Why the text layer is invisible — and why that’s the right design
A searchable scan has two layers: the original page image you see, and machine-readable text positioned at the exact coordinates of each printed word underneath. The text is drawn with zero opacity, so it never alters the document’s appearance — a pixel-perfect scan of a signed contract stays pixel-perfect. But viewers search, select, and copy against that hidden layer, and indexing systems can read it. This is the same architecture commercial OCR software (Adobe Acrobat, ABBYY) produces, because it preserves the scan as legal evidence of the original while making it usable as data.
The alternative — replacing the image with recognized text — would be destructive: OCR is never 100% accurate, and every recognition error would permanently corrupt the document. The invisible-layer approach means errors only affect search quality, never the document itself.
The privacy angle is bigger here than for most tools
Think about what people OCR: signed contracts, court filings, medical records, old IDs, financial statements — the paper trail of their lives. Uploading those to a cloud OCR service means the most sensitive documents you own sit on a third-party server during processing. Running the recognizer in your browser — Tesseract compiled to WebAssembly, with the language model cached locally — removes that exposure entirely: the only thing downloaded is the public model, and nothing about your document goes the other way. You can verify it live in the Network tab while a recognition runs. For a single photographed page rather than a PDF, the Image OCR tool does the same job; to get plain text out of an already-digital PDF, the PDF Converter’s text output is instant and needs no recognition at all.