How to Redact a PDF Safely and Verify Hidden Content Is Gone

Blur and pixelation are visual effects, not reliable removal methods. A solid overlay can also be unsafe while it remains editable. FiloTool's Redact PDF Secure Flatten workflow rasterizes each affected page, draws solid redactions into the pixels and builds a replacement page; unaffected pages are preserved where possible.

Blur and pixelation transform the pixels you can see; they do not remove the information those pixels encode. Pixelation replaces a region with a grid of averaged blocks, which is a deterministic function of the original. When the hidden value comes from a small, predictable set—a six-digit account number, a date, a name from a known list—an attacker can render candidate values through the same transformation and compare the output until one matches. Blur is a convolution, and approximate inverses for it are well studied.

There is a second, simpler failure. If the blur or pixelation is applied as an overlay object rather than baked into replacement pixels, the untouched content is still sitting underneath it in the file. FiloTool's Pixelate Image tool exists for creative and preview purposes on images, and is deliberately not offered as a document redaction method.

Text search locates exact matches inside the embedded text layer, with case-sensitivity and whole-word controls. Pattern suggestions cover four shapes: email addresses, digit runs that look like phone numbers, IPv4-style addresses, and http or https URLs. These are heuristics, and they are deliberately loose—the phone pattern matches most separated digit runs of nine or more characters, so it will also flag order numbers and reference codes, and the IPv4 pattern accepts impossible values such as 999.999.999.999.

Matched values are masked in the review list, showing the first and last two characters with the middle replaced by dots, so that reviewing a long list of hits does not put the sensitive values back on your screen. Scanned pages have no embedded text layer for any of this to search, so mark those regions by drawing areas over them instead.

Only pages carrying at least one redaction area are rebuilt. Each of those pages is rendered to a canvas at 1.75x scale on Standard quality or 2.5x on High, your redaction rectangles are filled into those pixels as solid colour, and any replacement label you added is drawn and clipped to the rectangle. The finished canvas is encoded as a JPEG—quality 0.88 on Standard, 0.94 on High—and embedded as a single full-page image at the original page dimensions.

The consequence is that a rebuilt page has no text layer, no annotations, no form fields and no links, because it is now one flat image. Pages you did not mark are copied across untouched and keep their original text, vectors and searchability. That preserves quality where nothing needed removing, and it creates the limitation described further down this page.

Work from a copy and define what must be removed before drawing. Include enough surrounding area to cover every glyph, shadow or edge. Secure Flatten invalidates existing digital signatures and removes searchable text, links, forms and annotations from affected pages.

Secure Flatten does not simply save and hand over the result. After writing the new document it re-reads its own output and fails loudly rather than returning a file that did not meet its conditions. It confirms the saved bytes begin with a valid %PDF- signature, reopens the document and confirms the page count still matches the source, then re-parses every page it redacted and checks two things: that no extractable text remains on that page, and that no annotations remain on it.

If any of those checks fail you receive an error message instead of a download. That is a useful floor, and it is not a guarantee. The check only inspects pages that carried a redaction area, so it can confirm that the pages you marked were flattened correctly—it cannot know about a page you forgot to mark.

Verify the downloaded file, not the editor preview. The preview is a rendering of your intent; the download is the artifact you will actually send. These three checks are independent, which is the point—each one can catch a failure the others miss.

The realistic risk is not a defeated redaction. It is a value you never marked. Because unmarked pages are copied through with their text layer intact, and because the built-in verification only inspects the pages you redacted, a sensitive string sitting on page 34 of a 40-page contract is fully extractable from the exported file and nothing in the workflow will complain.

So run the search across the whole exported document rather than page by page, and search for the value itself rather than trusting that you found every occurrence visually. Repeated headers, footers, page-level watermarks, appendices and signature blocks are the usual places a second copy hides.

Flatten PDF converts supported form controls into ordinary page content. It is useful when form values must stop being editable, but flattening a form is not the same as securely removing sensitive pixels or text. Use the dedicated redaction workflow for content removal.

Redact PDF accepts one file at a time, up to 100 MB and between 1 and 500 pages. A rendered page is capped at 40 million pixels, which is why a very large page can fail on High quality and succeed on Standard—the lower scale factor produces a smaller canvas.

Rebuilding pages as images has costs worth planning for. Redacted pages are no longer selectable or searchable, which also means they are no longer readable by a screen reader; if the document must remain accessible, redact the smallest possible set of pages. The file can also grow, because a JPEG of a text page is often larger than the text instructions that drew it, so compare the exported size against the original before sending.

Text search depends on the embedded text layer, so unusual encodings and subset-embedded fonts can defeat it, and pattern suggestions can be both over- and under-inclusive. Existing digital signatures are invalidated, since the pages they signed no longer exist in their original form. And no browser workflow can reach copies that already left your machine.