Interoperability
Content provenance in the AI era is not one standard — it is a layered set of them, each answering a different question. Made Mark fills the slot none of the others claim: human-readable authorship disclosure. Everything else, it composes with.
Four questions, four layers, one carrier
A single image or web page can carry all of these at once. They don't compete for the same job — they stack.
Is this provenance record authentic and unaltered?
A cryptographically signed manifest — the secure envelope. It can wrap the authorship claim and prove who asserted it.
Who — or what — made this?
Three plain-language marks: Human Made · Human Designed, AI Made · AI Made. A controlled vocabulary a person reads without tooling.
mm: → https://mademark.org/ns/
What may you do with it?
Licenses and preference signals — usage and AI-training terms. Beside authorship, never overlapping it.
Who to credit, contact, bill?
The established photo-metadata fields. Made Mark adds the human-versus-AI axis they never had.
On the open web
<script type="application/ld+json"> in the head
Inside the file
APP1 / iTXt / <metadata> — travels with the bytes
How Made Mark relates to each
The test of a standard is whether implementers can adopt it without giving anything up. Made Mark is designed to be additive to every row below.
| Standard | Answers | Carrier | Trust model | Made Mark's relationship |
|---|---|---|---|---|
| C2PAContent Credentials | Is this provenance record authentic and unaltered? | Signed manifest in the file | Cryptographic — X.509 signing | Rides inside it. The Made Mark statement is carried as a C2PA assertion; C2PA proves it hasn't been tampered with, Made Mark makes the manifest legible to a human. |
| Schema.orgstructured data | What does the open web know about this page? | JSON-LD in the page head | Declarative — crawler-trusted | Is a carrier for it. The mark expresses as a CreativeWork extension, making authorship machine-discoverable at web scale. |
| Creative Commonslicenses + signals | What may you do with it — including AI training? | XMP/RDF, HTML, dataset signals | Declarative — legal + preference | Sits beside it. CC expresses usage; Made Mark expresses authorship. Non-overlapping namespaces on the same carrier — a file can hold both. |
| IPTC / XMP Rightsphoto metadata | Who to credit, contact, and bill? | XMP/RDF in the file | Declarative | Neighbors it. Made Mark coexists with creator, credit, and rights fields and adds the human-versus-AI authorship axis they lack. |
One file, every layer
Because the layers stack rather than compete, a single image's metadata can carry all of them together — each in its own namespace, wrapped by a C2PA manifest that makes the whole record tamper-evident.
<!-- XMP / RDF inside portrait.jpg -->
<rdf:Description rdf:about=""
xmlns:mm="https://mademark.org/ns/"
xmlns:cc="http://creativecommons.org/ns#"
xmlns:dc="http://purl.org/dc/elements/1.1/">
<!-- authorship — Made Mark -->
<mm:mark>https://mademark.org/marks/human-designed-ai-made</mm:mark>
<mm:model>Claude Opus 4.8</mm:model>
<!-- usage — Creative Commons -->
<cc:license rdf:resource="https://creativecommons.org/licenses/by/4.0/"/>
<!-- credit — IPTC / Dublin Core -->
<dc:creator>Alex Rivera</dc:creator>
</rdf:Description>
<!-- …the file is then signed as a C2PA manifest that
references the above, making the record tamper-evident. -->
Four standards, four namespaces, one file — none contradicting another. Verify any of it on the verify page.
The Creative Commons fit
Creative Commons has spent two decades on the question of what you may do with a work — and is extending that into the AI era with preference signals for the commons. Made Mark answers the adjacent question that vocabulary was never built to carry: who or what made it.
Creative Commons
“Can this be reused — and can AI train on it?”Usage and permission. Licenses and preference signals.
Made Mark
“Was this made by a person, an AI, or both?”Authorship disclosure. Three human-readable marks.
They share the same RDF/XMP carrier, use separate namespaces, and never contradict each other — so a work can be labeled with both at once. That is the case for treating Made Mark as a complement to the commons, not a competitor for the same slot.
Implementing across the stack
The schemas for each carrier — JSON-LD, XMP, and the C2PA assertion — live on the Implement page, with copy-paste examples. How the vocabulary and URIs stay stable as the ecosystem evolves is set out in Governance.