Turn a scanned PDF into searchable text

What OCR does to a scanned document

OCR stands for optical character recognition. Each page is drawn as an image, the recognition engine (Tesseract.js, running as WebAssembly in your browser) reads the letters in it, and the text is recorded together with where each word sits on the page. For a searchable PDF, that text is laid invisibly behind the original page image, so the page looks unchanged but you can search it, select it and copy from it.

You can also download the recognized text as a plain .txt file, which is handy for pasting into an editor or feeding another tool. You choose the languages first, then whether to skip pages that already contain text, and the resolution used to render pages (200 or 300 DPI). Besides PDFs, the tool accepts JPG and PNG images as input.

Documents are never uploaded. The one thing that is fetched over the network is the OCR engine and the language data, which are downloaded from a public CDN (jsDelivr) the first time you use the tool, a few megabytes in total. The CDN can see your IP address and standard request headers, but not your documents.

Scanned pages versus pages that already have text

Not every PDF needs OCR. A file exported from a word processor already contains real text, and running OCR over it would only add work and risk small errors. A scan, on the other hand, holds images only. A quick test: open the file and try to select a word with the cursor. If nothing highlights, or Ctrl+F finds nothing, the pages are pictures.

Mixed files are common, for example a digital contract with a few scanned signature pages. The option to skip pages that already have text handles them: the digital pages are left untouched and only the scanned ones are processed. Once done, the searchable PDF supports every text tool, so you can go on to Compress PDF, PDF to Word or PDF to Text with useful results.

• Test first: if you can select text, the page is not a scan.
• Use the skip option for mixed documents.
• Search for a distinctive word afterwards to confirm the result.

Choosing languages and understanding scripts

The engine reads better when it knows what to expect, so select the language of the text and not just the one you happen to speak. If a page mixes French and English, select both; recognition takes slightly longer but it stops mistaking one language's accents for noise. Selecting many languages you do not need slows things down and can cause stray misreadings, so pick only the ones that are on the page.

There is one important difference between scripts. For Latin-script languages (English, French, Spanish, German, Italian, Portuguese, Dutch and Turkish), the tool can build a searchable PDF whose invisible text layer sits behind the page image. For Arabic and Russian, which use other scripts, that searchable layer is not available in this tool. It still recognizes them, but delivers the result as a .txt file. If you need a searchable Arabic PDF, this tool will not produce it.

• Bilingual page: select both languages.
• Arabic or Russian: use the .txt download.
• Numbers, dates and Latin names inside an Arabic page are best proofread by eye.

Getting the most accurate result

Image quality matters more than any setting. The most reliable input is a page scanned at 300 DPI, straight on the glass, in even light, with dark text on a light background. Skewed photos, shadows, creases, faded ink and very small print all reduce accuracy. If you are capturing a page with a phone, hold it parallel to the paper, avoid your own shadow and fill the frame with the page.

Use the 300 DPI option for small print, footnotes and dense pages, and 200 DPI when text is large and you want faster processing. Higher resolution also means more memory and more time per page. Tables and multi-column layouts are read line by line, so the order of the text may not match the visual layout; check those areas by eye.

• Scan at 300 DPI or higher where possible.
• Straighten and crop pages before OCR if they are tilted.
• Do not expect reliable handwriting recognition.
• Proofread names, figures and dates against the original page.