# PDF OCR > Convert Scanned PDF to Searchable PDF - 100% in Your Browser Free PDF OCR. Convert scanned PDFs to searchable PDFs in your browser. Tesseract.js + pdf-lib. 13 languages, 150-300 DPI. No upload, 100% private. URL: https://tools.scoreroute.com/tools/pdf-ocr/ Markdown: https://tools.scoreroute.com/tools/pdf-ocr/.md ## Make Scanned PDFs Searchable Upload a scanned PDF, pick a language, and get back a real searchable PDF with an invisible text layer you can Ctrl+F through. No upload, no signup, no daily limit. ### OCR Complete Your PDF now has a searchable text layer. ## Why Scanned PDFs Need OCR Privacy section: All file processing happens inside your browser. We do not upload your PDF anywhere, no analytics record the file, no accounts are needed. Open the browser DevTools Network panel and you will see zero outbound traffic carrying the document. This matters because scanned PDFs often contain receipts, contracts, medical records, tax forms, and other documents with personally identifiable information or business confidential data. Sending those files to a remote server you cannot audit means handing over control to that server security posture, retention policy, breach history, and jurisdiction. Our tool keeps every byte local: the PDF is read with the standard File API, parsed by pdfjs-dist, rendered to a canvas, sent through Tesseract for character recognition, and reassembled into a new PDF by pdf-lib, all inside the same browser tab. Problem section: A scanned PDF is a photograph of a page wrapped in PDF metadata. The page may look like text to a human, but to a computer it is just pixels. You cannot search inside it, you cannot copy text out of it, you cannot paste a quote into an email, you cannot feed it to a translation tool, you cannot have a screen reader speak it for accessibility, and you cannot extract structured data from a table. Every one of those tasks fails because the PDF contains no text layer. Optical character recognition, or OCR, is the bridge between pixels and characters: it inspects each glyph shape, matches it against learned letter forms, and emits the text along with positional confidence. Our PDF OCR tool adds that invisible text layer back into your PDF so the file becomes a living document you can search, copy, quote, translate, and archive. ## How It Works How It Works section: Upload the PDF, pick the document language so Tesseract uses the right character model, choose the resolution between 150 and 300 DPI, and click Start OCR. The tool renders each page to an offscreen canvas at the selected DPI, hands the image to Tesseract, reads the recognized words along with their bounding boxes, and writes a fresh PDF where every page has the original rasterization plus an invisible text overlay positioned to match. The output looks identical to the input but now behaves like a born-digital document. You can search across pages with Ctrl+F, copy text with Ctrl+C, and run any downstream tool that expects a real text layer. For accessibility, screen readers will read the text aloud. For translation, you can paste content into a translator. For data extraction, you can pipe the text into a spreadsheet or script. ## Common Pitfalls Common Pitfalls section: Selecting the wrong language is the single most common failure mode. English-only models will produce nonsense when given Chinese characters, and a Chinese model will fail on Latin script. If your document mixes languages, pick the dominant script and accept some accuracy loss on the other. Low DPI on the render step makes characters blurry, which costs accuracy. Below 200 DPI, expect character confusion on small fonts. Below 150 DPI, expect rejection rates above ten percent even on clean scans. Highly stylized fonts, decorative scripts, and heavy noise or smudges degrade recognition. Multi-column layouts where text wraps tightly can confuse the line-order heuristic and cause output to interleave between columns. Tables without grid lines are particularly hard: Tesseract reads them as flowing prose and the cell boundaries are lost. For those documents, run OCR first to get the text, then manually reconstruct the table structure. ## Why AI Chatbots Fall Short Why AI Chatbots Fall Short section: Large language models like ChatGPT, Claude, and Gemini process text tokens. They can describe a scanned page in words, but they cannot perform the actual pixel-to-character mapping that OCR requires. Even when a chatbot accepts an image input, the output is a textual description, not a modified PDF. To make the PDF searchable, the chatbot would need to read every pixel, run a recognition model, generate a brand-new PDF, embed an invisible text layer at the correct coordinates, and stream the bytes back as a download. No general-purpose chatbot has that pipeline. Some vision-capable models can extract text from images, but their output is unstructured prose without bounding boxes, so you cannot embed it back into a PDF as a searchable layer. Our tool runs the actual Tesseract OCR engine plus the pdf-lib PDF writer inside the browser, producing a real searchable PDF that AI chatbots cannot produce on their own. ## Technical Details This tool chains three client-side libraries without any server round trip: - [pdfjs-dist](https://mozilla.github.io/pdf.js/) parses the PDF, walks the page tree, and renders each page to an offscreen canvas at the chosen DPI. - [Tesseract.js](https://github.com/naptha/tesseract.js) runs the Tesseract OCR engine compiled to WebAssembly inside a Worker, returning per-word bounding boxes. - [pdf-lib](https://pdf-lib.js.org/) builds the output PDF, embeds the rendered page image as the visual layer, and overlays the recognized text invisibly (PDF text rendering mode 3) at the OCR-reported positions using the standard Helvetica font for Latin coverage. ## Authoritative References - [Tesseract OCR](https://github.com/tesseract-ocr/tesseract) - the open-source engine Tesseract.js wraps (sponsored by Google since 2006). - [ISO 32000-2:2020 (PDF 2.0)](https://www.iso.org/standard/63534.html) - the PDF specification that defines the text rendering modes used to embed an invisible search layer. - [MDN Canvas API](https://developer.mozilla.org/en-US/docs/Web/API/Canvas_API) - the browser API that lets pdfjs-dist hand Tesseract a renderable image. - [MDN File API](https://developer.mozilla.org/en-US/docs/Web/API/File_API) - the browser API used to read the uploaded PDF without ever sending it over the network. ## Frequently Asked Questions **Will my PDF be uploaded to a server?** No. The entire pipeline runs in your browser. The PDF is read with the standard File API, parsed by pdfjs-dist, rendered to an offscreen canvas, sent through Tesseract.js for character recognition, and reassembled into a new PDF by pdf-lib. No bytes leave your device. You can confirm this by opening your browser DevTools Network panel before clicking Start: you will see zero outbound requests carrying the document content. Only the Tesseract language pack (a few MB, cached after first load) is fetched from a CDN. **What does searchable PDF mean?** A searchable PDF contains both a visual image of the page (what you see) and an invisible text layer (what computers read). Search engines, screen readers, copy-paste, and translation tools all rely on that text layer. A scanned PDF without OCR is just a photograph of text inside a PDF wrapper - you cannot search inside it, copy text from it, or quote a paragraph. Our tool adds the invisible text layer by running Tesseract.js OCR on each page and embedding the recognized words at their original positions. **Which languages are supported?** English, Spanish, French, German, Italian, Portuguese, Dutch, Russian, Chinese (Simplified and Traditional), Japanese, Korean, Arabic, and Hindi. The Tesseract language pack is downloaded on demand and cached by your browser. Note: the invisible text layer in the output PDF is rendered with the standard PDF Helvetica font, which fully covers Latin scripts. For non-Latin scripts (Chinese, Japanese, Korean, Arabic, Hindi, Russian), the visual page image stays accurate, but the embedded text layer may show garbled glyphs when copy-pasted - the plain text download always preserves the original Unicode correctly. **What DPI should I choose?** 150 DPI is the fastest and produces small output files; accuracy is acceptable for clean printed text. 200 DPI is the recommended default - good accuracy on most fonts, modest file size. 250 DPI gives noticeably better accuracy on small fonts and low-contrast scans at the cost of larger files. 300 DPI is the upper limit and matches typical office scanner resolution; it produces the largest files and is worth using only when 250 DPI is not enough. Below 150 DPI, expect significant accuracy loss. **Can I OCR only specific pages?** Yes. In the Page Range field, enter ranges like 1-5 for the first five pages, 1,3,5 for non-contiguous pages, or 2-10,15,20-22 for a mix. Leave it blank or type all to process every page. Pages outside the range are copied into the output PDF unchanged (with no text layer added). **How accurate is the OCR?** On clean printed text at 200 DPI, Tesseract.js typically achieves 95-99% character accuracy. Common failure modes: low-resolution scans (use higher DPI), handwritten text (Tesseract is designed for printed text, not handwriting), heavy noise or smudges on the original, multi-column layouts where text wraps tightly, and tables without visible grid lines (Tesseract reads tables as flowing prose). For those documents, run this tool first to get the text, then manually reconstruct any lost structure. **Why can't AI chatbots OCR my PDF?** Large language models like ChatGPT, Claude, and Gemini process text tokens. When given an image, they can describe what they see in prose, but they cannot perform the actual pixel-to-character mapping that OCR requires. Even vision-capable models output unstructured text without bounding boxes, so there is no way to embed that text back into a PDF as a searchable layer at the correct coordinates. Our tool runs the actual Tesseract OCR engine plus the pdf-lib PDF writer inside the browser, producing a real searchable PDF that AI chatbots cannot produce on their own. **Is there a daily limit?** No. Unlike SmallPDF (2 tasks per day on free tier), iLovePDF (hourly limits and 25 MB cap), or Adobe Acrobat (requires sign-in plus a $20+/month subscription), this tool has no daily limit, no file size tier restrictions, and no signup. OCR as many PDFs as your device memory and patience allow. The only practical limit is that very large PDFs (hundreds of pages at 300 DPI) take proportionally longer because every page must be rendered and recognized. ## Related Tools - [PDF Compressor](/tools/pdf-compressor/) - shrink a PDF before OCR to speed up processing. - [PDF Merger](/tools/pdf-merger/) - combine multiple scanned PDFs into one before running OCR. - [PDF Splitter](/tools/pdf-splitter/) - extract specific pages to OCR separately. - [PDF to Word](/tools/pdf-to-word/) - export OCR'd text to editable DOCX. - [Image to Text (OCR)](/tools/image-to-text/) - OCR a single image instead of a full PDF. - [PDF to Excel](/tools/pdf-to-excel/) - pull tabular data out of OCR'd PDFs. ## References - [Wikipedia — Portable Document Format](https://en.wikipedia.org/wiki/Portable_Document_Format) - [Adobe — PDF SDK Documentation](https://opensource.adobe.com/dc-acrobat-sdk-docs/)