Make a Scanned PDF Searchable
Making a scanned PDF searchable means running optical character recognition on each page image and embedding the recognized words as an invisible text layer at their exact positions over the scan. The document looks identical afterward, but its text can suddenly be selected, copied, searched in any viewer, and indexed by Spotlight or Windows Search. This tool does the whole job on your device: Tesseract OCR compiled to WebAssembly reads each page and rebuilds it as the original image plus hidden, perfectly aligned text.
How it works
A scanned PDF contains no text at all — each page is one big picture of a page, which is why Ctrl-F finds nothing and nothing can be selected. This tool fixes that in three stages. First, each page is rendered to a bitmap at high resolution. Second, that bitmap goes through Tesseract, the open-source OCR engine, compiled to WebAssembly so it runs inside your browser's sandbox; the English recognition model (about 11 MB) is fetched once on first use and cached, and from then on everything — image data included — stays on your device. Tesseract returns not just the recognized text but a bounding box and confidence score for every single word.
Third comes the trick that makes the output searchable without changing its appearance: each output page is rebuilt as the original scan image with the recognized words drawn on top in an embedded font, at the position and size of each word's bounding box, with the text set to zero opacity. The words are genuinely in the page's content stream — a viewer's text engine sees them, so search, selection and copy work — but the rendering contributes no visible pixels, so what you see is still the untouched scan. Select a sentence in the result and the highlight tracks the printed words underneath, because each invisible word sits exactly over its printed twin.
Honest limits, so you know what to expect. Accuracy follows scan quality: clean, straight 300 DPI scans of printed text recognize very well; faxes, phone photos, skewed pages and unusual fonts do worse, and handwriting mostly fails. The bundled model is English — other languages need other models this tool does not ship. Processing is capped at 50 pages per run because your own CPU is doing the work at a few seconds per page. And because pages are rebuilt as image-plus-text, any vector content a page had (rare in true scans) is rasterized into the page image. If you want the recognized words as a plain text file rather than a searchable PDF, the OCR PDF tool uses the same engine for exactly that.
Worked example: a 22-page scanned lease
A tenant has lease-scan.pdf: 22 pages scanned at 300 DPI, 14.8 MB, zero selectable characters. First run on this page downloads the 11 MB English model (kept in the browser cache for next time), then recognition proceeds at roughly 3–4 seconds per page — about 80 seconds total on a mid-range laptop. The output is 15.1 MB: the original page images plus around 9,400 invisible words and one embedded font. Visually the two files are indistinguishable, but Ctrl-F for "security deposit" now jumps to clause 4.2 on page 7, the clause can be copied into an email, and within a few minutes macOS Spotlight surfaces the file when searching for the landlord's company name — text that previously existed only as pixels.
Frequently asked questions
How do I make a scanned PDF searchable without uploading it?
Load the scan into the tool on this page. OCR runs as WebAssembly inside your browser and writes an invisible text layer over each page, so the document becomes searchable without ever leaving your device.
Will the OCR change how my scanned pages look?
No. The recognized words are written with zero opacity on top of the original scan image, so every page looks exactly as before. Only the behavior changes: text can now be selected, copied and found.
How accurate is the text recognition?
Accuracy follows scan quality: a clean 300 DPI office scan typically recognizes well above 95 percent of words, while skewed, low-resolution or handwritten pages do markedly worse. Search still works around occasional misread words.
Can Windows or macOS search find text inside the OCRed PDF?
Yes. Operating-system indexers like Spotlight and Windows Search read the embedded text layer just as viewers do, so the file starts appearing in system-wide searches by its contents.
Does this tool support languages other than English?
The default recognition model is English, which also handles numbers and common punctuation. Text in other languages will be recognized poorly or not at all, so accented and non-Latin scripts need a dedicated OCR application.
Is there a limit on how many pages I can process?
Yes, 50 pages per run, because recognition happens on your own processor at a few seconds per page. Split a longer document first, process the parts, then merge them back together.