How to Make a Scanned PDF Searchable With Free OCR
The short answer
A scanned PDF is a stack of pictures. It looks like text, but the file holds only pixels, which is why Ctrl+F finds nothing and you can’t copy a sentence out of it. OCR (optical character recognition) fixes that by reading the letters in each picture and writing them into the PDF as an invisible text layer, lined up over the words you see. The pages look exactly the same; they just become searchable and copyable.
To do it without uploading the scan, open OCR PDF, add the file, pick the document’s language and press “Make searchable”. A copy named with “-searchable” on the end downloads when it’s done.
How well it works depends far more on the scan than on the software. The rest of this guide covers what to check first, how to scan for good results, and where OCR still gets things wrong.
First, check whether you need OCR at all
Many PDFs that feel like scans already contain real text. Try this: open the file and drag across a line of text. If it highlights word by word, there’s a text layer and you don’t need OCR. If it draws a box over the whole page, or nothing happens, it’s an image.
| What you have | What to use |
|---|---|
| PDF with selectable text, and you want the words out | PDF to text: instant, no OCR, no recognition errors |
| Scanned PDF you want to search, copy from and keep as a PDF | OCR PDF |
| A photo, screenshot or single scanned image, and you just want the text | Image to text |
| A mix of typed and scanned pages | OCR PDF with “Skip pages that already have text” ticked |
PDF to text also tells you which pages came back empty. If all of them did, the file is a scan.
What OCR actually adds to the file
Our OCR PDF tool does three things to each page, one page at a time so long scans fit in memory:
- Render the page as an image at 300 DPI. The Tesseract documentation says “Tesseract works best on images which have a DPI of at least 300 dpi.” On phones the tool caps the image at 5 megapixels (about 230 DPI for a Letter page) so the browser tab doesn’t run out of memory.
- Recognise the text with Tesseract.js, a WebAssembly port of the open-source Tesseract engine. It finds lines and words and gives each word a position and a confidence score.
- Write each word into the page as invisible text, stretched to cover the word in the picture. The scanned image isn’t touched, so nothing looks different and the file grows only slightly.
Because the hidden words sit exactly over the visible ones, searching highlights the right spot and selecting text works like in any typed PDF.
Scan settings that make OCR accurate
If you control the scanner, these settings matter more than anything else. They follow the Tesseract documentation’s advice:
| Setting | Recommendation | Why |
|---|---|---|
| Resolution | 300 DPI | Tesseract works best at 300 DPI or more. 150 DPI scans and faxes lose small print. |
| Colour | Grayscale or black and white for text pages | Text needs contrast, not colour, and the files are smaller. |
| Alignment | Page straight on the glass | Skew degrades line finding. Tesseract’s docs recommend rotating until text lines are horizontal. |
| Orientation | Every page upright | The tool doesn’t detect sideways or upside-down pages. Turn them first with Organize PDF. |
| Text colour | Dark text on a light background | Current Tesseract versions expect it. |
| Edges | Trim dark scanner borders | Dark edges can be misread as characters. |
Phone scans
A phone works if you treat it like a scanner: hold it flat above the page, fill the frame, use daylight, and keep your hand’s shadow off the paper. A tilted photo of a page reads badly for the same reason a skewed scan does.
Quick arithmetic: is your text big enough?
At 300 DPI, a Letter page (8.5 × 11 in) renders at 2,550 × 3,300 pixels, about 8.4 megapixels. Ten-point body text is 10 ÷ 72 inches tall, so its em height is about 10 ÷ 72 × 300 ≈ 42 pixels. That’s comfortably readable for OCR. At 150 DPI the same text is about 21 pixels, and footnotes or fine print drop to a size where errors become common. This is why rescanning at 300 DPI fixes more problems than any setting in the software.
Languages
OCR reads letters using a model for each language. Pick the wrong one and accuracy falls: English data misreads accented letters such as ñ and ü, and can’t read Hindi, Arabic or Chinese at all.
Our tool offers nine: English, Spanish, French, German, Portuguese, Hindi, Arabic, Chinese (Simplified) and Urdu. For a non-English document, the “Also read English words” option is on by default, which helps with English names, email addresses and numbers mixed into the text, at a small cost in speed. Only English is downloaded up front; other language files (0.6 to 1.7 MB) download the first time you choose them.
Step by step
- Open OCR PDF and add the scan. If it’s password-protected, you’ll be asked for the password.
- Choose the language.
- Choose pages if you don’t need them all, using ranges like 1-3, 8. Leave it empty for every page.
- Leave “Skip pages that already have text” ticked for mixed files. A page counts as having text if it already holds at least 20 selectable letters or digits; skipping it avoids a duplicate layer.
- Press “Make searchable”. Expect roughly 2 to 6 seconds per page on a recent laptop, after a one-time download of the OCR engine (about 3.7 MB) and English data (about 1.9 MB). A 20-page scan usually takes one to two minutes; phones are slower. Cancel stops at once.
- Check the result. Open the new file, press Ctrl+F and search for a word you can see. The recognised text is also shown on the page, where you can copy it or save it as .txt.
If the original had a password, the searchable copy is saved without it and without any print or copy restrictions. Lock it again with Password protect PDF if it needs protecting.
How accurate to expect it to be
OCR guesses each letter from the picture, so treat its output as very good, not perfect.
- Sharp, straight, 300 DPI scans of printed text: nearly every word right.
- Faxes, skewed phone photos, small print, stamps and textured backgrounds: expect mistakes, often in exactly the places you care about: numbers, names and short words.
- Tables: words are found, but Tesseract’s documentation notes it struggles with table layout, so columns may not come out as you’d like.
- Handwriting: poor. Tesseract is built for printed text; typing handwritten notes is usually faster than correcting OCR of them.
The tool shows an average confidence score across all words. Under 70% means you should expect errors. For anything that matters, like an account number, a dollar amount or a date, check it against the image rather than trusting the hidden text.
Privacy
The documents people OCR are often the most sensitive they have: signed contracts, IDs, medical records, tax returns. Online OCR services upload them. Here, the engine and language data are downloaded from this site once and cached, and every page is recognised in your browser. The scan is never sent anywhere.
Common mistakes
- OCRing a PDF that already has text. You get every word twice in searches. Check by selecting text first.
- Scanning at low resolution to save space. You save a few hundred KB and lose the small print. Scan at 300 DPI, OCR, then compress if needed.
- Leaving pages sideways. Rotate first; orientation isn’t detected.
- Trusting numbers blindly. A 5 read as an S, or a 1 as an l, still “looks” right in search results. Verify figures.
When your scan is straight and sharp, OCR PDF turns it into a searchable file on your own device. For a single photo or screenshot, Image to text gives you the words directly.