Page-by-page OCR: complete, selective or assisted
For PDFs up to 20 pages, process every page, specify ranges such as 1-3,5, or inspect locally which pages appear scanned. Skipped pages retain their existing native text and receive no new OCR layer. Automatic detection is a heuristic: for pages mixing selectable text and scanned images, choose all pages to avoid missing content. Out-of-range selections are rejected before recognition starts.
Searchable PDF without uploading your file
The file is processed in your browser. The OCR engine and language model may be downloaded when required, but your document is not sent to the OCR provider. For PDFs, MakeFileWork preserves the original page appearance and adds a positioned text layer.
What OCR can and cannot guarantee
OCR accuracy depends on scan quality, resolution, rotation, typography and language. The output is reopened and checked for a searchable text layer, but recognized words should still be reviewed before relying on them for legal, financial or safety-critical use.
When OCR is the right workflow
Use OCR when you can see words on the page but searching, selecting or copying them does not work. This is common with scanned contracts, invoices, letters, forms and photographs of documents. If the PDF already contains selectable text, OCR may add no value and a different repair or extraction workflow may be more appropriate.
Searchable PDF and extracted text are separate outputs
The searchable PDF keeps the page image and adds text positioning so viewers can search and select recognized words. The text output is useful when you need plain recognized content for review or another application. MakeFileWork verifies that the generated PDF reopens with a text layer; it does not treat that structural check as proof that every recognized word is correct.
How searchable-text verification works
After OCR, the generated PDF is reopened and every distinct normalized positioned token is compared with text extracted from the output pages and, for existing PDFs, with the original selectable text. Words already present in the source do not count as evidence of a newly added OCR layer. If no positioned words exist, or the PDF text layer cannot be confirmed, the job fails instead of offering a falsely verified download. Repeated words are checked by occurrence count, not just by whether a word appears once. Word coordinates and per-page text volume are bounded before embedding. This check does not establish transcription accuracy; compare important values against the scan.
Verified downloads and mismatched files
Before recognition, the tool compares the actual file signature with its filename and declared MIME type; a renamed PDF masquerading as a PNG is rejected. After OCR, both the PDF and the TXT must match their original verification evidence. The PDF is structurally reopened and its page count checked; SHA-256 is recalculated when you download. Changing the input or language invalidates the previous result. This protects the delivery step, not the correctness of every recognized word.
Cancel a long OCR job safely
Use Cancel OCR if you selected the wrong file or language. The current result is discarded and the OCR worker or PDF render is stopped where supported. Loading a language model or finishing a browser operation may still take a moment; cancellation is not a guarantee of instantaneous memory release. No cancelled output is offered for download.
Safe limits
Files are limited to 20 MB, PDFs to 20 pages and pages or images to 25 megapixels. These bounds prevent a single job from consuming unbounded memory or CPU and make failures more predictable on phones and lower-memory browsers.
Reopen and verify batch results
After recognizing a batch, download the combined ZIP or its individual searchable PDFs and UTF-8 text files. Each accepted output has a SHA-256 checksum, and the ZIP contains a machine-readable manifest, a JSON report and a spreadsheet-safe CSV audit. A saved ZIP can be reopened later; supply an independent SHA-256 if you need to establish that the archive has not been replaced. A matching internal manifest alone cannot authenticate its origin. OCR verification checks structure and text availability, not transcription accuracy.
Batch OCR for mixed documents
Recognize up to six PDFs or images sequentially, download verified searchable PDFs and UTF-8 text in one ZIP, and inspect a per-file failure report. A failed file is never labeled as recovered. The OCR model may require a download; source documents stay in the browser.
Retry rejected files without repeating successful OCR
After a partial OCR failure, choose only one or several rejected originals, in any order. The saved name, size and SHA-256 must match before processing. Previously verified PDFs and TXT files remain byte-for-byte unchanged; unselected failures stay rejected. A failed retry is never marked successful. Older reports lacking source SHA-256 require every original file in its original order.