A black rectangle drawn over a name hides it from the eye and leaves it in the file — selectable, copyable, extractable. Real redaction removes the underlying text object so there is nothing left to recover. Then it verifies the exported file.
Standard text and metadata redaction is free. Image-based removal uses AI and is metered.
Most people redact by drawing a filled rectangle over sensitive text. It looks right on screen and fails completely in practice.
A rectangle is a drawing on top of the page. The text object underneath is untouched. Anyone receiving the file can select it, copy it into a document, or extract it with a parser in seconds. This has caused real disclosure incidents in legal filings.
The text object itself is removed from the content stream, along with its font resources and any structured references to it. What remains is a page that simply does not contain the words any more.
Open the redacted file, select all, and paste into a plain text editor. If the sensitive string appears, the file was covered, not redacted. You can also extract the text with a command-line parser:
pdftotext redacted.pdf - | grep -i "confidential"
If that returns anything, the content is still there.
Redaction touches more than the visible sentence. These are the places sensitive content hides and the places a proper removal has to reach.
Names, account numbers, addresses, clause text. Removed as objects from the content stream, including text split across runs and text inside form fields.
Author, organisation, title, keywords and timestamps — the fields that identify people and companies even when the page is clean.
Review notes, highlights and replies. Deleted as objects rather than hidden, so they cannot be toggled back into view.
White-on-white text, content positioned outside the visible page box, and text left behind by earlier edits.
Embedded files riding inside the document are removed, not just unlinked from the page.
Visible watermarks in scanned pages sit in the image rather than the text layer. Removing them needs image-level processing, which is where AI-assisted removal comes in.
Removing the text is one step. What people actually need is to find what should be removed, remove it, mark what happened, and hand over a file they can stand behind.
Scan the document for sensitive content and hidden data before deciding what to redact.
Remove the text objects and metadata permanently, at object level.
Mark the redaction with a reason or approval note where the recipient needs context.
Re-parse the output to confirm nothing remains, then export to PDF or the format you need.
When done at object level, yes — the text no longer exists in the file. This is different from covering, where the text remains underneath a drawn shape.
Scanned pages contain no text layer, so there is nothing to remove until OCR runs. Once the page is recognised, text can be redacted; watermarks and stamps inside the image itself require image-level processing.
You choose. You can leave a black bar, a white gap, or a redaction label with a reason. The removal happens either way — the mark is only what the reader sees.
Text and metadata redaction runs locally in your browser. Image-based watermark removal needs AI processing and sends the file, which is stated before you confirm.
Export, then re-scan the output. Verification is built into the workflow rather than left to you — a redaction you cannot verify is a redaction you cannot rely on.
Standard text and metadata redaction is free. AI-assisted removal of image watermarks is metered under a plan.
Object-level removal with verification built in — not a rectangle over words that are still there.
Redact a PDF