Convert a Scanned PDF to Editable Word With OCR (Step by Step)

You drop contract-2019-signed.pdf into a PDF-to-Word converter, open the result, and every page is a single giant image floating in Word. Nothing is selectable. Nothing is editable. You did not do anything wrong: the file never contained text in the first place.

A scanned document is a photograph of paper stored in a PDF wrapper. Converting it to DOCX just moves the photograph into a new container. To get a scanned PDF to editable Word, you need one extra step before conversion: OCR, which reads the pixels and writes real characters underneath them. Get that order right and the rest is straightforward.

This guide covers how to tell whether your PDF needs OCR, the two practical routes (online and command line), which language packs to install, why the layout drifts and what to proofread before you send the file to anyone.

First check: does your PDF actually contain text?

Thirty seconds of checking saves an hour of confusion. Open the PDF in any viewer and try this:

• Try to select a word with your cursor. If you get a blue rectangle around the whole page instead of a text highlight, it is an image.
• Search for a word you can clearly see (Ctrl+F, type "Agreement"). Zero results on a page that obviously says Agreement means there is no text layer.
• Check the file size. A 12-page text PDF is usually 80–300 KB. A 12-page scan is often 4–20 MB, because each page is a JPEG or TIFF image.

Mixed files are common in legal work: a born-digital contract with three scanned signature pages appended. Those pages behave exactly like full scans and need the same treatment. If you want the background on how character recognition actually works and where it goes wrong, read our OCR basics article before you start on a big batch.

The correct order: OCR first, convert second

Almost every failed attempt comes from doing these in the wrong order, or skipping the first one entirely.

• Step 1 — OCR the PDF. A recognition engine analyses each page image, detects characters and writes an invisible text layer positioned exactly over the pixels. The page still looks identical; it is now searchable and copyable.
• Step 2 — Convert the searchable PDF to DOCX. The converter reads that text layer, groups it into paragraphs, tables and headings, and builds a Word document.
• Step 3 — Repair and proofread. Fix the handful of misread characters and rebuild anything the layout engine mangled.

Skip step 1 and step 2 has nothing to read, so it embeds the images and calls it a day. That is the "empty Word file" everyone complains about.

Route A: the online route (no installation)

This is the right choice for a handful of files, a shared office machine where you cannot install software, or a paralegal who needs one deposition transcript editable by lunch.

• Run your scan through an OCR step so the PDF gains a text layer. Confirm it worked by searching for a visible word in the output — if Ctrl+F finds it, you are ready.
• Feed the searchable PDF to our PDF to Word (DOCX) converter and download the result.
• Open the DOCX in Word and check the first and last page before anything else — those are where headers, footers and signature blocks go wrong most often.

If your source is a pile of phone photos or TIFFs rather than a PDF, combine them into one PDF first; the other tools on the document converter hub handle the image and office formats around this workflow.

One practical warning for legal and HR staff: confidential records have retention and disclosure rules. If the document contains client data, medical history or personal identifiers, check your firm's policy on uploading files before you convert anything, and prefer the local route below when in doubt.

Route B: the command line (batches and confidential files)

When you have 400 scanned personnel files in a folder, clicking through a web form is not a plan. Two free tools do the whole job: ocrmypdf for the text layer, soffice (LibreOffice) for the DOCX.

Step 1 — add the OCR layer:

ocrmypdf --deskew --rotate-pages --clean \
  -l eng \
  contract-2019-signed.pdf contract-2019-ocr.pdf

What those flags buy you: --deskew straightens pages fed in crooked, --rotate-pages fixes sideways scans, --clean removes speckle noise from old photocopies. On a typical office machine a 12-page scan takes well under a minute. Your 9 MB input often comes back around 9–10 MB; the text layer itself is tiny, add --optimize 3 if size matters.

Step 2 — convert to Word:

soffice --headless --convert-to docx:"writer_pdf_import" \
  --outdir ./word-out contract-2019-ocr.pdf

Step 3 — do the whole folder:

for f in scans/*.pdf; do
  ocrmypdf --deskew --rotate-pages -l eng "$f" "ocr/$(basename "$f")"
done
soffice --headless --convert-to docx --outdir word-out ocr/*.pdf

Two flags worth knowing: --skip-text leaves pages that already have text alone (perfect for those mixed contracts), and --redo-ocr re-reads pages whose existing text layer is garbage from an old scanner utility.

Language packs: the single biggest accuracy lever

OCR engines use a language model to decide between visually similar shapes. Run a German contract through the English model and für becomes fiir, Straße becomes StraBe, and every umlaut turns into noise. Install the right pack and the same page comes out clean.

On Debian or Ubuntu, language data for Tesseract installs as separate packages:

sudo apt install tesseract-ocr-deu tesseract-ocr-fra tesseract-ocr-spa
tesseract --list-langs

Then pass them to the OCR step. Bilingual documents take several codes joined by +:

ocrmypdf -l eng+deu lease-bilingual.pdf lease-ocr.pdf

Do not list six languages "just in case" — each extra model slows recognition and gives the engine more wrong answers to choose from. Two is usually the sensible maximum. Online tools expose the same idea as a language dropdown; set it before you upload rather than accepting the default.

Scan quality matters almost as much. 300 DPI, black and white or greyscale, flat on the glass is the sweet spot. 150 DPI faxes produce broken characters no engine can rescue; 600 DPI mostly produces bigger files. If a document is genuinely unreadable at 300 DPI, rescan before you blame the software.

Why the layout changes, and what to fix

PDF stores positions; Word stores flowing paragraphs. The converter has to guess which pixels form a paragraph, which form a table and which are a header. On a clean scan it guesses well. On a 1998 photocopy with a stamp across the margin, it guesses creatively. Expect these:

• Text boxes instead of paragraphs. Click a line; if it sits inside a frame you can drag, select all, cut, and paste as unformatted text into a clean document, then re-apply styles.
• Tables that became tab-separated lines, or the reverse: ordinary text wrapped in an invisible table. Ruled tables with visible borders survive far better than whitespace-aligned columns.
• Line breaks at the end of every visual line, which makes editing a paragraph miserable. Find and replace is faster than manual cleanup.
• Headers, footers and page numbers landing in the body text on every page.
• Signatures, stamps and handwriting staying as images — correct behaviour, since OCR does not read cursive reliably.

The practical takeaway: if you need perfect formatting, budget time to rebuild it. If you need the words — to search, quote, redact or paste into a template — you already have everything after step 2.

Proofread like the document depends on it

OCR does not misspell words the way humans do; it produces plausible-looking wrong characters that spellcheck happily ignores. In a contract or a medical record, that is the risk that matters.

• Digits and letters: 0/O, 1/l/I, 5/S, 8/B. Check every amount, date, case number, invoice number and postcode by eye against the scan.
• rn vs m — modern becomes modem, turn becomes tum. Spellcheck never flags it.
• Decimal points and commas in figures: 1.000,00 and 1,000.00 are one speck of dust apart.
• Negations. A dropped "not" changes a clause completely; read obligations and exclusions slowly.
• Names and addresses, which no dictionary can validate for you.

A fast review method: put the scan and the DOCX side by side, turn on Word's whitespace marks, and read the numbers first, then the clause headings, then the body. For records with legal weight, keep the original scanned PDF archived alongside the Word file — the scan stays the authoritative copy, and the DOCX is the working one.

A workflow you can repeat tomorrow

1. Try to select text in the PDF. No selection means it needs OCR; do not convert yet.
2. Scan or rescan at 300 DPI greyscale if the source is poor.
3. Add the OCR layer with the correct language pack (ocrmypdf -l eng, or the dropdown in your online tool).
4. Verify with Ctrl+F that a visible word is now findable.
5. Convert the searchable PDF to DOCX with our PDF to Word converter or soffice --headless --convert-to docx.
6. Rebuild tables and headings in Word; paste as unformatted text when frames get in the way.
7. Proofread digits, names and negations against the original, then archive the scan.
8. For volume, script steps 3 and 5 over a folder and spot-check 10% of the output.

The rule to remember: no converter turns a scanned PDF into editable Word without OCR, because there is nothing to convert. Add the text layer first and the rest of the job becomes ordinary editing.

FAQ

Why is my converted Word file just a picture? The PDF had no text layer, so the converter embedded the page images. Run OCR first, then convert again.

Can Word itself OCR a scanned PDF? Word's "Open PDF" feature reconstructs documents that already contain text. For image-only scans it gives you pictures. Do the OCR step outside Word.

How accurate is OCR on a typical office scan? A clean 300 DPI typed page with the right language pack is usually very accurate — but "very accurate" still means a few wrong characters per page, which is why proofreading digits is non-negotiable. Faxes, carbon copies and tight dot-matrix print are much worse.

Will OCR read handwriting or signatures? No, not reliably. Printed text yes; cursive, no. Signatures remain images in the Word file, which is normally what you want.

My PDF is half typed text, half scanned pages. What then? Use ocrmypdf --skip-text, which leaves existing text alone and only processes the image pages.

Should I convert to DOCX or keep a searchable PDF? If you only need to search, quote or archive, stop at the searchable PDF — it preserves the layout perfectly. Convert to DOCX only when someone has to edit the wording.

Does OCR change how the page looks? No. The recognised text is placed invisibly behind the original image, so the page prints exactly as before.