Start with the question you have
| If you are asking… | Read this | Then check a file |
|---|---|---|
| "Was this image or video made with AI, and can I verify it?" | Content Credentials (C2PA) guide | Metadata viewer |
| "Why did YouTube put an AI label on my video, and can I appeal it?" | Why is my video labeled AI? | Metadata viewer |
| "I blacked out text in a PDF — is the text really gone?" | PDF redaction check | Redaction checker |
| "What does the Apple Reference Image badge prove about a photo?" | Apple Reference Image guide | Metadata viewer |
The guides
Content Credentials (C2PA)
What a signed provenance manifest records, where it lives in each file type, the difference between "detected" and "verified", and the 2026 rules that made it matter.
Why is my video labeled AI?
The three routes to a platform AI label, which labels cannot be removed, how to check what your own file carries, and why stripping metadata is not a fix.
PDF redaction check
Why text under a black box often stays selectable, the 2026 audit numbers, a 60-second self-check, and what a real redaction has to remove.
Apple Reference Image
What the iPhone capture signature proves, what it does not, and how it differs from C2PA and SynthID.
What the guides have in common
Every guide answers the same kind of question: what does this signal actually prove? Metadata can show that a file was edited, that a device signed a capture, or that a tool wrote a claim. It rarely proves that the scene, the person or the event was real.
Two distinctions come up in all four guides. Presence is not verification: a manifest that exists can still be invalid, expired or unrelated to the pixels. And absence is not guilt: most real files carry no provenance data at all.
That is also why Docfy works locally. Reading metadata should not require handing the file to a server — and checking a credential should not upload the evidence to a third party.