How to OCR a PDF for Free (2026)
A scanned PDF looks fine, but try to search it or copy a line of text and nothing happens. That's because your "PDF" is really just a stack of images with no text layer underneath. OCR — optical character recognition — reads those images and writes real, searchable, selectable text back into the file. This guide walks through how OCR works, when you actually need it, how to do it for free in your browser, and the handful of mistakes that produce unusable output.
What OCR Actually Does
A normal PDF stores text as text — each letter is a character with a font, size, and position, which is why you can copy it, search it, and resize it without losing quality. A scanned PDF is different. When you run a document through a scanner, a phone camera, or a "print to PDF" of an image, the result is a picture of text, not text itself. The computer sees pixels that happen to look like the letter A, but it has no idea what that A is.
OCR reads the image, identifies each character based on its shape, and writes the recognized text back into the PDF as an invisible layer aligned behind the original image. The document looks identical to the original, but now Ctrl+F works, screen readers can read it, copy-paste returns real characters, and anything that processes text (contract software, e-discovery platforms, accessibility checkers) can actually see the content.
Modern OCR engines are good. In our tests of the tool, clean printed pages at 300 and 150 DPI, and a blurred, noisy typewriter page tilted 0.8° at 200 DPI, all came back with zero character errors in about 5 to 6 seconds per page. Contrast is what broke it: a faint, low-contrast photocopy at 200 DPI produced mostly garbage, with about four times as many character errors as there were characters, after 55 seconds. Handwriting is also outside what Tesseract is trained for.
When You Actually Need OCR
Not every PDF needs OCR. If you can already highlight text with your cursor and the text comes out clean, the file already has a text layer and you're done. Running OCR on a file that's already text-based just adds processing time and can occasionally damage the existing layer. A quick test: open the PDF, press Ctrl+F, and search for any word you can see on the page. If the search finds it, OCR is unnecessary.
You do need OCR when: you've scanned a paper contract, receipt, or archive document; you've saved photos of whiteboards, menus, or pages as PDF; you've received a "PDF" from someone that's really just screenshots glued together; you need to search or redact a large batch of old records; or you need an accessible version of a document for screen readers. Any time Ctrl+F comes back empty on a file that clearly has words on it, OCR is the fix.
Legal and accessibility requirements are a quiet driver of OCR work that's worth flagging. Under Section 508 and WCAG 2.1, PDFs distributed by public institutions or large employers generally need a real text layer to be considered accessible. Scanned-only PDFs fail that test. If you're publishing PDFs on a website, especially for a government, school, healthcare, or enterprise audience, OCR isn't optional — it's the baseline.
Method 1: UDT PDF OCR (Free, Browser-Based)
The UDT PDF OCR tool runs Tesseract, the open-source OCR engine first developed at HP and maintained for years by Google, compiled to run in your browser. Nothing uploads. The scan runs locally using WebAssembly. That matters for privacy-sensitive documents: medical records, contracts, HR files, and legal exhibits never leave your computer.
The workflow is:
- Open the PDF OCR tool and drop your scanned PDF into the upload zone.
- Pick the document's language. English is the default, and the tool offers 12 languages (English, Spanish, French, German, Italian, Portuguese, Dutch, Russian, Japanese, Chinese (simplified), Korean, Arabic); accuracy drops significantly if you pick the wrong one for your document.
- Let the engine process. Expect 2–6 seconds per page on a modern laptop; older hardware or very long documents take proportionally longer. The tool shows per-page progress.
- Download the result. You get a new PDF that looks identical to the original but now has a selectable text layer and searches in Ctrl+F, along with a plain .txt export of the extracted text if you just need the words.
Because everything runs in-browser, there are no page limits, no watermarks, no account signups, and no upload-size caps beyond what your own computer's memory can handle. For long documents, splitting the file into chunks with the PDF splitter and processing each chunk is faster than waiting on one huge job.
Method 2: Adobe Acrobat Pro
Adobe Acrobat Pro (the paid version — Reader won't do it) has OCR built in under Tools → Scan & OCR → Recognize Text. Adobe's implementation is excellent: highly accurate, preserves page layout well, and handles mixed-language documents and difficult scans better than most alternatives. For professionals who already pay for Creative Cloud or Acrobat Pro, it's a reasonable default.
The trade-offs are real, though. Acrobat Pro is a paid subscription. It uploads the document to Adobe's cloud during processing if you use the web version of Acrobat, and the desktop version requires installing a large application. There is also a known behavior where Acrobat occasionally "fixes" what it thinks are errors in the source document — auto-correcting proper nouns, forcing standard fonts, or re-flowing paragraphs — which is fine for casual use but can be a problem for legal documents where the original scanned appearance needs to be preserved exactly.
The short version: if you OCR PDFs for a living, Acrobat Pro is worth the subscription for edge-case handling. If you OCR a handful of documents per month, a browser-based free tool does the same job with no subscription and no cloud round-trip.
Getting Accurate Results
OCR accuracy is mostly about what you feed the engine, not the engine itself. The single biggest factor is scan resolution. Resolution mattered less than you might expect: clean text scanned at 150 DPI read perfectly in our test. Contrast and noise matter more, so 300 DPI grayscale remains a safe default rather than a magic number. If you're scanning originals specifically to OCR them, set your scanner to 300 DPI grayscale and don't overthink it.
Skewed pages tank accuracy. Straighten skewed pages first; the 0.8° tilt in our test was harmless, but larger skews make lines harder to separate. Before running OCR, use the PDF page rotator to straighten any pages that scanned crooked. Most good OCR tools will auto-deskew, but the less the engine has to guess, the better the output.
Clean scans beat filtered scans. Scanner software often applies auto-contrast, sharpen, or "enhance" filters that look nice visually but create letter-shape artifacts that confuse OCR engines. If your scans look heavily processed — harsh black-and-white, aggressive edge sharpening, visible halos around text — re-scan in plain grayscale without filters. The raw file looks worse to the eye but recognizes dramatically better.
Language selection matters more than people realize. Tesseract ships with models trained on specific languages and character sets. Running an English model on a Spanish document doesn't just miss the accented characters — it mis-recognizes whole words because the language model biases toward English letter patterns. If your document is bilingual, some tools let you pick multiple languages at once; otherwise, process each section separately.
Common Pitfalls to Avoid
Don't OCR over an existing text layer. If a PDF already has selectable text and you run OCR on it anyway, some tools overwrite the clean text layer with a lower-accuracy OCR version, replacing good text with worse text. Check first with Ctrl+F, and if the file is already searchable, skip OCR entirely.
Don't trust OCR output for high-stakes data entry without human review. Even at 99% accuracy, a 10-page contract has thousands of characters — that's tens of potential errors, often in exactly the places that matter (dollar amounts, dates, names). For legal filings, financial records, or medical data, treat OCR as a starting point that still needs a human read-through against the original.
Don't OCR a password-protected PDF without unlocking it first. The image content is encrypted, so the engine can't read the pixels. Use the PDF unlock tool first (on documents you have the right to unlock), then OCR the decrypted copy.
Don't assume the output is redaction-safe. An OCR'd PDF has both an image layer and a text layer on top of each other. If you cover information with a black rectangle visually, the underlying text layer still contains the original words and anyone can copy them out. For actual redaction, use the PDF redactor, which removes both layers.