A black box is not redaction
Most tools draw a rectangle on top of the words. The words are still in the file — select the area, and they come out. This one shows you how many characters are left before you export.
What this actually does to the file
The pages you mark are re-rendered as images at 144 DPI, with the box filled black. Once a page is an image, its text layer no longer exists, so nothing on that page can be selected, searched or extracted — including the text you did not mark. Pages you leave untouched are copied as-is and stay selectable. That is the whole trade: the marked pages stop being text.
The alternative you will see elsewhere is editing the content stream to delete only the targeted text. It keeps the rest of the page searchable, but a missed duplicate of the same string elsewhere in the file — in a header, a comment, an attachment — stays readable. We choose the blunter method and tell you that it is blunt.
Output files are larger, because a page that used to be a few kilobytes of text becomes a JPEG. There is no way around that.
Healthcare documents
If you work with patient records, the practical question is usually whether the file ever leaves the device. Here it does not: the PDF is opened by the browser, rendered by the browser, and saved by the browser. No server receives it, so there is no transmission of PHI and no third party that would need a Business Associate Agreement to handle it.
That is a statement about architecture, not a certification. HIPAA also expects access controls, audit logs and encryption at rest, and a browser tool produces none of those. If your compliance process requires a signed BAA or an audit trail, this tool does not satisfy it and should not be presented as if it does.
What this cannot promise. A clean check here means the marked pages carry no extractable text layer. It does not mean the document is suitable for a legal filing, and it says nothing about text that lives in annotations, attachments or the XMP metadata block — that is a separate channel, handled on the remove metadata page.
Scans are a different case, and the check says so. A scanned PDF has no text layer to begin with. Counting characters there returns zero no matter what you draw, and a zero that came from a missing text layer is not evidence that anything was removed. The check reads the page's text layer first and tells you which of the two zeros you are looking at: nothing to copy because the box missed the words, or nothing to copy because the page never had words. The competition does not draw that line — most tools rasterize and report success either way.