How to OCR a scanned PDF and make it searchable

A scanned PDF looks like a document but behaves like a photograph: you cannot search it, select a sentence, or copy a number out of a table. OCR (optical character recognition) fixes that by reading the pictures of letters and adding real text to the file. This guide explains what OCR does, how to run it, what accuracy you can expect, and the scanning habits that make the biggest difference.

What OCR produces

Most OCR tools, including OCR PDF, produce a searchable PDF: the original page images stay exactly as they were, and an invisible layer of text is placed underneath, positioned to match the words in the image. The page looks identical, but now Ctrl+F finds words, text can be selected and copied, and screen readers can read it aloud. Search engines and document management systems index it.

The alternative output is plain text (or Word), where the recognised words are extracted without the images. That is what you want when the goal is to edit or reuse the content rather than preserve the look of the scan.

Running OCR on webutilitypro.com

1. Open OCR PDF and upload the scanned PDF.
2. Choose the language of the document. Picking the right language matters: the recogniser uses dictionaries and letter-shape models per language, and a German document processed as English will have many errors in words with umlauts.
3. Click Run OCR. Processing takes a few seconds per page; a 50-page scan can take a minute or two.
4. Download the searchable PDF. Test it by searching for a word you can see on the first page. Files are removed from our server after 60 minutes.

If you then need the text in an editable form, run the searchable PDF through PDF to Word or PDF to Text; both read the text layer OCR created.

What accuracy to expect

On a clean, straight, 300 DPI scan of a printed document in a common font, modern OCR gets well over 99 percent of characters right, which means a page of 3,000 characters has perhaps 10–30 errors, often in punctuation or in rare words. Accuracy falls with:

• Low resolution. Below 200 DPI, small print and thin fonts become ambiguous (rn versus m, 1 versus l versus I).
• Skew and curl. Pages scanned at an angle, or book pages that curve into the spine.
• Poor contrast. Faded typewriter text, coloured backgrounds, photocopies of photocopies.
• Unusual layouts. Multi-column text, tables without lines, text wrapped around images, and footnotes can be recognised correctly but assembled in the wrong order.
• Handwriting. Standard OCR does not read handwriting. Specialised handwriting recognition exists and is far less accurate than print OCR.
• Mathematical notation, chemical formulas, musical scores. These need specialised tools.

OCR output is a draft. For anything that will be quoted, relied on legally, or used as data, proofread it against the scan.

Scanning so OCR works

The best OCR improvement is a better scan. If you control the scanner:

• 300 DPI for normal documents; 400 DPI for small print (footnotes, financial tables). 600 DPI rarely helps and makes files large.
• Grayscale rather than colour for text documents; it halves the file size and OCR engines convert to grayscale internally anyway. Colour only when colour carries meaning (highlighted text, coloured charts).
• Straight pages. Use the scanner's document feeder or a guide, and enable auto-deskew if the scanner offers it.
• Flat pages. For books, press the spine down or use a book scanner setting. Phone scanning apps do a decent job of flattening and deskewing and are often better than a flatbed for books.
• Clean glass and good contrast. Dust specks become stray characters.

Does the PDF already have text?

Many scanners and copiers OCR automatically. Before running OCR, open the PDF and try to select text. If words highlight individually, there is already a text layer; running OCR again is harmless but unnecessary. If the whole page highlights as one block or nothing selects, there is no text layer.

Languages and mixed documents

Choose the primary language of the document. For documents mixing two languages (an English contract with French clauses), choose the dominant one and expect more errors in the other. Documents in a script the OCR engine was not trained for (some Indic scripts, historical typefaces, Fraktur) need an engine with that language pack; if the result is gibberish, the language was the problem, not the scan.

What OCR does not change

• The page images stay as they are; OCR does not clean, straighten or sharpen them (some scanning software does that separately).
• Layout is not reconstructed. A searchable PDF is still a picture of the page with text underneath. For a reflowable document you need conversion to Word afterwards, with the usual clean-up described in our PDF to Word guide.
• File size usually increases slightly, because the text layer is added to the images rather than replacing them. Running Compress PDF afterwards can reduce the image part.

OCR without uploading

For sensitive scans, OCR locally. macOS Preview and Photos recognise text in images on recent versions (Live Text), and Adobe Acrobat has Scan & OCR. On any system, ocrmypdf input.pdf output.pdf (open source, built on Tesseract) produces a searchable PDF and is what many professional archiving workflows use. These are slower to set up than a web tool but keep the documents on your own machine.

A workflow for an archive of scans

Scan at 300 DPI grayscale with auto-deskew, OCR each file as it is created, name the files by date and subject, and store them in a folder that your operating system indexes. From then on, finding a document is a search for a word you remember rather than a hunt through thumbnails, which is the whole point of OCR.