Redact a PDF before you hand it to an AI
To redact a PDF is to remove the information a reader has no need to know: a name, an address, a case number, an IBAN. A PDF file is an assembly of objects, fonts and metadata, so redacting it properly means dealing with each of them.
With Biffer, the text is rewritten in place, on your computer, offline. This page explains what redaction means, what precautions a PDF calls for, how our redaction software works, and how it handles scanned PDFs, using character recognition and blacked-out pixels.
Redact, black out, strike out: what are we talking about?
Key points
- Redacting a document means removing certain information from the file itself, permanently.
- A PDF carries text in its content stream, its secondary objects and its metadata: Biffer acts on all three.
- Biffer replaces each item with an alias or a mask directly in the text stream, keeping the layout.
- Everything happens offline, on your computer; scanned PDFs (images) are read by OCR and redacted by blacking out pixels, permanently.
Our product name comes from French: "biffer" means to strike out, the word clerks and registrars use when they cross out an entry with a line so that it no longer counts. Striking out, blacking out and redacting are three shades of the same gesture: crossing out, covering in black, withdrawing from view. In every case, the goal is that the removed data stays out of the recipient's reach, whether they read the document or try to reconstruct it.
Redaction was long a matter of paper: a stroke of marker, a photocopy, and the entry was gone. Redacting a file today means changing a data structure you cannot see on screen. A redaction tool therefore works on the content of the file, beyond its appearance.
We use "redact a PDF document" in a broad sense here: removing personal or confidential data before sending the file to a colleague, to a public body or, more and more often, to an artificial intelligence. Redacting the file before you drop it into an online assistant keeps the identifying part on your computer; we cover this on our page on personal data and AI. The differences between masking, pseudonymization and anonymization are covered on our page redact a document.
What a PDF needs to be redacted correctly
A PDF looks like a printed page, but it follows a logic of its own. The format describes a page as a list of objects: text fragments positioned by coordinates, embedded fonts, images, annotations, forms, and a metadata dictionary. A name can be split into three fragments, placed on a different layer from the rest of the paragraph, or repeated in an invisible form field.
This structure explains what a complete redaction has to cover:
- The text stream. Biffer replaces the data in the file's actual text: a selection, a copy and paste or a search returns the alias or the mask.
- Text spread across objects. The replacement applies to every copy of the data: bookmarks, form fields, annotations and hidden layers.
- Embedded fonts. When a font is embedded as a subset, it only contains the glyphs in use; the preview shows you how each alias renders before you save.
- Metadata. Author, company, title, keywords, application name, sometimes the path of the original file: all of this is stored separately, and Biffer clears these fields.
Redacting a PDF document properly therefore means acting at three levels: the text stream, the secondary objects (annotations, forms, bookmarks) and the metadata. That is what redaction software designed for this format does, and that is what we built with Biffer.
How Biffer rewrites the text in place
Our approach comes down to one word: rewrite. Biffer reads the PDF's text stream, finds the data to remove, then replaces each occurrence with an alias or a mask directly in the file's objects. The document keeps its pages, fonts, tables and positions; only the affected strings change. The process has four stages.
1. Local analysis
The file is opened on your computer. Biffer looks for people, email addresses, phone numbers, postal addresses, French company numbers (SIREN and SIRET), French social security numbers, IBANs, license plates and dates, as well as the fields specific to your profession. Every fragment of text stays on the machine.
2. Highlighted preview
Each detected item appears highlighted in a preview. You untick false positives and add whatever is missing. A selection you add is applied to all its occurrences at once, which also covers the second mention of a name on page 14.
3. Rewriting
Depending on the policy chosen for each data type, the value is replaced by a stable alias (PERSONNE_001, EMAIL_002), by a mask ([PERSONNE]) or kept. The text is rewritten in place in the PDF's objects, the layout is preserved, and the metadata (author, company, comment authors) is cleared.
4. Optional mapping table
If you need to restore the original values in the document an AI or a third party sends back to you, Biffer can save a mapping table, encrypted (Argon2id and AES-256-GCM) or in plain text. Without a table, the operation is permanent.
Each step is described in detail on the features page. Rewriting in place produces a PDF that is readable, consistent (the same person carries the same alias throughout) and clean down to its secondary objects.
Scanned PDFs, review and responsibility: you stay in control
Biffer adapts its processing to the type of file, and tells you so. On a PDF that contains text, it rewrites the text stream with aliases and keeps the layout. On a scanned PDF, meaning a file made only of page images, it works differently: built-in character recognition, run offline, rebuilds the text, detection is applied, then the identified areas are redacted by blacking out the image's pixels. This redaction is permanent: the blacked-out area truly disappears from the image, which puts the data out of the recipient's reach. Quality follows that of the scan, and the preview lets you review carefully; when the original exists (a Word file, an export from your business software), it also gives you aliases and the mapping table.
Second point, common to any automatic detection: the highlighted preview leaves you in charge of false positives and additions. A street name that looks like a person's name is unticked in one click, and a nickname that only context reveals is added to all its occurrences. Review is part of the method: Biffer prepares, you approve, and you have the final say on the document you send.
Third point, and the decisive one: redacting a file makes full sense when the file stays with you until redaction is finished. That is why we designed Biffer as offline software, for Mac and PC (Windows). Romain Rissoan, a trainer and consultant based in Lyon, France, designed it for this use. The redacted PDF can then be submitted to ChatGPT, Claude or Gemini, since the identifying part of the file stayed on your computer; we describe this use case on the page anonymize before ChatGPT.
Finally, a legal reminder: the GDPR defines personal data broadly (Article 4 of the regulation), and replacing a name with an alias counts as pseudonymization as long as a mapping table exists. We explain this distinction on the page anonymize a document.
Frequently asked questions
Is drawing a black box over a PDF enough to redact it?
A complete redaction removes the text from the file itself. A black rectangle added with an annotation tool is a graphic object placed over the text, which remains in the file and in the metadata (author, company). Biffer rewrites the text in place, applies the replacement to every occurrence and clears the metadata. You can check the result by opening the PDF produced and searching for the original data: it is gone.
Can Biffer redact a scanned PDF?
Yes. A scanned PDF is a series of page images. Biffer runs built-in character recognition on it, offline on your computer, then the same detection as on a text file. A name printed in an image is redacted by blacking out the pixels of the identified areas directly: the image is truly modified and the blacked-out area is permanently empty. This redaction is permanent, with no mapping table. The preview lets you review carefully, which is valuable on a faint or skewed scan. If you have the original (a Word file, a text export from your business software), that version also gives you aliases and reversibility.
Is the PDF layout preserved after redaction?
Yes, because the text is rewritten inside the existing objects rather than the document being rebuilt. Positions, fonts, tables and page breaks stay as in the original. An alias such as PERSONNE_001 has its own length, which can slightly change how a line fills; the document remains readable and consistent. With a font embedded as a subset, some characters may render differently; the preview shows you this before you save.
How much does it cost to redact documents with Biffer?
Biffer is free for 5 files per month, with every feature included and no time limit: the first time you open it, you confirm your email address through the link you receive, once per computer, and the software gets its key and then works offline. Beyond that volume, Biffer costs €10 excl. VAT (€12 incl. VAT) per computer, paid once (Mac or PC), a lifetime license for the version purchased, one license per computer, unlimited files, and you can assign several licenses to colleagues from your account. Details are on the pricing page, and the download is on the download page.
Your business documents anonymized, ready to be published or handed to an AI tool.
Biffer runs on your computer without sending a single file. Free for five files a month, unlimited for €10 excl. VAT per computer.
100% offline · Free: 5 files a month · No file sent