The 10-second test: do you actually need OCR?
Open your PDF and try to drag-select a sentence. Then press Ctrl+F and search for a word visible on the page.
This test takes ten seconds and prevents the most common failure in this space: running a scan through a non-OCR converter and wondering why the output is empty.
- Text selects and search finds it: you have a real text layer. You don't need OCR — use a standard converter (see our free PDF-to-Word guide: https://all-done.xyz/blog/pdf-to-word-free-no-signup-limits-tested).
- Nothing selects, search finds nothing: it's image-only. You need OCR. Everything below is for you.
A note on our own tool (honesty first)
All Done's PDF-to-Word tool (https://all-done.xyz/tools/pdf-to-word) is a basic text extractor — it recovers text from PDFs that have a text layer, entirely in your browser. It has no OCR. It will not help with a scanned PDF, and its own disclaimer says so: "Basic text recovery — formatted PDFs and image-only scans may not extract cleanly."
I'm telling you this upfront because the rest of this guide sends you to other tools, and I'd rather you hear the limitation here than discover it after uploading your scan. For text-based PDFs where privacy matters, our extractor is the right call. For scans, use the OCR methods below — they're free, and they're better at this job than any text extractor could be.
Method 1: Google Drive "Open with Google Docs" OCR (the best free option)
This is the method that deserves to be famous and isn't. Google Drive has a full OCR engine built in — the same technology that powers Google Lens — and it applies it automatically when you open a PDF or image with Google Docs. It's free, it handles multi-page PDFs, and its accuracy on clean print is genuinely excellent.
**Strengths:** best-in-class free accuracy on clean printed text; handles dozens of pages; preserves rough page order; free with any Google account; output downloads as proper .docx.
**Limits:** formatting is not preserved — you get plain editable text, not a reconstructed layout. Tables come out as linear text you'll need to rebuild. Handwriting is unreliable (see accuracy section). And the privacy tradeoff is real: your scan uploads to Google's servers. For sensitive documents, weigh that against the alternatives below.
**Pro tip:** if the OCR looks rough, the scan quality is usually the culprit, not the engine — see the scan-quality tips section before blaming the tool.
- Go to drive.google.com and upload your scanned PDF (drag and drop, or New > File upload).
- Right-click the uploaded file > Open with > Google Docs.
- Wait. Drive runs OCR on every page — a 20-page scan takes a minute or two.
- A new Google Doc opens: your scanned pages as images at the top, with the OCR'd editable text below each image.
- Delete the images, clean up the text, then File > Download > Microsoft Word (.docx).
Method 2: Microsoft Word's built-in File > Open
If you have Microsoft Word (desktop, any recent version), you already own an OCR-capable converter. Word's PDF import runs its own text recognition on image-only pages.
**Strengths:** no upload — everything happens on your machine, so it's the most private of the free OCR routes. No page limits, no account needed beyond your Word license. Output is a native .docx you can keep editing.
**Limits:** requires a Word license (not free software, though many people have it through work or school). Accuracy on scans is good but generally a notch below Google's engine, especially on lower-quality scans. Like all OCR here, formatting is not preserved — expect plain text needing cleanup.
**When to choose it:** sensitive scans you don't want to upload anywhere. The privacy advantage is decisive for contracts, medical documents, or anything confidential.
- Open Word > File > Open > Browse, and select your scanned PDF (change the file filter to "All Files" or "PDF Files" if needed).
- Word shows a prompt: "Word will now convert your PDF to an editable Word document." Click OK.
- Wait while Word processes the file — image-only pages get OCR'd during this step.
- Review the result. Text from clean scans is editable; complex layouts need cleanup.
Method 3: OneNote "Copy Text from Printout"
Microsoft OneNote (free) has a lesser-known OCR feature that works per-page. It's the most manual of the three, but it's free, private, and surprisingly capable.
**Strengths:** completely free (OneNote is free), fully local (no upload), decent accuracy on clean print.
**Limits:** per-page manual process — impractical past ~10 pages. No batch export; you're copying page by page. Formatting is plain text only.
**When to choose it:** short scans (a page or two) where you want free, private OCR without a Google account or Word license. For a single-page signed letter, this is the fastest route.
- Open OneNote and create a new page.
- Insert > File Printout, and select your scanned PDF. OneNote imports each page as an image printout (this can take a while for long documents).
- Right-click a page image > Copy Text from Printout.
- Paste the recognized text wherever you need it — into Word, for example.
- Repeat per page.
The three methods compared
The decision: Google Drive for the best free accuracy on long documents you're comfortable uploading; Word for sensitive scans (local processing); OneNote for short free jobs without accounts or licenses.
- Google Drive OCR — Cost: free (Google account). Privacy: uploads to Google. Multi-page: excellent (batch). Accuracy (clean print): best of the three. Output: .docx download. Formatting preserved: no — plain text.
- Word File > Open — Cost: requires Word license. Privacy: fully local. Multi-page: good (batch). Accuracy (clean print): good. Output: native .docx. Formatting preserved: no — plain text.
- OneNote printout — Cost: free. Privacy: fully local. Multi-page: poor (page by page). Accuracy (clean print): good. Output: copy-paste text. Formatting preserved: no — plain text.
Honest accuracy expectations by document type
OCR is pattern recognition, not reading. Set expectations by what you're feeding it:
The universal rule: OCR gives you the words, never the layout. Anyone promising "perfect formatting preservation" from a scan is selling something. Plan your workflow around plain editable text, and you'll be satisfied; expect a reconstructed document, and you'll be disappointed.
- Clean printed text (laser printer, 300 DPI scan): very good. Expect ~98%+ character accuracy on body text — a few errors per page, mostly on unusual characters, which a spellcheck pass catches. (Illustrative estimate based on typical OCR performance; your results depend on the scan.)
- Photocopied or faxed pages: good to fair. Each generation of copying degrades edges; expect more errors on small text and punctuation. Still far faster than retyping.
- Tables and complex layouts: the words are recovered accurately, but the structure is not. Tables come out as linear text runs. Budget time to rebuild tables by hand — this is normal and expected, not a failure.
- Handwriting: variable to poor. Neat, consistent handwriting on clean paper: partially readable. Cursive, cramped, or stylized handwriting: largely unreliable. No free OCR handles handwriting well — this is a known hard problem across the industry, not a limitation of these specific tools.
- Mixed languages or unusual fonts: accuracy drops. Decorative fonts, heavy italics, and mixed scripts confuse the recognition engine. Stick to the cleanest pages for the best results.
Scan-quality tips that move the needle
OCR accuracy is dominated by input quality — these five tips improve results more than switching tools:
Illustrative example: the same contract page OCR'd from a 150-DPI skewed phone photo versus a 300-DPI flatbed scan typically goes from "needs heavy correction" to "needs a spellcheck pass." The tool didn't change — the input did. (Illustrative; described to show the relative impact, not measured data.)
- Scan at 300 DPI. Below 200 DPI, character edges blur and error rates climb. Above 300 DPI adds file size without meaningful accuracy gains for standard text.
- Straighten the pages. Skew is an accuracy killer — even a few degrees of tilt degrades recognition. Most scanner software has auto-deskew; use it. For phone photos of pages, retake crooked shots rather than OCR'ing them.
- Maximize contrast. Black text on white paper is the ideal. Gray, yellowed, or shadowed backgrounds hurt. If you're photographing pages, use even lighting and avoid shadows across the text.
- Clean the pages first. Remove staples, sticky notes, and margin scribbles before scanning — the engine tries to "read" all of them, injecting garbage into your text.
- One language per run. If the OCR tool lets you set the document language, set it. Auto-detection on mixed content is a common error source.
Privacy note for sensitive scans
Scans are often the most sensitive documents people OCR — signed contracts, ID documents, medical records, financial statements. Two rules:
And the complementary note: if your PDF is not a scan — it has selectable text — you don't need OCR at all, and you don't need to upload anything either. All Done's PDF-to-Word extractor (https://all-done.xyz/tools/pdf-to-word) pulls the text out entirely in your browser: no upload, no account, no trace. For sensitive text-based PDFs, that's the highest-privacy option available.
- Prefer local processing for sensitive material. Word's File > Open and OneNote never upload your file. Google Drive's OCR is excellent but your scan goes to Google's servers — fine for a textbook chapter, wrong for a medical record.
- Delete cloud copies when done. If you do use Drive, delete the uploaded PDF and the OCR'd doc from Drive and trash afterward. "Deleted" from trash is the actual deletion.
The bottom line
Scanned PDF to Word is a two-step job — OCR first, cleanup second — and the free tools cover it well once you pick the right one: Google Drive for accuracy and long documents, Word for privacy, OneNote for quick short jobs. Feed the engine a clean 300-DPI straight scan, expect plain text rather than preserved layout, and budget a cleanup pass. Do that, and a 50-page scan becomes an editable document in under an hour instead of a retyping project.
