OCR PDF - Make Scanned PDFs Searchable
Run optical character recognition across a scanned PDF so its text becomes selectable, searchable and copyable.
Turn a scanned PDF into one you can search
OCR (optical character recognition) reads the words in a scanned PDF and places an invisible copy of them behind each page image, so the file looks the same but Ctrl+F finds its words and you can copy a paragraph out. Piqav's OCR runs in your browser with the open-source Tesseract engine, so the scan is never uploaded. It reads printed text in eleven languages - English, German, French, Spanish, Portuguese, Italian, Dutch, Hindi, Marathi, Bengali and Tamil - and is not built for handwriting.
Does your PDF need OCR?
Open it and press Ctrl+F (Cmd+F on a Mac), then search for a word you can see on the page. If the search finds nothing, or you cannot drag across a line to select it, the page is a picture of text: a scan, a phone photo saved as PDF, or a fax. That is what OCR is for.
If the search does find the word, the PDF already has real text, the kind a PDF exported from Word or a website has. Running OCR on it replaces sharp text with a picture of it plus a recognised copy, which is worse on both counts.
What we measured, in every language
On 10 October 2026 we typed the same tenancy notice in each of the eleven languages: 50 to 80 words with a rent, a deposit and a number of days. We made a page image of each at 300 DPI and ran it through this tool exactly as the screen does. Each page was read twice: once clean, and once with a paper tint, a slight blur, a 0.6-degree skew and grain added, like an office photocopy.
| Language | Words read, clean | Words read, rough | Amounts right |
|---|---|---|---|
| English | 67 of 67 | 67 of 67 | 3 of 3 |
| German | 59 of 59 | 59 of 59 | 3 of 3 |
| French | 77 of 77 | 77 of 77 | 3 of 3 |
| Spanish | 74 of 74 | 74 of 74 | 3 of 3 |
| Portuguese | 67 of 67 | 67 of 67 | 3 of 3 |
| Italian | 66 of 66 | 66 of 66 | 3 of 3 |
| Dutch | 66 of 66 | 66 of 66 | 3 of 3 |
| Hindi + English | 80 of 80 | 80 of 80 | 3 of 3 |
| Marathi + English | 52 of 53 | 53 of 53 | 3 of 3 |
| Bengali + English | 58 of 59 | 58 of 59 | 3 of 3 |
| Tamil + English | 46 of 49 | 47 of 49 | 3 of 3 |
Each page took about two seconds. In the finished PDFs a search found every recognised word, except one Bengali word of 58. Three things came out of the test:
- The language order matters for numbers. Read as Hindi first, the Hindi notice came back with 1,450 as 7,450 and 10 as 0. With English first, every figure was right, so the Indian languages run that way.
- Indian-language text is now searchable. Until 10 October the invisible text layer could only hold Western European letters, so a Hindi word was recognised but could not be found by a search. It now uses fonts for Devanagari, Bengali and Tamil, with the letters in the order they were typed, which is what a search looks for. A first version missed Bengali words ending in the full stop ।, which Unicode files under Devanagari; that is fixed.
- More resolution did not help. The tool reads each page at 144 DPI. We also tried 300 DPI in English and Hindi, at 11-point and 8-point type: every word read the same, and the output files were roughly three to eight times larger.
Limitations. These scans were simulated, not made on a scanner, and each language had one paragraph in one font. Real paper adds creases, stamps, uneven light and handwriting, so expect lower figures on a poor original and check anything that matters. Earlier, on 27 September, a six-page English report rendered as page images read 198 of its 204 words.
Getting better results
- Choose the document's language. The recogniser uses the language's dictionary to settle letters it is unsure of, and as the Hindi test shows, the wrong choice can change figures.
- Scan at 300 DPI. Lower resolutions blur small type; higher ones rarely help and make huge files. Why 300 is the usual answer is in what DPI to scan at.
- Straighten pages first. Recognition is weakest on sideways pages. Rotate PDF fixes pages that scanned in the wrong way up.
Why some words come out wrong
OCR guesses each letter from its shape, so anything that changes the shape costs accuracy. That includes a page fed in at an angle, faint or grey print, a coffee stain or a fold across a line, and type smaller than about 8 points. Heavy JPEG compression hurts too, because it leaves smudges around the edges of letters. Numbers suffer most, because there is no dictionary to catch a 1 read as a 7. Compare figures, names and account numbers against the original before you rely on the copied text.
What the output is
A new PDF built from an image of each page with the recognised words behind it. Everything visible comes across, including filled-in form answers, as part of the picture. What does not come across is anything that lived outside the page image: bookmarks, links, fillable fields, attachments.
How to make a PDF searchable
Frequently asked questions
What does OCR actually do to my PDF?
It reads the words off each page image and lays an invisible copy of them over the picture, word by word, in the same place. The page looks exactly as before, but you can now search it, select a line and copy it. You can also copy all the recognised text at once from the result screen.
Which languages are supported?
English, German, French, Spanish, Portuguese, Italian, Dutch, Hindi, Marathi, Bengali and Tamil. Choose the document's language before you start. The four Indian languages are read together with English, because documents in them usually carry English names, figures and terms.
How long does OCR take?
About two seconds a page in our tests on a laptop, in every language. The first run also downloads the language model once, between 0.7 MB (French) and 3 MB (English, Dutch); for Hindi, Marathi, Bengali and Tamil the English model comes too, about 4.3 to 4.9 MB in all. Phones and long scans take longer.
How do I know if my PDF needs OCR?
Press Ctrl+F (Cmd+F on a Mac) and search for a word you can see on the page. If nothing is found, or you cannot select the text, the page is a picture and needs OCR. A PDF exported from Word or a website already has real text and gains nothing from it.
Why were some words missed or wrong?
Recognition drops on skewed pages, faint or low-contrast print, heavy JPEG compression and very small type. Straighten pages with Rotate PDF, rescan at 300 DPI if you can, and choose the right language: read as Hindi alone, our test turned 1,450 into 7,450. Always check names and figures by hand.
Does OCR keep my formatting?
The page keeps its exact look, because the page image stays. OCR adds an invisible text layer behind it, so copied text comes out as plain text without fonts or columns. Bookmarks, links, fillable fields and attachments in the original are not carried into the new file.
Can it read handwriting?
Not reliably. It is built for printed text - typed letters, books, receipts, forms. Neat block capitals sometimes work; joined-up handwriting mostly does not.