A PDF can remember who wrote itafter the words have changed.
A PDF keeps an info dictionary and often an XMP packet: author, creator software, dates, and sometimes a C2PA Content Credential. Deleting those keys in place is not enough — leftover objects stay in the bytes. Gifi clears the fields and rebuilds the file so the old metadata is not sitting in an unused stream.
Reviewed by Can Balkaya
Published
Process
How to verify the job
01
Inspect
Preview the document info, XMP, and credential markers before the file leaves your browser.
02
Rebuild
The worker clears the metadata and rewrites the PDF structure so unused objects are not left behind.
03
Reinspect
Open or scan the returned PDF and compare its action list with the original findings.
Scope
Supported and unsupported surfaces
| Surface | Gifi action | Verification limit |
|---|---|---|
| Named metadata or invisible Unicode | Inspect, remove, and report named actions | Reinspect the returned bytes independently |
| Statistical patterns in ordinary words | Rewrite the prose | No certificate against a private detector |
| Visible logos, pixels, audio, or video | Unsupported | No removal claim |
Where a PDF hides its history
The document info dictionary is the obvious place: Title, Author, Creator, Producer, CreationDate, ModDate. Many files also carry an XMP stream attached to the catalogue, which repeats the same facts in XML and can hold a C2PA manifest.
PDFs are a pile of objects with a trailer. If you only rewrite the dictionary and leave the old object in the file, a determined reader can still recover it. That is why a serious strip has to rebuild the structure, not just edit a field.
Why this job leaves the browser
PNG, JPEG, WebP, SVG, HTML, Markdown, DOCX and ODT can be cleaned in-process. A trustworthy PDF strip needs a tool that can linearize and drop unused objects. Gifi sends PDFs to a worker that can do that. One credit, same as any other file.
A format we cannot fully clean is refused. A PDF you believe is clean, but is not, is worse than one we declined.
What this does not do
It does not rewrite the sentences on the page. Invisible Unicode in the text layer of a PDF is a different problem from the info dictionary. It does not remove a watermark printed into the appearance of the page. And if a C2PA soft-binding fingerprint was taken, a remote manifest can still re-attach.
Questions this usually raises
- Is the title removed?
- Document info and XMP are cleared. If you need a title to remain, say so in the download name — do not rely on the PDF’s Author or Title fields after a strip.
- Can I recover the old metadata after a clean?
- Not from the cleaned file, if the rebuild finished. That is the point of the structural pass. We cannot speak to copies you still have, or to a remote C2PA registry.
- Does this affect the look of the pages?
- The strip targets metadata and unused objects, not the drawing of the page. If a job would have to guess, it fails and the credit is refunded.