Schema untuk Content Provenance: Jangan Samakan dengan C2PA

Ada satu slide yang belakangan makin sering muncul di meeting content trust.

“Provenance.”

Di bawahnya ada:

Author schema.

datePublished.

publisher.

citation.

C2PA.

Content Credentials.

Watermark.

AI disclosure.

Semua ditumpuk dalam satu kotak seolah fungsinya sama.

Padahal tidak.

Structured data bisa membantu menjelaskan siapa membuat content, kapan diterbitkan, siapa publisher, apa sumber yang dirujuk, dan relationship antar-CreativeWork.

C2PA punya job berbeda.

C2PA adalah standard teknis provenance dan authenticity yang menggunakan manifest, assertions, content bindings, serta cryptographic signatures untuk memberi bukti tentang asal dan history suatu digital asset.

Dua-duanya bicara “provenance” dalam sense luas.

Tetapi level jaminannya berbeda.

Schema adalah semantic assertion.

C2PA Content Credential adalah cryptographically verifiable provenance mechanism.

Dan bahkan C2PA sendiri secara eksplisit mengingatkan bahwa provenance information tidak otomatis membuktikan content itu true, accurate, atau factual.

Ini distinction yang sangat penting di era synthetic media.

Schema Bisa Menjawab “Siapa Mengatakan Apa”

Article structured data dapat menyatakan:

author.

publisher.

datePublished.

dateModified.

headline.

URL.

citation atau related work dalam contexts yang sesuai.

CreativeWork relationship dapat menunjukkan sebuah work berdasarkan, merujuk, atau terkait work lain melalui properties yang tersedia dan benar secara semantic.

Person schema bisa mengidentifikasi author.

Organization schema bisa mengidentifikasi publisher.

Dataset bisa punya creator.

Report bisa punya author/publisher.

Jadi structured data membantu provenance editorial.

Kita bisa membaca graph:

Article A
→ author
→ Person B

Article A
→ publisher
→ Organization C

Article A
→ citation
→ Report D

Dataset E
→ creator
→ Organization C

This is useful.

It makes attribution explicit.

But what verifies that Organization C really signed that exact file?

Schema itself doesn't.

JSON-LD Is an Assertion, Not a Cryptographic Proof

Anyone who can edit HTML can write:

author:
“The New York Times.”

publisher:
“NASA.”

creator:
“University X.”

It can be false.

Validator can still parse.

Structured data syntax does not prove control over named identity.

Search systems can use other signals to evaluate.

But schema itself isn't authentication.

This is why it should not be presented as:

“Content authenticity certificate.”

It isn't.

It's machine-readable statement.

Trust depends on source, context, corroboration, platform processing, and other controls.

C2PA introduces cryptographic verification of provenance data.

Different class of mechanism.

Apa yang C2PA Sebenarnya Lakukan?

C2PA specification 2.4 defines Content Credentials around C2PA Manifests.

A Manifest combines:

Assertions.

A claim.

Content bindings.

A claim signature.

The manifest can be cryptographically validated.

The specification explains that manifests represent provenance data of an asset.

C2PA uses hashes/signatures so unauthorized tampering with asset provenance can be detected during validation.

It can also represent ingredient relationships when assets are composed from other assets.

This is much closer to a signed history.

Example conceptual:

Camera creates photo.

Manifest records assertions.

Asset bound through hash.

Claim signed.

Later editor modifies.

New manifest can reference ingredient/history.

Consumer validates.

Now user can inspect provenance chain.

This is not what Article JSON-LD does.

C2PA Bukan “Schema yang Lebih Kuat”

Calling C2PA advanced schema is misleading.

Schema.org:
semantic vocabulary for describing things and relations.

C2PA:
technical standard for content provenance/authenticity with cryptographic validation.

Different architecture.

Schema is often embedded in web page JSON-LD.

C2PA manifests can be embedded or associated with supported assets, with detailed format-specific mechanisms.

C2PA 2.4 specification includes handling for multiple media/document formats and even considerations for HTML/structured text.

But implementing Content Credentials is a provenance system, not adding another JSON-LD block.

Needs:

Claim generator.

Signing credential.

Manifest.

Binding.

Validation.

Trust model.

Possibly repository/external manifests.

Lifecycle.

User experience.

Don't assign this to SEO intern with “install plugin” checklist and assume done.

C2PA Juga Tidak Membuktikan “Konten Ini Benar”

This is perhaps most important.

Official C2PA explainer directly states provenance information alone cannot tell you whether digital content is true, accurate, or factual.

Why?

Because a trusted publisher can publish something wrong.

A camera can authentically capture a staged scene.

An AI system can legitimately generate a fictional image and sign provenance correctly.

A bad actor can sign their own misleading content under their own identity, depending trust model.

What C2PA can help answer:

Where did asset come from?

Who/what signed assertions?

Has provenance data/content binding been tampered with?

What edits/ingredients are represented?

Then human/consumer decides trust.

It does not output:

TRUTH = TRUE.

This is why Content Credentials should never be marketed as universal misinformation detector.

Provenance supports evaluation.

Not factual adjudication.

Structured Provenance Punya Nilai Meski Tanpa C2PA

Does that mean ordinary websites should ignore author/date/source schema unless they deploy cryptographic signing?

No.

Editorial provenance remains valuable.

Reader needs:

Who wrote?

Who reviewed?

When updated?

What sources?

Which dataset?

Publisher?

Correction?

Version?

Visible page should communicate.

Structured data can mirror.

This helps:

Search understanding.

Internal governance.

Machine interoperability.

Audit.

Content management.

C2PA adds a different verification layer for digital asset provenance.

You can have both.

They complement.

Don't make one prerequisite for the other.

Article Provenance vs Asset Provenance

This distinction helps.

Article page:
Editorial work.

Inside:
Hero image.

Chart.

Video.

PDF report.

Each asset can have its own provenance.

Article JSON-LD can say:

author Person A.

publisher Organization B.

image ImageObject.

citation Report C.

The image could have C2PA Content Credential.

The PDF could too, depending implementation.

Now:

Editorial graph describes relationship.

C2PA verifies provenance metadata of specific asset.

Two levels.

If image reused in social platform and metadata survives or can be recovered through supported mechanisms, provenance can travel in ways page JSON-LD cannot.

Conversely, article context includes relationships asset-level manifest may not fully express.

Layering.

C2PA Identity Is Not the Same as Schema sameAs

Schema sameAs:
identity reference URL.

C2PA trust model:
signing identity/credential and validation.

Do not say:

“sameAs verifies signer.”

It doesn't.

Do not use Wikidata link as cryptographic proof.

And don't assume a valid C2PA signature means signer is globally trusted; trust decisions depend on credential/trust model and validation context.

C2PA specification explains trust is tied to signer identity associated with cryptographic signing key and credential validation.

This is real identity assurance architecture.

Different from public entity graph.

Good implementation can connect them conceptually:

Organization public page.

Organization schema.

C2PA signing identity.

Human-readable trust page.

But verification layers stay distinct.

Watermark Juga Bukan C2PA

Visible or invisible watermark can indicate ownership/origin.

It can be cropped.

Removed.

Copied.

Faked.

C2PA uses cryptographic manifests/bindings.

Watermark may complement.

Not equivalent.

Likewise file metadata EXIF:

Can contain provenance-like data.

Can be modified.

C2PA is designed around verifiable claims and bindings.

Do not call any metadata “Content Credential”.

Use term accurately.

Content Credentials specifically refers to C2PA technology/preferred nontechnical term for C2PA Manifest in spec context.

Terminology matters.

AI Disclosure Bukan Otomatis Provenance

Article says:

“Created with AI assistance.”

That's disclosure.

Useful.

Could be required by policy.

But it doesn't cryptographically prove which model, prompt, or edits.

Schema may include creation metadata where vocabulary fits.

C2PA AI/ML assertions/guidance may provide structured provenance possibilities for assets in supported workflows.

But plain sentence is policy transparency.

Different.

Again:

Disclosure.

Semantic metadata.

Cryptographic provenance.

Three layers.

Don't collapse.

Provenance Graph Should Track Versions

Institutional publication:

Report v1.

Correction v1.1.

Dataset v2.

Chart regenerated.

PDF updated.

Visible version history.

Schema can connect version/source relationships where applicable.

Internal graph should store:

work ID.

version.

supersedes.

derived from.

dataset.

author.

publisher.

correction.

C2PA can preserve provenance for specific files/assets.

If PDF v2 is new signed asset:
new manifest/history.

This is stronger than silently replacing file at same URL with no record.

Content trust requires lifecycle.

Not one badge.

Correction Record Is Provenance Too

Publication was wrong.

Correction issued.

Don't hide.

Visible correction note.

Date.

What changed.

Original version if policy allows.

Updated structured metadata.

C2PA manifest/asset history can potentially show editing history in implementations, but editorial correction semantics still need human-readable context.

Cryptographic proof that file changed doesn't tell reader why.

Schema can link relevant CreativeWorks.

Content page can explain correction.

A mature provenance system has both:

Technical lineage.

Editorial explanation.

C2PA Doesn't Prevent Manifest Removal Entirely

C2PA security considerations explicitly note that the standard does not protect against complete removal of C2PA manifests from assets, although mechanisms such as soft bindings/manifest repositories can help in some scenarios.

Important limitation.

If image platform strips metadata, provenance may be missing.

Absence of credential doesn't prove image fake.

This must be said clearly.

Otherwise users learn bad heuristic:

Credential present = true.
Credential absent = fake.

Wrong.

Presence means provenance data can be validated under its trust model.

Absence means no C2PA credential was found through available mechanisms.

No further universal conclusion.

Schema Bisa Hilang Juga

Article syndicated.

Copied as plain text.

JSON-LD not copied.

Screenshot.

PDF export.

Schema provenance is page-local unless consumer preserves.

C2PA is designed specifically to bind provenance to asset or reference manifest, giving different portability.

This is one reason asset provenance standards exist.

But HTML publishing context remains important.

A news article's author/publisher page still useful.

No one layer survives every transformation.

Architecture should decide which provenance must travel.

Use Case: Institutional Report

GEO.or.id publishes:
“AI Search Measurement Report Q3 2026.”

Provenance stack could be:

Visible page:
authors, publisher, method, version, date.

Schema:
Report + authors + publisher + related Dataset.

PDF:
Content Credential using C2PA if organization implements.

Dataset:
version, creator, license, hash in internal evidence system.

Correction page:
change log.

Organization identity:
canonical page.

Now trust is layered.

User can see.

Machine can parse.

Validator can verify asset provenance.

Still, none of this automatically proves report conclusions scientifically valid.

Method and evidence determine that.

This is mature claim scope.

Use Case: AI-Generated Image

Publisher uses AI image.

Visible caption:
illustration generated with AI.

ImageObject metadata:
creator/publisher/license if true.

C2PA Content Credential:
may record provenance/assertions from creation/editing tools depending workflow.

Reader:
gets disclosure.

Validator:
can inspect credential.

Editorial:
explains context.

If image depicts imaginary scene:
don't present as documentary photo.

C2PA doesn't absolve editorial responsibility.

It can prove provenance chain.

Not intent.

Not interpretation.

When Does Website Need C2PA?

High-value provenance use cases:

News/media assets.

Research visual evidence.

Official institutional images.

Government communications.

Brand assets vulnerable to impersonation.

AI-generated/synthetic media disclosure workflows.

Documents where source/history matters.

Not every blog icon needs.

Implementing C2PA has operational cost.

Signing credentials.

Tool support.

Asset pipeline.

Storage.

Validation.

UX.

Revocation/trust.

Start where provenance risk is meaningful.

Don't deploy because “AI SEO”.

There is no basis to say C2PA boosts search ranking or AI citation universally.

It's trust infrastructure.

Content Provenance Checklist

For web content:

Author visible?

Publisher?

Date?

Version?

Source?

Dataset?

Correction?

Canonical URL?

For structured data:

Matches page?

Stable IDs?

Creator/publisher relationships?

Source relations accurate?

For C2PA asset:

Manifest?

Signature validates?

Credential valid under trust model?

Asset binding valid?

Ingredients/history where applicable?

Manifest accessible?

User can inspect credential?

For interpretation:

Does team avoid calling provenance “truth”?

Does absence avoid “fake” label?

Does invalid credential trigger investigation rather than instant factual verdict?

This is the governance layer.

Schema dan C2PA Bisa Berjalan Bersama tanpa Saling Meniru

Schema says:
“This report was published by Organization X.”

C2PA says:
“This asset carries a signed provenance manifest associated with a credential and can be validated for tampering/history according to the specification.”

Visible evidence says:
“Here is methodology.”

Research process says:
“Here is why findings are credible.”

These are separate statements.

When kept separate, trust grows.

When mixed, marketing overclaims.

The future of content provenance will likely involve more layers, not one magical badge.

Machine-readable authorship.

Cryptographic content credentials.

Source evidence.

Identity systems.

Correction logs.

Human editorial judgment.

So if someone asks:

“Can we just add provenance schema instead of C2PA?”

Answer:

No, if what you need is cryptographically verifiable asset provenance.

And if someone asks:

“Kalau sudah C2PA, kita nggak perlu author/schema/evidence lagi?”

Juga no.

C2PA proves a kind of provenance.

Schema describes content and relationships.

Evidence supports claims.

Editorial process establishes responsibility.

Don't make one layer carry another layer's job.

That is the most important rule in content provenance.