Visible disclosures vs. machine-readable labels

Why visible notices and machine-readable provenance solve different problems and become stronger when used together.

Visible notices reach people directly in the image or its immediate context. They do not require a specialist inspection tool and can remain understandable when a publishing platform strips technical metadata. Their text, position, size and contrast still need to survive scaling and cropping.

Machine-readable labels can carry structured information, signatures and declared editing actions. They require compatible tools and may be lost during transcoding.

What a visible disclosure provides

A label such as “AI generated” or “AI manipulated” is immediately perceivable. It can be embedded in an image, shown within a video or placed in closely associated copy. An embedded overlay usually remains visible after download, while a caption can become separated from the asset.

A visible label is not cryptographic protection. It can be removed, covered or added later. Its strength is direct communication to people.

What machine-readable provenance provides

XMP fields can carry structured disclosure text. C2PA Content Credentials can additionally contain signed assertions, asset bindings and declared actions. Compatible systems can read this information automatically and report its technical status.

The meaning depends on the method used. An unsigned text field has a different technical quality from a valid C2PA signature. A valid signature also does not automatically assess the truth of the depicted content.

Why the layers work together

A robust workflow combines a clear visible statement for the audience with technical provenance for supporting systems:

  1. Choose wording that accurately describes the role of AI.
  2. Place the visible notice so that it survives in the destination channel.
  3. Add machine-readable information when the workflow supports it reliably.
  4. Export the final asset.
  5. Inspect the visible label, metadata and Content Credentials in that exact file.
  6. Repeat the inspection after platform processing.

If one layer is lost, the other may remain useful. They should not be presented as interchangeable: visibility informs people, while machine-readable provenance supports technical inspection.