Redact sensitive data from a PDF, and actually remove it
Drawing a black box over a line hides it from you, not from the computer: that text can usually still be selected, copied and found with a search. Here every redacted page is rebuilt as an image, so there is nothing left underneath the bar to recover. It all happens in your browser, and the file is never uploaded.
Open your PDF in the Redact tool. It stays in the browser, nothing is uploaded.
Drag a box over whatever has to go, or let the scan propose the personal data it recognises.
Check the preview, then download. On redacted pages the covered text is no longer there.
Covering text is not the same as removing it
This is the mix-up that sends supposedly safe documents out of law firms every week. In a PDF the text sits on its own layer, so a black rectangle drawn on top of it, with a PDF reader's highlighter or any editor, is decoration: the letters underneath are still there. Whoever receives the file can select them, copy them, search for them. Most of the redaction failures that made the news started exactly here, with a black box that was only a drawing.
- 1A rectangle added on top leaves the text selectable and copyable.
- 2The black highlighter in most PDF readers does not touch the content either.
- 3Searching the document still finds the words that look deleted.
- 4Real redaction means taking the text off the page, not hiding it from view.
What happens to the page you redact
Worth knowing up front, because it is a deliberate choice with visible consequences. Every page holding at least one redaction is rebuilt as an image: the black box is not drawn over the text, the text is simply no longer there. It is the fail-closed route, the one that shuts rather than hopes, and it costs something. Pages you did not touch are not rebuilt: they stay the document you started with.
- 1On a redacted page the whole text layer goes, not just the part under the box: the rest of that page stops being selectable and searchable too.
- 2On that same page links, form fields and annotations do not survive: measured, a link and a fillable field present before were gone after.
- 3The look is kept: the page reads as it did, it just is not text any more. Visually preserved does not mean structurally intact.
- 4Pages with no redaction are copied from the original and keep their text, links and fields: measured on a two-page document redacted on the first page only.
How the redaction works, step by step
These are the actual steps, in the order you meet them. None of them sends your document to a server: the work happens in your browser's memory, and you can confirm that yourself in the browser's developer tools (F12, Network tab).
- 1Pick your PDF. It opens in the browser rather than being uploaded.
- 2Optionally run the personal data scan, which proposes what it recognised, grouped by type.
- 3Review the proposals and drop the ones you want to keep. You decide what disappears.
- 4Draw boxes by hand over everything else, straight on the document.
- 5Apply the redaction with Redact & review.
- 6Look at the preview before saving. That is where a misplaced box shows up.
- 7Download the file, then reopen it and try to select or search what you redacted.
Redaction, anonymisation, sanitising: same thing?
No, and the difference matters. Blacking out describes the gesture and redaction is its formal name: taking parts out of a document before handing it over, which is what this tool does. Anonymising is an outcome rather than an operation: nobody should be able to identify the person, and that does not follow from the boxes alone, because a role, a date and a place together can be enough. Sanitising is wider still and covers the whole file: text, metadata, attachments, hidden elements. Redaction is therefore one part of anonymising and one part of sanitising, not a synonym for either.
Getting a PDF ready for ChatGPT or Claude
Once a document goes into an AI service it stops depending on you alone. How long it is kept, who can see it and whether it contributes to training depend on the provider, the plan and the account settings: those are things to read in that service's terms, not to assume either way. Stripping the personal data first reduces what you share regardless of the policy: the work happens here, on your own machine, and only an already clean file travels. It does not replace a privacy assessment, it makes one easier.
What the scan recognises, and what it does not
Automatic detection is not magic, and it helps to know where it is exact and where it is only suggesting. Data with a fixed shape is matched by structure; names, addresses and company names are matched by inference, so they always deserve a second look. Anything outside these lists you redact by hand, and that is just as permanent.
Matched by structure
- Email addresses
- Tax codes
- VAT numbers
- IBANs
- Card numbers
- Phone numbers
- IP addresses
- Reference numbers
Proposed, review these
- People's names
- Street addresses
- Company names
- Dates of birth
- Monetary amounts
By hand only
- Health information
- Contract terms and figures
- Photographs and signatures
- Any other area you choose
Who uses it, and on what
- Law firms
- Filings, expert reports and exhibits going to a court or an opposing party, with third-party details taken out.
- HR teams
- CVs and personnel files passed to a hiring manager without exposing contact and identity details.
- Accountants and finance
- Invoices, statements and contracts shared with account numbers and tax identifiers removed.
- Healthcare
- Reports and records used for a second opinion or a study, with nothing left that identifies the patient.
- Public sector
- Documents published under transparency rules, with personal details blacked out.
- Consultants
- A client's material shown to someone else as an example, cleaned of everything that identifies them.
- Anyone using AI tools
- Documents to hand to ChatGPT or Claude after the personal data has been removed locally.
What you can verify yourself
- The document is never uploaded.
- Open the developer tools (F12), Network tab, and work on your PDF: no request carries the file out. It is also one of the automated checks that must pass before any code change ships.
- The redacted text is no longer in the file.
- Reopen the downloaded PDF and try to select or search what you covered. On redacted pages there is nothing left to find.
- Pages you did not redact are copied over unchanged.
- Measured: in a two-page document with one page redacted, the untouched page keeps its text layer, its link and its form field, while the redacted page keeps none of them. The automated checks also confirm on the pixels that a word inside the redacted area is gone and one just outside it is still visible.
- If it cannot guarantee the result, it stops.
- A protected or unreadable PDF is refused with a clear message, rather than handed back as a file that looks fine and is not.
The limits, up front
A tool meant to protect data should also say what it does not do. These are the cases that need care, or one extra step.
- Password-protected or encrypted PDFs
- They are not opened: the tool says so and stops. The protection has to be removed first, by whoever holds the password.
- Scans and photographed pages
- When a page is an image with no text layer, the scan tries OCR, but not every page is read and the reading can be wrong. Draw the box by hand there. It works exactly as it does on text.
- Malformed documents
- A damaged file is refused rather than processed halfway.
- Metadata is a separate job
- Author, software and dates stay in the file's properties. Removing those is the Metadata tool, not this one.
- A human still has to look
- The automatic proposals need reading: an unusually written name can be missed, and something you wanted to keep can be flagged. The last word is yours, which is what the preview is for.
FAQ
How do I redact text in a PDF?
Open the PDF in the Redact tool, drag a box over the text that has to go, press Redact & review and download the file. On redacted pages the covered text is no longer part of the document.
Does a black box really delete the text?
No, and that is the whole point. A rectangle drawn on top is decoration: the text stays underneath, selectable and copyable. Here the redacted page is rebuilt as an image, so the covered text is not in the file any more.
Can redacted text still be copied?
Not on the pages you redacted: there is no text layer left to select, and that holds for the whole page, not only the part under the box. Pages you did not touch stay as they were, with their text still copyable.
What else does a redacted page lose, besides the covered text?
It becomes an image, so it loses the entire text layer and with it any links, form fields and annotations on that page: measured, not assumed. The appearance is unchanged. Pages without redactions are not rebuilt and keep everything.
Can I redact personal data without uploading the PDF?
Yes, and it is the only way this works: the document stays in your browser and is never sent to a server. You can confirm it in the Network tab of your browser's developer tools.
How do I remove names, addresses, emails or account numbers?
Draw the boxes by hand, or run the automatic scan: emails, tax codes, VAT numbers, IBANs, card and phone numbers are matched by structure, while names and addresses are proposed and should be reviewed before you apply.
Can I anonymise a PDF before sending it to ChatGPT?
Yes, and it is one of the reasons this tool exists. You strip the data on your own machine and only upload the already clean file: the original never leaves your browser.
Is the redaction permanent?
On the file you download, yes, for the pages you redacted: the covered text cannot be recovered because it is not there. Your original file is not modified, since what you download is a new copy.
Does it work on scanned PDFs?
Yes, by drawing the boxes by hand: on a page that is an image, redaction behaves exactly as it does on text. The automatic scan will try OCR, but it does not read every page and can misread, so check those by eye.
Does it work on password-protected PDFs?
No. An encrypted document is not opened, and the tool says so plainly instead of returning an incomplete file. The protection has to be removed first.
Does redacting text also remove the metadata?
No, and it is worth knowing: author, software and dates stay in the file properties. Removing those is the separate Metadata tool.
How do I check the result?
There is a preview before you save, showing the page as it will be. After downloading you can reopen the file and search for what you redacted: not finding it is the confirmation.
What is the difference between redacting, blacking out, anonymising and sanitising?
Blacking out describes the gesture on the page and redacting is its formal name; anonymising looks at the outcome, that nobody can identify the person, which sometimes means removing details that look harmless on their own; sanitising usually includes the wider clean-up such as metadata. They ask the tool for the same thing.
From here you can also
- Open the Redact tool and work on your own PDF
- Check whether a PDF still carries risky data or fake redactions
- See how the automatic personal data scan works
- Get a document ready before handing it to ChatGPT
- Strip the metadata, which redaction does not touch
- Password-protect the redacted file
- Understand why everything runs in the browser and nothing is uploaded
More local-first PDF tools
Part of the Secure Local Suite: manage PDFs and anonymize sensitive text, all in your browser.