How Privio redacts
In short
- Privio finds your search terms in the PDF’s text layer, renders every page to an image, burns black boxes into the pixels over the matched words, and builds a new PDF from those images. Nothing from the original file’s structure is copied into the new one.
- It can only redact what it finds. Text inside images is not searched. Pages Privio could not check are listed on the result screen, and it asks you to confirm before you export; anything it could not check on them stays visible. The known limitations are listed below.
- Check every redacted file before you share it: how to check.
- Privio’s redaction has not been audited or certified by an independent third party. This note describes the method as implemented in the app’s source code; it is not a test report.
- This page covers the Redact tool for PDFs. Redact photo is a separate tool: you draw the boxes yourself on a photo or screenshot (privacy policy, section 4.7).
1. The method, step by step
- Search terms. You enter one or more terms. Matching ignores upper and lower case unless you turn on case-sensitive matching; without it, the Turkish letters I, ı, İ and i all count as the same letter.
- Working copy. Privio first makes a working copy of the document with each page’s rotation cleared, so that text positions and the rendered image line up; the rotation is put back on the output pages. Form fields and annotations are left as they are: they are not searched (see the limitations).
- Finding matches. Privio reads each page’s text layer and joins its words into one continuous run of text. Spaces, line breaks, soft hyphens, invisible formatting characters such as direction marks, and quotes, apostrophes, hyphens and dashes in every form are ignored; typographic ligatures such as “fi” are expanded and full-width characters narrowed; and unless you turn on case-sensitive matching, upper and lower case, accents (é, ü, ç and the like), ß and the Greek final sigma are folded as well. So “O’Brien” also matches “OBrien”, “123-45-6789” also matches “123456789”, and a word broken across two lines is still found. Privio looks for each term, prepared the same way, in that run. For every match it takes the bounding boxes of the words the match touches, as reported by the text layer, and pads them slightly (1 point on the left, right and top, 2 points below). Where the text layer gives a line without word boxes, the whole line counts as one word. In a PDF made searchable with Privio’s OCR PDF tool from version 1.1.5 on (its creator field reads “Privio OCR”, and the mark survives Merge, Split, Delete pages and Rotate), the word positions are approximate, so every line with a match is blacked out across the full width of the page.
- Pages that cannot be checked. Privio lists a page as unchecked when it has no searchable text (usually a scanned page, or a page whose text is only inside images); when its text layer covers only part of what is printed on it — for example a scan that carries a Bates stamp or a court header, or a photo of a document — which Privio detects by comparing the dark ink inside and outside the text layer’s words on a small rendering of the page; when its text cannot be extracted; and, on iOS, when its crop box differs from its full size. Privio runs this check even when no term matches. The result screen names every unchecked page and Privio asks you to confirm before you export; anything it could not check on those pages stays visible.
- No match, no change. If no term matches, Privio tells you so, names any page it could not check, and produces no new file; your document is unchanged.
- Rendering every page. Every page — not only the pages with matches — is rendered to a JPEG image about 2,000 pixels on its long edge (roughly 180 dpi for a US Letter page; JPEG quality 88), using the platform’s PDF renderer (Apple Core Graphics on iOS, Android PdfRenderer on Android). If a page cannot be rendered, or its image cannot be decoded, the run stops with an error instead of producing a file without it.
- Burning. On pages with matches, Privio paints solid black rectangles into the image pixels over the matched word boxes and re-encodes the image. On iOS pages whose crop box differs from their full size, nothing is blacked out: a box would land in the wrong place, so the page is drawn as it is and listed as unchecked. If every match is on such pages, Privio says that nothing could be redacted. The result screen shows how many redactions it made on each page.
- Rebuilding. Privio builds a brand-new PDF from the page images, with each page’s original size and rotation. The original file’s text layer, fonts, metadata, attachments, form fields, links, bookmarks and earlier revisions are not copied into it.
- OCR-back (“Keep non-redacted text searchable”, on by default). Privio runs on-device text recognition — Apple Vision on iOS, Google ML Kit on Android — on each redacted page image and adds the recognised text as an invisible text layer over the image, so the remaining text can be searched and copied. The recognizer only sees the burned images, so the black boxes contribute no text. Rotated pages are left image-only, without an OCR-back layer, because their recognised text would land in the wrong place. With OCR-back off, the new PDF contains page images only and has no text layer at all.
Everything above runs on your device; no page image or text is uploaded. (On Android, ML Kit sends Google its own diagnostic metadata, described in the privacy policy.) The free version and Pro run the same pipeline. Pro exports the result as it is. In version 2.0, the first file you export without Pro, from any tool, is also exported as it is; every later export without Pro adds the Privio watermark on top of every page, after redaction (see the last item under Known limitations). In versions 1.1.6 and 1.1.7, every export without Pro adds that watermark, which reads “PDF Privio” there; in 1.1.5, exporting requires Pro. The page images are written to the app’s private temporary folder during the run and deleted when it ends, whether it finishes, fails or is cancelled; if the app crashes or is force-quit mid-run, they are deleted the next time the app starts. The on-device audit log records the number of search terms, matches, affected pages and unchecked pages — never the terms themselves.
2. Known limitations
- Only the text layer is searched. Text that is part of an image — a scanned page, a photo, a signature or a chart — is not found. Run a scanned PDF through the OCR PDF tool first; matches then depend on how accurately the OCR read the page.
- Form fields and annotations. Values typed into form fields, and annotations such as comments, text boxes and stamps, are not part of the page’s text layer, so they are not searched. On iOS the page renderer does not draw them either, so they are missing from the redacted copy; on Android this has not been checked. Fill in form fields after redacting, or flatten the form in another app first and then check the output.
- Short terms can match across words. Because spaces, quotes and dashes are ignored, a term can match letters that span two words: searching for “ten” also covers “late night”. Privio errs towards covering too much rather than too little; use longer terms where you can.
- Other character variants. Apart from the folds above, the term must match the characters stored in the text layer: look-alike characters from other alphabets, or a text layer that stores characters in an unexpected way, can prevent a match. With case-sensitive matching on, accents and case must match exactly.
- Detecting scans behind a stamp is a heuristic. It compares dark ink inside and outside the text layer’s words; a very faint or light-coloured scan may not be flagged. Check scanned pages yourself.
- Whole words are covered. A match inside a longer word blacks out the whole word: searching for “Ann” also covers “Annual” and “Planning”.
- Boxes follow the text layer’s positions. If a PDF’s text layer does not line up exactly with the visible letters — this happens in some files — a box can miss part of a word.
- Pages with a crop box (iOS). On iOS, a page whose visible area (its crop box) differs from its full size is listed as unchecked and is not blacked out at all. Redact such pages another way, or remove the crop box first.
- Scans made searchable with Privio before 1.1.5. Earlier versions of the OCR PDF tool did not mark their output, and their invisible text layer’s word boxes may not line up with the visible words, so black boxes can be offset. Run OCR PDF on the original scan again in 1.1.5 or later before redacting.
- Rotated pages are not searchable in the output. With OCR-back on, rotated pages get no recognised text, so searching the redacted file finds nothing on them. Check rotated pages by eye.
- OCR-back languages. Text recognition for OCR-back is set up for English and Turkish; text in other languages may be recognised less accurately. This affects only searching the output, not what was redacted.
- The output is an image-based PDF. It is usually larger than the original. Its only selectable text is the OCR-back layer, which can contain recognition errors (and, on a free copy, the watermark). Form fields, links and bookmarks no longer work.
- A free copy carries a watermark. After your first free export from any tool, a redacted file exported without Pro gets the Privio watermark after redaction: a large diagonal “Privio” in black at 28% opacity and a footer line on a mostly opaque white strip, about 8 to 12 pt tall, at the bottom centre of each page. (In versions 1.1.6 and 1.1.7 it reads “PDF Privio” and is added to every export without Pro.) The watermark’s words are text, so they can be selected and found by a search. The strip hides what lies under it without removing it from the file, so check the bottom edge of each page in the original as well. A Pro export has no watermark.
3. How to check a redacted file
- Compare page counts between the original and the redacted file.
- Search the exported file for every term, including variants: different case, spacing, line breaks, abbreviations and possessives. Keep OCR-back on for this, or run the exported file through OCR first — with OCR-back off the file has no text apart from a free copy’s watermark, so a search finds nothing either way. Rotated pages have no OCR-back text, so check them by eye.
- Look at every page at a zoom where you can read the text. Check that each box covers the whole term, and look for anything a search cannot see: images, stamps, handwriting, headers and footers, comments and form fields. Take extra care with the pages listed as unchecked, with filled-in forms and with scanned pages. On a free copy, the watermark’s footer strip covers the bottom edge of each page: check that edge in the original.
- Open the file in a second viewer (for example Adobe Acrobat or Apple Preview) and try to select and copy text around a box.
- For important productions, treat Privio’s redaction as one step and follow your own quality-control process.
4. Changes in 1.1.5
- Pages that could not be checked — including scans whose only text is a stamp or header — are listed on the result screen, and Privio asks you to confirm before you export.
- PDFs made searchable with Privio’s OCR PDF tool are marked, their text layer is fitted to each line’s width, and a redaction of such a PDF blacks out every matching line across the page.
- Matches that run across a line break or a line-ending hyphen, matches whose spacing differs from your search term, matches that differ in quotes, apostrophes, hyphens or dashes, matches interrupted by invisible formatting characters and, without case-sensitive matching, matches that differ only in accents are found. On iOS, the search field no longer turns typed quotes and dashes into curly ones.
- Rotated pages are redacted in the right place and keep their rotation in the output. (In earlier versions the boxes on pages rotated by 90, 180 or 270 degrees landed in the wrong place.)
- A page that cannot be rendered stops the run with an error. (In earlier versions it was left out of the output.)
- On iOS, pages whose crop box differs from their full size are listed as unchecked instead of being blacked out in the wrong place.
- The result screen shows how many redactions were made on each page, and says so when matches were found but none could be blacked out.
- The new PDF keeps each page’s original size and orientation, with OCR-back on or off. (Earlier versions made each OCR-back page as large as its 2,000-pixel image — a US Letter page became about 21.5 × 27.8 inches — and, with OCR-back off, squeezed landscape pages into portrait pages.)
- Temporary files left by an interrupted run, including copies made by the file picker, are deleted when the app starts.
5. Changes in 1.1.6
- A redacted file can be exported without Pro; it then carries the PDF Privio watermark, added after redaction (see Known limitations). In 1.1.5, exporting a redacted file requires Pro.
6. Changes in 2.0
- The app is now called Privio. The redaction method on this page has not changed.
- The first file you export without Pro, from any tool, has no watermark. Later exports without Pro carry the Privio watermark.