How to Quote Text From a Scanned PDF Accurately

To quote text from a scanned PDF, convert the page with OCR, copy the passage from the plain text export, and paste it into your draft without formatting. Then read the quotation against the scanned page image, character by character, and take the page reference from the original document, not from the converted file.

Retyping a paragraph from a scan is slow and introduces its own mistakes. Copying it from OCR output is fast and introduces different ones. The method here uses the software for speed and your eyes for accuracy, and it puts the checking at the one moment when it is cheap: before the quotation goes into a draft that other people will copy from.

Why is quoting from a scan risky?

Quoting from a scan is risky because quotation marks are a promise that the words are exact, and OCR output is an estimate. A recognition error inside a quotation looks deliberate. It may change the meaning, and it will certainly be noticed by the one reader who has the original open.

In ordinary use a few wrong characters on a page do no harm. You can still find the document, read it and understand it. A quotation is different in three ways.

• The tolerance is zero. A summary can be roughly right. A quotation is either what the document says or it is not.
• The errors are plausible. OCR rarely produces nonsense on a decent scan. It produces "may" for "must" on a smudged line, "2012" for "2017", "$15,000" for "$75,000". Each reads naturally in context, so your eye passes over it.
• The error travels. A passage quoted in a memo is pasted into a letter, then a brief, then a reply. Nobody goes back to the scan after the first use.

There is a subtler risk as well. A quotation must reproduce the source's own mistakes: its misspellings, its odd capitalization, its missing comma. Recognition software can be more helpful than you want here, returning the dictionary spelling of a word the document spelled wrongly. The post on AI OCR for lawyers explains why newer systems are prone to this. For quoting, the lesson is that a passage with no visible errors has not been verified by that fact.

None of this is a reason to retype. A person copying by hand skips lines and transposes digits too. It is a reason to verify.

A workflow for lifting a quotation from a scanned page

The workflow has six steps: find the passage, convert only the pages you need, copy from plain text, paste unformatted, verify against the image, and record the source. The first four take a couple of minutes. The fifth is the one that matters.

1. Find the passage in the scan and note its location. Write down the printed page number, the Bates number if there is one, and the section or line. Do this from the scan, before any conversion.
2. Convert the pages. Extract the page with the passage plus one page either side, and upload that short PDF. Schema Law's free plan takes a document of up to three pages a day and includes the plain text export, which fits this job exactly. For longer documents the pricing page shows the page limits on each plan.
3. Copy from the TXT export. Open the text file, find the passage and copy it. Plain text carries no font, no style and no hidden line endings from the scan's layout. The post on scanned PDF to text describes what else the file contains.
4. Paste as unformatted text. Use your word processor's paste-without-formatting option, then apply your block quote or inline quote style.
5. Verify against the page image. Compare the pasted passage with the scan phrase by phrase. Do it now, not at cite-check.
6. Record what you did. Add a comment or footnote draft with the pin cite and a note that the quotation was checked against the image, with the date and your initials.

Then restore what plain text removed. Italics, underlining and bold in the original are not in a text file. Put them back from the scan, so that you can say truthfully whether emphasis is original or added.

Handle alterations after verification, not before. First get the quotation exact. Then make your ellipses, bracketed substitutions and capitalization changes according to the citation manual your court or office follows. If you alter first, you no longer have an exact text to compare with the page.

Once the words match the image, drop the passage into a pleading or a memorandum of law. Put a pinpoint citation beside it from the page and line on the scan, using Bluebook citation form or your office form. The same block can move into an appellate brief. Add it to the table of authorities only after the pin cite matches the image.

A common mistake is to copy from the on-screen view of a searchable PDF instead. It works, but selections in a PDF can pick up text out of order or drag in a header, and the pasted text may bring formatting with it. If a searchable PDF is all you have, paste into a plain text editor first, then copy from there.

How do you verify a quotation against the page image?

Put the page image and your pasted quotation side by side and compare them in small units, pointing at each word. Read for characters, not meaning. Then check the items most likely to be wrong a second time: numbers, names, negatives and punctuation.

Reading for meaning is what makes proofreading fail. Your brain supplies the word that ought to be there. These techniques interrupt that habit.

• Zoom the scan. Set the page image to at least 150% so that commas and periods can be told apart.
• Go phrase by phrase. Compare three or four words at a time. Move a ruler, a sheet of paper or the edge of a window down the scan so you cannot lose the line.
• Read it backwards for the hard parts. For a string of figures or a citation, compare from the last character to the first. Meaning cannot help you, so you see what is there.
• Read aloud with a second person for anything long. One reads from the scan, including punctuation and capitals ("capital T, The Tenant shall comma"), while the other follows the draft. It is slow, and for a block quotation in a dispositive brief it is worth it.
• Count. Count the sentences in the passage on the scan and in the draft. A dropped line usually takes a whole clause with it and leaves grammatical text behind.

Then run this checklist over the quotation.

If you cannot read a word on the scan yourself, the software could not either, whatever it output. Get a better copy, or quote around the illegible word and say so. Do not let a guess stand inside quotation marks.

The broader method for proofreading recognized text, for whole documents and not single passages, is in the post on why OCR output is full of errors.

Characters OCR most often gets wrong in quoted passages

The characters OCR confuses most are the ones that look alike at low resolution: 1, l and I; 0 and O; 5 and S; rn and m; and the small punctuation marks. In a quotation these matter most inside numbers, citations, names and short function words.

Use this table as a list of things to look at twice.

Three patterns do not fit in a table.

Missing small words. A fleck or a crease can erase "not", "no" or "or". The sentence still parses. Compare the word count of each sentence when the passage contains a negative.

Words fused or split. "the Tenant" becomes "theTenant", or "herein" becomes "here in". Spell-check flags the first and passes the second.

Marks that are not text. A stamp, an underline drawn by hand, a hole punch or a fax header can be read as characters and inserted mid-sentence. The reverse also happens: text beneath a stamp goes missing. The post on whether OCR can read stamps over text covers that case.

Spell-check is of limited help. It catches "dause". It does not catch "modem", "may" for "must" or a wrong digit. Treat a clean spell-check as no evidence at all about a quotation.

Pin cites, page numbers and line numbers after conversion

Take every pin cite from the scanned original: the page number printed on the page, the Bates number stamped on it, or the page and line of the transcript. The converted file has its own page positions, and they usually do not match the numbers your reader will use to find the passage.

A converted file can mislead you in several ways.

• File page versus printed page. A PDF's fifth page may be printed page 3, because of a cover sheet and a table of contents, or page 212, because the file is an excerpt. Exports follow the file. In a TXT export from Schema Law, pages are separated by a page break character, and in the Word export each page of the scan starts a new page. Both tell you which page of the uploaded file you are on, and neither tells you the printed number.
• Reflowed pages. Once you edit a Word conversion, delete a header or change a font, text moves between pages. Page numbers in an edited conversion mean nothing.
• Bates numbers. A Bates stamp is text printed on the image, and it is usually recognized along with everything else, with the usual risk of a misread digit. Read the number off the image. The post on how to OCR Bates numbered documents goes into keeping pages tied to their numbers.
• Line numbers. Transcripts and pleading paper print line numbers down the margin. After conversion those numbers may sit at the start of each line, appear in a block of their own, or be missing. You cannot count lines in the converted text and get the right answer.
• Internal numbering. Section and paragraph numbers are part of the text and normally survive, but a misread "1.1" for "11" is easy to miss. Confirm the section number on the image.

So the routine is: locate with the converted text, cite from the image. Search the text file to find the passage, use its page position to open the right page of the scan, then write down the printed page, the Bates number and the line range from what you see there.

Courts and citation manuals differ on how record materials are cited and whether parallel references are needed. Check the rules that govern your filing. Whatever the form, the reference has to lead a reader with the original to the words.

Quoting from transcripts versus contracts

Transcripts and contracts fail in different places. A transcript's risk is its layout: line numbers, speaker labels and four-page condensed sheets that scramble reading order. A contract's risk is its content: defined terms, figures, cross-references and long sentences where one dropped word reverses an obligation.

Transcripts

A transcript page has a fixed number of numbered lines, and citations are to page and line. When you quote testimony:

• Check that the question and the answer are both attributed correctly. "Q." and "A." labels are short and sit at the margin, where they are easily dropped or attached to the wrong line.
• Watch for line numbers embedded in the copied text. A stray "14" in the middle of an answer is a line number, not testimony.
• Take the page and line range from the image, for the first and the last line you quote.
• Be careful with condensed transcripts that print four pages to a sheet. Text from neighboring panels can be interleaved, so confirm that every line you copied came from the same panel.
• Preserve the reporter's rendering exactly, including dashes for interruptions, "(indicating)" and "[sic]" where the reporter put it.

The post on making deposition transcripts searchable deals with page and line fidelity across a whole transcript.

Contracts

A contract quotation usually has to carry its context with it.

• Capitalized defined terms must stay capitalized. "the Premises" and "the premises" are different things in a lease.
• Quote the section number and heading from the image, and check any cross-reference inside the passage.
• Read provisos and exceptions to the end of the sentence. Contract sentences are long, and the operative "provided, however" may be on the next page of the scan.
• Check whether the passage was amended by hand. Interlineations and initialed strike-outs are part of the executed text and are the parts recognition handles worst.
• Confirm you are quoting the executed version. A scan of a signed copy and a Word draft with the same title may differ.

For both kinds of document the last step is the same. Keep the scan, the converted text and your verification note together in the matter file, so that when someone asks where a quotation came from, the answer takes a minute to give. If you are choosing an export for a different purpose, the format list on the homepage shows what each one is meant for.