OCR PDF to Word: The Choices That Change the Result

When you OCR a PDF to Word, four decisions change the file you get: whether the Word output flows or copies each page, which language the recognition expects, which pages you convert, and what happens to headers, footers and page numbers. Settle them first, then test two pages before the whole PDF.

None of the four needs special software. Each is a question about what the Word file is for, and you can answer every one with the scanned PDF open and a pen in your hand. Converters differ in which of these choices they expose as settings, so part of the work is finding out what yours lets you control and what it decides for you.

Which decisions shape an OCR PDF to Word conversion before it starts?

Four decisions shape an OCR PDF to Word conversion: the layout of the output, the recognition language, the page range, and the treatment of running headers, footers and page numbers. A fifth step, a two-page trial, tells you whether you chose well. Each one changes how long cleanup takes afterwards.

The table sets the four side by side. Read the last column against your own task.

What your converter decides for you

Not every converter offers every choice. Before you upload, look for three things in the tool's own help pages or settings screen: an output or layout option, a language option, and a page-range option. Write down which exist. Where a choice is missing, the converter has made it already, and you find out which way by running a test.

A converter's own description of its output is the other place to look. Schema Law's homepage, for example, describes its Word export as editable text with its headings, lists and tables. The list of export formats shows what each of the other formats carries.

What the decisions do not cover

Two neighbouring questions sit outside this list. Tables have their own failure modes in a Word file, and the post on keeping tables as tables in a converted Word file deals with them. The mechanics of uploading and downloading are the subject of the walkthrough of an online OCR PDF to Word conversion in the browser.

Flowing text or a page-for-page layout in the Word output

Choose flowing text when anyone will edit the Word file, and a page-for-page layout only when the file will be read and never changed. Flowing text rewraps as you type. A page copy holds every line where it sat on paper, and it resists editing for the same reason.

The two outputs come from two different goals. A flowing conversion asks what each block of the page is, a heading or a paragraph or a list, and writes it as ordinary Word content. A page-for-page conversion asks where each block sits and pins it there. The post on scanning text into Word explains why an exact-layout conversion fills a Word file with text boxes and how to rescue the words from one.

Matching the layout to the job

The last two rows deserve a second look. When appearance is what matters, a Word file is a poor container, because the same file can paginate differently on another computer with other fonts installed. A PDF does not. If what you want is the look of the page plus text you can search, ask for a searchable PDF and leave Word out of it.

A worked example

Suppose you hold a scanned 22-page supply agreement between Marlowe Tool Co. and Pendry Fabrication, and the client wants the payment clause and the term clause amended. You will type into two clauses and renumber what follows. That is editing, so flowing text is the right choice, and the loss of the original fonts and spacing costs you nothing, since the amended agreement will go out in your firm's template anyway.

Now suppose the same client sends a one-page scanned certificate that must be reissued with a new date and otherwise look identical. A flowing conversion gives you the words and none of the look. Here the quicker path may be to retype the certificate into a template that already matches, using the conversion only as a source for the wording.

If you cannot choose

When the converter offers no layout option, run one page and click into the middle of a paragraph. Type a sentence. If the lines that follow move down and rewrap, the output flows. If your sentence spills out of a frame or hides behind the next block, you have a page copy.

Does the recognition language matter for an English legal document?

The recognition language matters for an English legal document wherever the page leaves plain English: accented names, section and paragraph signs, currency symbols, and foreign words and phrases. Plain English prose uses none of these. They sit in citations, names and figures, where one wrong character does real harm.

Recognition software has to decide which character each printed shape is, and some converters let you tell them which languages to expect. Whether yours has that setting, and what the setting changes, is a question for its documentation. The post on whether OCR uses a dictionary covers how word lists can alter what is recognized.

Where an English document stops being plain English

Go through the scan and mark each of these before you convert. They are the characters to check afterwards, whatever the tool.

• Accented names. A party named Zoë Albrecht-Muñoz, or a company registered as Société Générale de Transport, carries letters that an English alphabet does not. A dropped accent changes the spelling of a name in the record.
• Section and paragraph signs. Statutory citations lean on § and ¶. Read as "S", "5" or "8", a section sign turns "§ 12" into a number that looks plausible and is wrong.
• Currency symbols. £, € and ¥ sit directly in front of the figures that matter most.
• Latin and other borrowed phrases. Inter alia, pari passu and force majeure are spelled as their own languages spell them. Software that tidies unfamiliar words toward English can alter them.
• Foreign place names and addresses. A registered office in Zürich or Kraków appears in the notices clause, where a wrong character matters to anyone who posts a letter.
• Whole passages in another language. A bilingual contract or a certified translation bound with its original is a different job from an English page with one accented name.

Deciding what to set

If the converter has a language option and the document is English with scattered foreign names, read the documentation to see whether you may select more than one language, and test both ways on the page that holds the names. If the document has whole pages in another language, split those pages out and convert them as their own file with that language selected. If there is no language option at all, convert, then search the Word file for each marked item.

A search plan helps here. Before converting, write a short list from your marked pages: every accented name, every § citation, every figure with a currency symbol. After converting, search for each one as it should appear. A search that finds nothing tells you the characters came through differently, and the scan tells you how they should read.

Converting a page range to Word instead of the whole PDF

Convert a page range instead of the whole PDF when you will work on only part of the document, when the PDF holds several documents, or when a plan limits pages. Fewer pages means a shorter Word file, a shorter proofread and fewer chances for a stray error to pass unnoticed.

A scanned PDF can be a bundle: a cover letter, an agreement, three exhibits and a fax header sheet, all in one file because they went through the feeder together. Converting the bundle produces one long Word document in which the agreement begins on page four. Converting the agreement alone produces the file you wanted.

When you convert only the agreement, the Word file is the working copy of that contract. Check the indemnification clause and any third-party beneficiary sentence it points to. If a delegation of duties sits on the next page, include that page in the range so the draft does not drop the end of the clause.

When a partial conversion is the better job

1. You need one clause. To quote or amend the indemnity clause, convert the pages that hold it plus the definitions page it relies on.
2. The PDF holds more than one document. Convert each document you intend to edit as its own file, so each Word document starts on its own first page.
3. Some pages have nothing to recognize. Photographs, plans, blank backs and separator sheets add pages to the count and nothing to the text.
4. Some pages need different handling. A schedule of figures may belong in a spreadsheet, and a handwritten rider needs a closer check than print does.
5. The plan counts pages. Every page converted comes out of an allowance, on a free plan or a paid one. The pricing page shows the page limits for each Schema Law plan.

When the whole PDF is right

Convert everything when the document is one continuous instrument that you will redraft from start to finish, when cross-references run throughout, or when you want to search the whole text for a term before deciding where to work. A partial conversion of an agreement that defines its terms on page 2 and uses them on page 19 leaves you checking definitions against a file that does not contain them.

Doing it cleanly

If the converter accepts a page range, enter it and note the range in the file name. If it does not, extract the pages into a new PDF with your PDF program first, then upload the extract. The post on free conversions gives the steps for cutting a scanned PDF down to a page cap. Either way, write down which pages of the source the Word file represents. Marlowe supply agreement pp 4-25 of scan (OCR).docx answers the question a colleague will otherwise ask in six months.

Keep the page arithmetic visible too. Page 1 of an extract that began at page 4 of the scan is page 4 to anyone holding the paper, and a citation taken from the Word file's own page count will be off by three.

Running headers, footers and page numbers: keep them in the Word file or strip them?

Strip running headers, footers and page numbers when you will redraft the document, and keep them when you will cite the original by page. On a scan they are ordinary printed text near the page edge, so a converter can deliver them as lines of body text in the middle of your sentences.

The post on whether OCR keeps formatting explains what happens to footnotes, headers and page numbers in conversion. The decision here is narrower: once you know they can arrive in the body, do you want them there?

What each choice costs

A rule you can apply

Sort page furniture into two kinds before deciding.

Furniture that is information. A Bates number, an exhibit sticker, a "Page 7 of 34" on a document whose completeness is in question, or a fax transmission line showing a date and sender. Keep these, or record them somewhere before stripping, because they say something the body text does not.

Furniture that is decoration. A firm name repeated at the foot of each page, a document reference code, a draft stamp from a superseded version. In a working draft these only interrupt.

Stripping without losing track

If the converter has an option to leave headers and footers out, use it for a redraft. If it does not, remove them in Word. A repeated header is identical on every page, so find and replace clears it in one pass. Page numbers vary, so search for the fixed part, such as "Page " or " of 34", and delete each hit after looking at it. Do this before you rejoin paragraphs that were split at a page end, while the furniture still marks where each page stopped.

When you need the original page boundaries later, a bracketed marker is a tidy middle course: replace each stripped page number with [p. 14] in the running text. The marker reads as an editorial note, survives editing, and can be removed in one search when the draft is final.

Trying two pages before committing a long PDF to Word

Try two pages before converting a long PDF: the plainest page of body text and the hardest page in the file. Two pages show how the output flows, how symbols and names come through, and where page furniture lands. Changing a setting after a two-page trial means rerunning two pages, not the whole document.

Picking the two pages

The plain page is your baseline. If ordinary paragraphs do not come through cleanly, nothing else will, and the scan itself may need attention. The hard page is whichever one holds the most of what worries you: the accented names, the § citations, a table, a header crowded against the first line of text, a faint fax line. In an agreement, the notices clause and the first page of a schedule are good candidates.

What to look at in the trial output

• Click into a paragraph and type. Does the text under it move?
• Search for one accented name and one § citation from your list.
• Find the page header and footer. Are they in the body, in Word's header area, or absent?
• Read the last line of the first trial page and the first line of the second. Is a sentence cut in two?
• Look at any table. Is it a Word table, loose text, or a picture?
• Count the paragraphs on the plain page against the scan.

Acting on what you see

Change one thing at a time. If you alter the layout option and the language option together and the result improves, you will not know which did it, and the next document will send you back to guessing. Run the two pages, change one choice, run them again, and compare.

Write the outcome down where the next person can find it. A three-line note in the matter file is enough: which converter, which choices, which two pages you tested and on what date. Then convert the full range, and proofread the result against the scan as you would any typed copy, because a good trial lowers the number of errors and never removes them. Recognition is strongest on clean print, and no converter can promise that page 19 behaves like page 2.