Guide
Making scanned PDFs searchable without uploading them
A scanned PDF is a stack of photographs of paper. It looks like a document, but your computer sees only pixels: search finds nothing, you can’t copy a sentence, screen readers stay silent and PDF-to-Word converters return empty pages. OCR (optical character recognition) reads the words in those pictures and adds them to the file as real text. This tool does it entirely in your browser, which is rare for OCR and important for the kind of documents people scan.
What the result looks like
Nothing visible changes. The recognised words are placed as an invisible layer exactly over the printed words on each page. In any PDF reader you can now:
- search for a name, amount or clause,
- select and copy text,
- have the document read aloud or navigated with a screen reader,
- convert it with PDF to Word or PDF to Text.
Because the original page image is kept untouched, there is no loss of quality and the file grows only slightly.
Getting good results
OCR accuracy depends mostly on the scan:
- Resolution: 300 DPI is ideal for normal print. Leave quality on Accurate, which renders each page at 300 DPI for recognition.
- Straightness: tilted lines are read less reliably. Straighten crooked scans first with Deskew PDF.
- Contrast: dark text on light paper works best. Faded, shadowed or colored backgrounds reduce accuracy.
- Content: printed text in ordinary fonts is read well; handwriting, stylised fonts and text over pictures are not.
After the run, the tool shows the average confidence. Above 85% is typical for clean scans. Below 60% you get a warning; improve the scan and run it again.
Languages
The recognition model is trained for English. It also reads most text in other languages written with the basic Latin alphabet, such as Spanish, French, German or Indonesian, though accented letters are recognised less reliably. Other scripts, including Devanagari, Arabic, Cyrillic and Chinese, are not supported yet.
Pages that already have text
Many PDFs mix scanned and digital pages. Pages that already contain real text are skipped by default, which saves time and avoids doubling the text. Turn on Also read pages that already contain text if a page’s existing text is garbage, as sometimes happens with old scanner software.
Speed and the first run
Recognition runs on your device, page by page, typically 5 to 15 seconds per page on a computer and longer on phones. The first time, the browser downloads the recognizer, about 7 MB, from this site. It is then cached, so later runs start immediately and work offline.
Why local OCR matters
OCR is usually a cloud service: you upload the scan and get text back. The scans people need to search are often the most sensitive: contracts, medical letters, tax records, ID documents. Here, the recognizer runs in your browser, the pages never leave your computer and nothing is stored anywhere but your downloads.
After OCR
Copy or download the recognised text below the result, or use the searchable PDF. To reduce its size, use Compress PDF with the Light level, which keeps the text layer.
A worked example: an archive of letters
An office has a 40-page PDF of scanned letters and wants to find every mention of a supplier.
- Open the PDF, keep Accurate and run OCR; on a laptop this takes a few minutes.
- Check the average confidence shown with the result.
- Open the searchable PDF in a reader and search for the supplier’s name; every page where it appears is found.
Checked against the current version of the OCR PDF on .