How to OCR a Scanned PDF and Make the Text Searchable

To make a scanned PDF searchable, run it through OCR, short for optical character recognition. OCR looks at the picture of each page, works out which letters are on it, and adds those letters back as an invisible text layer. The page looks the same, but you can now search it, select words and copy text out of it. With FileCrank's OCR PDF tool, that means uploading the file, choosing a language and downloading the result.

OCR is not magic, and the quality of your output is largely decided before you ever upload the file. This guide explains why scans aren't searchable in the first place, how to prepare scans that OCR well, what the tool can and can't do, and where you should still expect mistakes.

Why a scanned PDF isn't searchable

There are two very different kinds of PDF, and they look identical on screen.

A born-digital PDF is exported from Word, a browser or a design tool. It contains actual text characters, fonts and positions. Your viewer can search it because the letters exist as data.

A scanned PDF is made by a scanner or a phone camera. Each page is a single image, a grid of pixels that happens to show letters. To a computer, the word "invoice" in that image is no different from a photo of a tree. Search finds nothing, copy and paste gives you nothing, and screen readers have nothing to read.

You can usually tell which kind you have in ten seconds. Open the PDF and try to select a single word with your mouse. If the highlight snaps neatly around the word, there's text. If you get a rectangle over part of the page, or nothing at all, it's an image.

This matters for more than convenience. Searchable text lets you find a clause in a 200-page contract, lets accessibility software read the page aloud, and makes archives usable. It also lets you pull the content into other tools, such as converting the document with PDF to Word, where an image-only PDF would give you nothing useful.

What OCR does, step by step

Knowing the process makes the quality advice below make sense.

1. Render the page as an image. FileCrank turns each PDF page into a raster image at 150 DPI.
2. Find the text. The engine locates blocks, lines and words on the page.
3. Recognise the characters. It matches shapes against what it has learned about each language's letters and uses language knowledge to choose between similar candidates, such as "rn" and "m".
4. Place the text invisibly. The recognised words are positioned underneath the visible image at the spots where they appear.
5. Assemble the PDF. The pages are combined into a single searchable file.

FileCrank uses the open-source Tesseract engine for the recognition step. Note step 1: the output pages are built from that 150 DPI render of each page, which has consequences for both quality and file size that we'll come back to.

Prepare the scan: what makes OCR work

Most bad OCR results trace back to the scan, not the software. Here is what to check.

Resolution

Resolution is measured in DPI (dots per inch). The commonly recommended range for scanning text documents for OCR is around 300 DPI. Lower than that and small letters blur into each other; much higher gives larger files with little extra benefit.

A tool-specific caveat: because FileCrank renders pages at 150 DPI before recognising them, scanning at 600 DPI will not give you better recognition than scanning at 300. It just makes a bigger file to upload. Conversely, a scan made at a low resolution, such as a thumbnail-sized phone photo, cannot be rescued, since the detail simply isn't in the image. If your text is very small (footnotes, fine print, dense tables), expect weaker results at 150 DPI even from a good scan.

Straight pages

Skew, where the page is rotated a few degrees, is a classic cause of errors. Lines of text run uphill, and the engine's line detection suffers. Use the flatbed's guide edges, or when photographing, line the camera up directly above the page. Pages scanned upside-down or sideways need to be rotated before OCR; use Rotate PDF first.

Contrast and cleanliness

Dark text on a light, even background is ideal. Problems include:

• Faded or grey text, common with old photocopies and thermal-paper receipts.
• Shadows, especially down the spine of a book or from a phone held over the page.
• Show-through, where print from the back of a thin sheet appears faintly on the front.
• Coloured or patterned backgrounds behind text.
• Stamps, handwriting or highlighter over printed words.

When you can control it, scan in greyscale or black and white with sensible contrast, and avoid heavy "enhancement" filters that thicken letters until they merge.

Layout

Single-column text is the easy case. Multi-column pages, tables, forms and pages mixing text and images are harder, because the engine must decide the reading order. Even when each word is recognised, the order of words in copied text may jump between columns.

A quick pre-flight checklist

Turn photos into a PDF first

If your starting point is a folder of photos or scanner images rather than a PDF, convert them first. FileCrank's Scan to PDF accepts JPG, PNG, TIFF and WebP images, up to 50 MB in total, and combines them into one PDF. Each image becomes a page, and you can drag the thumbnails into the right order before converting. You can also choose a fixed page size (A4 or Letter) or let each page match its image.

Put the pages in the right order and orientation at this stage. Fixing it afterwards means extra steps. If a scan includes blank pages or covers you don't want, you can drop them later with the Remove Pages tool, and our guide to splitting, extracting and removing PDF pages covers that in detail.

How to OCR a PDF with FileCrank

1. Open the OCR PDF tool and upload your scanned PDF. Files up to 100 MB are accepted.
2. Choose the document language. At the moment the choices are English and Hebrew.
3. Start the process. OCR is the slowest operation on the site, so you'll see progress as pages are processed.
4. Download the result and test it. Search for a word you know is on the page, then select a sentence and paste it into a text editor to see what was recognised.

Limits to know about

• Languages. Only English and Hebrew are offered. Choosing the wrong language for a document gives poor results. For other languages, or documents that mix several, you will need different software.
• Time. OCR is limited to roughly two minutes in total per file, with a cap on each page. A long or complicated document may fail with a timeout. If that happens, split the PDF into smaller parts (see Split PDF), OCR each part, and merge them back with Merge PDF.
• Size. Uploads are limited to 100 MB per file.
• Rebuilt pages. Because pages are rendered at 150 DPI and the new PDF is built from those renders, the output is not a byte-for-byte copy of your original. Fine detail in the scan may look slightly softer, and the file size can differ from the input. If you need the original image quality preserved, check a sample page at full zoom, and keep your original file.
• Born-digital PDFs. There is no point in running OCR on a PDF that already has real text. It would be rebuilt from page images, which is worse than the original.
• Existing bookmarks and links. Since the output is assembled from page renders, don't expect bookmarks, form fields or clickable links from the original to survive. Keep the original if you depend on them.

If the resulting file is bigger than you can use, run it through Compress PDF.

How accurate is OCR, really?

Honestly: it's good on clean material and fallible on everything else, and there is no single accuracy figure that applies to every document. Anyone quoting one without knowing your scan is guessing.

Here is what tends to happen in practice:

• Clean printed text in common fonts comes out well, with the occasional error.
• Look-alike characters cause the classic mistakes: the digit 0 and the letter O, 1 and l and I, rn versus m, 5 versus S. These are particularly dangerous in numbers, codes and reference numbers, where no spell-checker will flag them.
• Punctuation and diacritics are often lost or mangled, because the marks are small and the first thing to vanish at a low resolution.
• Handwriting is generally not recognised reliably by this kind of engine. Don't plan on it.
• Tables are recognised word by word, but the structure of rows and columns isn't preserved in the searchable layer.
• Right-to-left text, such as Hebrew, works through its own language setting, and mixing it with English on one page (numbers, brand names, URLs) is a common source of oddly ordered output.

Practical consequence: use OCR output to find things and copy rough text, but proofread anything that matters. For legal citations, financial figures, names, addresses and identifiers, verify against the visible page. The text layer is invisible, so errors are easy to miss; the picture you see is always the scan, and the mistake lives underneath it.

A useful test: copy a full paragraph and paste it into a text editor. Mistakes that you'd never notice while reading the page become obvious.

Privacy: what happens to your document

OCR requires reading the content of your pages, so think about what you are uploading. FileCrank's OCR processes files on its servers: the PDF is uploaded over an encrypted HTTPS connection, processed, and deleted automatically within about an hour. No signup is required.

That works for most everyday documents like manuals, receipts and course notes. But consider carefully before uploading:

• Documents covered by confidentiality agreements or professional-privilege obligations.
• Medical, financial or identity documents about other people.
• Anything your employer or client has told you not to put into online services.

For those, run an offline OCR program on your own machine, where the file never leaves it. Our article on why browser-based tools can be safer than desktop software lays out the broader trade-offs, and it's worth reading with your own risk in mind rather than as a blanket answer. Whatever tool you use, delete temporary copies when you're finished.

When OCR is the wrong tool

• You need the text editable, not just searchable. OCR gives you a layer under an image. For editing, convert to a document format afterwards, and expect to clean up the formatting by hand.
• The scan is unreadable to you. If you can't read it comfortably at 100% zoom, the engine won't do much better. Rescan if you can.
• The document is handwritten. Type it up, or use a service built for handwriting.
• You only need one page. Extract it first with Extract Pages so you aren't processing 80 pages for one paragraph.

A workflow that works

1. Scan at about 300 DPI, upright and straight, with good contrast.
2. Combine images into one PDF with Scan to PDF, correcting order and orientation.
3. Remove blank or unwanted pages.
4. Run the file through the OCR PDF tool with the right language.
5. Test by searching for a few words and pasting a paragraph into a text editor.
6. Proofread numbers and names against the page, and keep the original scan.

Ready to try it? Upload your scan to OCR PDF. It's free, requires no account, and the files are deleted automatically afterwards. If you're starting from photos, begin with Scan to PDF.