Redacting a scanned PDF: where OCR helps and where to review
How to inspect image-only and mixed PDF pages, use OCR as a search aid and review the exported result without assuming recognition is complete.
A scanned PDF can look like ordinary text while behaving like a photograph. Searching for a name may return nothing even when the name is plainly visible. Before deciding that a value is absent, establish what kind of page you are reviewing.
This is a preparation checklist for an authorised copy you intend to share. It does not describe a way to recover someone else’s concealed information, and it does not make OCR a guarantee. The PDFRedactLab example is fictional and does not perform OCR or process documents.
Start with a page inventory
Try selecting a short paragraph and searching for a clearly visible word using normal viewer controls. A failed search is a clue, not a diagnosis: the page might be image-only, poorly recognised or encoded in an unexpected way. Inspect more than the first page. A dossier can mix a digitally created cover, scanned letters and photographs.
Record which pages need visual review and which appear to have a searchable layer. Rotation, faint ink, stamps and text crossing a fold deserve particular attention. If a page is illegible, do not treat automated detection as a substitute for understanding it. Obtain a clearer authorised source where possible.
OCR adds a way to search
Adobe’s OCR instructions explain how recognition creates searchable text from a scan. That can help locate known words. The page image and recognised text are different representations, so an effective redaction workflow must account for both rather than assuming a search hit describes the whole page.
Choose recognition settings supported by your tool and the document language. Check a few visible words against the recognised result before relying on search. Test ordinary words as well as the actual target: if basic text is not found, a missing name is weak evidence.
Build a review list around the recipient
Suppose you need to share a project letter. The recipient may need the project scope but not a private postal address. List those categories first, then search known variants where text recognition allows it. A name may be abbreviated, split across lines or repeated in a signature block.
Look for information that search cannot reliably represent: handwritten notes, photographed badges, signatures, embossed stamps and details inside a background image. A search for a phone number will not necessarily find the same number with different spacing. These are reasons for visual review, not a promise that a different pattern will find everything.
Apply removal to the actual content
Use the tool’s documented redaction method for images and text. Confirm that its scope covers the page image, any OCR layer and the saved result. Do not confuse placing a visual shape with applying redaction. If the tool cannot explain its handling of scanned pages, treat that as an unresolved limitation rather than assuming the output is ready.
Keep a separate original. Review the marked page at a readable zoom before applying the change, checking the edges of each zone and any wrapped value. Then verify that useful context has not been removed unintentionally.
Check the exported copy, not the editing view
Reopen the saved output and review every affected page visually. Use ordinary search and selection as additional checks when the output has a text layer. If the chosen method rasterises pages, searchability and accessibility may change; assess that trade-off for the recipient.
Inspect content outside the page as well. Adobe’s sanitisation guide identifies hidden information as a separate concern. The hidden-information checklist explains how to organise that review.
When to pause the handoff
Pause when a page cannot be read, the exported output differs unexpectedly, the method for images is unclear or the document contains material your tool cannot inspect. Record the limitation without reproducing the sensitive value in a new shared note. A useful workflow makes uncertainty visible; it does not turn a successful OCR job into a certificate of complete anonymisation.