came with them | One person, on the day of receipt, then nobody |
02_searchable | OCR copies, one for each original | Whoever runs the conversion |
03_work | Extracts, annotated copies, Word and Excel exports | The team |
00_log | The hash list and the comparison notes | Whoever maintains the file |
Naming
- Leave the original's file name alone, however unhelpful it is. A name such as
scan0042.pdfis part of how the file arrived. - Give the copy the same base name plus a suffix:
scan0042_ocr.pdf. Anyone sorting the folder by name sees the pair together. - Put the description in the log and not in the original's name: "scan0042.pdf is the Calloway ledger, 212 pages, received by email".
- If you rerun OCR, add a date or a version to the suffix so the copy you compared stays identifiable.
Do not let the converter be the archive
A conversion service is a place where work happens. It should not be the only home of the original. Every Schema Law plan lets you delete any document at any time, as the pricing page sets out. Your own file system or document management system holds the record. If your firm's document system keeps versions, check whether saving the searchable copy as a new version of the original hides the first version from casual view. Two separate documents are easier to explain than one document with a history.
A closed file raises the retention question again. Whether the original, the copy or both must be kept, and for how long, depends on your firm's policy and the rules that govern your practice.
Describing an OCR'd copy accurately when you send it to someone else
Describe an OCR'd copy as what it is: a copy of the document with machine-recognized text added. Say whether the page pictures were changed, say that the text has not been proofread unless it has, and say where the original is. A sentence or two in the cover note does it.
The recipient may search the file, copy from it, or load it into a system of their own. Each of those uses relies on the hidden text, and they cannot see from the page whether it is good. Telling them what you did lets them decide how far to trust it.
What a cover note should let the reader answer
- Is this the file as you received it, or a copy you processed?
- Were the page pictures altered in any way, and if so how?
- Was the recognized text checked, and against what?
- Can they have the unprocessed file if they ask?
Wording you can adapt
For a colleague or co-counsel:
Attached is a searchable copy of the Calloway ledger. I ran OCR on a copy of the scan we received; the page images are unchanged and I compared seven pages against the original. The hidden text has not been proofread, so check any figure against the page before you quote it. The file as received is in 01_as_received.
For a client:
This PDF is the same scan you sent us with searchable text added so you can find names and amounts. The text was produced by software and may contain misreadings. The pages themselves are the pictures you provided.
Where the pictures did change:
This copy was reduced in size for email. The pages were recompressed and are less sharp than the original scan. Please ask if you need the full-quality file.
What to leave out
Do not call the copy "the original", and do not call it "identical" unless a hash comparison says so, which it will not once text has been added. "Same pages, text added" is accurate. "Certified", "verified" and "authenticated" carry meanings in some settings that a spot check does not support.
When the recipient is a court, an agency or an opposing party, the description may be governed by a rule, an order or an agreed protocol. Read that first and follow its wording over anything suggested here. The post on searchable PDFs for court filing covers what courts ask of a filed scan, and the requirements differ from court to court.