Kenapa SameAs Berlebihan Bisa Membuat Entity Mapping Makin Berantakan

Ada pola yang kelihatan logis tetapi sering salah arah.

AI masih salah mengenali brand.

Solusi pertama: tambah sameAs.

Masih salah.

Tambah lagi.

LinkedIn.

Instagram.

Facebook.

YouTube.

TikTok.

Marketplace.

Founder profile.

Media article.

Partner page.

Directory.

Wikipedia.

Wikidata.

Profile conference.

Lalu Organization schema punya daftar sameAs sepanjang menu restoran.

Kelihatannya kaya.

Secara entity mapping, bisa justru semakin berantakan.

Masalahnya sederhana: sameAs bukan daftar “semua URL yang berhubungan dengan brand”. sameAs menyatakan bahwa URL lain merujuk ke entity yang sama.

Identity equivalence.

Bukan association.

Bukan mention.

Bukan endorsement.

Bukan backlink.

Schema.org mendefinisikan sameAs sebagai URL halaman referensi yang secara tidak ambigu menunjukkan identitas item tersebut. Kata pentingnya bukan URL. Kata pentingnya “identitas yang sama”.

Kalau relationship itu salah, kita bukan memperkaya graph.

Kita sedang membuat false merge.

Lebih Banyak Edge Bukan Berarti Graph Lebih Baik

Bayangkan entity graph sebagai peta.

Node utama:
PT Nusa Data.

Valid sameAs:
Official LinkedIn organization page.
Official Wikidata item.
Official profile di authoritative registry jika memang mewakili entity yang sama.

Lalu marketing menambah:

Founder LinkedIn.
Product marketplace.
Media interview.
Partner article.
Award page.
Customer review.

Semua punya hubungan dengan company.

Tapi bukan company itu sendiri.

Graph sekarang punya banyak garis.

Masalahnya, garisnya salah label.

Untuk manusia, kita masih bisa melihat context.

“Ini founder.”

“Ini product.”

“Ini media.”

Machine-readable statement sameAs tidak membawa nuance itu.

Ia berkata:

“Ini entity yang sama.”

Itu overstatement.

Founder Bukan sameAs Organization

Ini error klasik.

Organization:
PT Nusa Data.

Founder:
Rina Wijaya.

Rina punya LinkedIn.

Jika Organization.sameAs menunjuk LinkedIn Rina, machine representation mengatakan profile Rina adalah identity reference untuk Organization.

Padahal Rina adalah Person.

Hubungan yang benar:

Person Rina founded Organization.

Person Rina worksFor atau memiliki role di Organization.

Not sameAs.

Person profile bisa punya sameAs sendiri.

Company profile punya sameAs sendiri.

Pisahkan node.

Relationship lebih precise.

Jangan gunakan sameAs sebagai shortcut karena developer malas membuat Person object.

Product Listing Bukan sameAs Brand

Brand:
Langit Coffee.

Product:
Langit Cold Brew.

Marketplace listing:
Langit Cold Brew 250ml.

Brand sameAs marketplace product listing?

Tidak.

Product listing describes a Product.

Brand is Brand.

Organization may own Brand.

Merchant sells Product.

Different entity types.

Jika semua ditumpuk ke sameAs, external system bisa mendapat contradictory signals.

Apakah brand itu product?

Apakah marketplace seller adalah Organization owner?

Apakah product page merupakan company identity?

Semakin banyak false equivalence, semakin sulit mapping bersih.

Media Article Juga Bukan sameAs

Article:

“CEO PT Nusa Data Bicara AI Governance.”

Artikel membahas company.

Bukan company.

It can be citation.

Evidence.

Mention.

About.

Subject.

Not identity equivalent.

sameAs ke media article hanya karena media terkenal adalah semantic misuse.

Authority media tidak “ditransfer” lewat sameAs.

Tidak ada dasar untuk menganggap property ini sebagai trust booster.

Use it for identity.

Nothing more.

Skenario yang Sering Terjadi di Audit

Developer:
“Kenapa sameAs kita 34?”

SEO:
“Biar entity kuat.”

Developer:
“Ini ada link ke founder, dua product, tiga article.”

SEO:
“Semua tentang brand.”

Entity analyst:
“Kalau semuanya entity yang sama, berarti founder itu company dan product itu company?”

SEO berhenti.

Good.

Pertanyaan itu memaksa semantics kembali ke reality.

Graph harus bisa diterjemahkan ke sentence.

“PT Nusa Data is the same entity as this LinkedIn Organization profile.”

Yes.

“PT Nusa Data is the same entity as this founder interview.”

No.

Kalau sentence terdengar absurd, edge salah.

SameAs Berlebihan Bisa Membuat False Merge

False merge adalah ketika dua entity berbeda diperlakukan sebagai satu.

Ini bisa lebih berbahaya daripada missing relation.

Contoh:

Parent company.

Subsidiary.

Same brand family.

Jika subsidiary.sameAs parent, system bisa menggabungkan:

Leadership.

Location.

Legal identity.

Products.

Ownership.

Financial claims.

Salah.

Parent dan subsidiary related.

Not identical.

Likewise:

Global Organization dan local distributor.

Franchise operator dan brand owner.

Agency dan founder.

Brand dan Product.

Jangan merge hanya karena user mengenal semuanya dengan nama yang sama.

Entity architecture butuh boundary.

Nama Mirip Membuat Risiko Makin Tinggi

Brand bernama “Orbit”.

Official LinkedIn:
Orbit AI Indonesia.

Ada LinkedIn lain:
Orbit Technologies Singapore.

Marketing salah pilih URL karena logo mirip.

sameAs salah.

Sekarang markup mengassert identity ke entity lain.

Nama generic meningkatkan risk.

sameAs harus diverifikasi.

Check:

Domain.

Location.

Leadership.

Official website.

Company name.

Industry.

Unique identifier.

Jangan hanya logo.

One bad sameAs can create stronger ambiguity than no sameAs.

Directory Berkualitas Rendah Jangan Dipakai sebagai Identity Anchor

Scraped directory.

Data lama.

Address salah.

Founder salah.

Website benar.

Team tetap memasukkan ke sameAs karena “domain punya authority metric tinggi”.

Problem.

sameAs isn't backlink strategy.

If directory profile is inaccurate, you're explicitly telling machines:

“This profile represents us.”

That makes source conflict self-inflicted.

First correct directory.

Then decide whether it deserves sameAs at all.

Not every correct profile needs to be in sameAs.

Curate.

Jangan Punya Target Jumlah sameAs

“Minimal 10.”

“Optimal 20.”

“Enterprise 50.”

No universal evidence.

There is no documented magic number.

A company may only have:

LinkedIn.

Wikidata.

Official industry profile.

Three valid references.

Fine.

Another organization might have 12 legitimate authoritative identities across research systems, regulators, social networks, and public registries.

Also fine.

Quantity follows reality.

Not SEO target.

Entity strategy without quota is healthier.

sameAs Harus Punya Owner

Social accounts change.

LinkedIn slug changes.

Wikipedia pages merge.

Wikidata item corrected.

Directory acquired.

Brand rebrands.

If sameAs list is hardcoded and forgotten, old URLs become identity drift.

Set owner.

Quarterly or event-triggered review.

For each URL:

HTTP 200?

Still same entity?

Still current?

Redirected?

Changed ownership?

Hacked?

Private?

Deleted?

If not, update.

sameAs is living relationship.

Not one-time schema field.

Rebranding Membuat sameAs Sangat Sensitif

Old Brand becomes New Brand.

Social account renamed.

Same account, same entity.

Valid continuity.

But new brand may create new account.

Old account becomes archive.

Legal entity may stay.

Brand entity may change.

Don't blindly list old and new as sameAs.

First answer:

Is this same Organization?

Same Brand renamed?

New Brand replacing old Brand?

Acquisition?

Different legal entity?

Relationship changes based on answer.

Sometimes `alternateName` plus rebrand explanation is better.

Sometimes old URL redirects to current identity.

Sometimes entities remain separate historically.

No single sameAs recipe.

Schema is not migration logic by itself.

Canonical Page Tetap Lebih Penting

If Organization identity page is weak:

Name unclear.

Legal entity unclear.

Location conflicting.

Product relation vague.

Adding sameAs doesn't fix base node.

Start with canonical page.

Who are we?

What is legal/public name?

What do we do?

What brands/products?

Where operate?

Who leads?

What official profiles?

Then structured data.

sameAs references reinforce identity.

They don't create identity from nothing.

A clean canonical entity page with three valid sameAs is stronger than thin page with thirty URLs.

sameAs dan Citation Jangan Dicampur

Marketing may think:

“Put media citations in sameAs so AI sees them.”

No.

Evidence architecture should model evidence as evidence.

Article can be linked naturally.

Case study.

Press page.

Citation property where semantically appropriate for CreativeWork.

Internal graph can use `supportedBy`.

But do not misuse identity equivalence.

AI citation selection is external behavior.

No guarantee sameAs increases citation.

Keep facts and hypothesis separate.

Google Organization Structured Data Tidak Mendorong sameAs Maximalism

Google Organization documentation frames structured data as helping Google understand administrative details and disambiguate organization.

That is compatible with curated sameAs.

It does not imply “add every profile you can find.”

General structured-data guidelines also require true representation of page content and prohibit misleading data.

A false sameAs is misleading semantics.

Validation may not flag it.

Rich Results Test can't know founder profile is different entity just from syntax.

Human semantic audit required.

How to Audit Existing sameAs

Export all sameAs URLs.

Columns:

URL.

Entity it actually represents.

Type.

Official?

Current?

Verified?

Canonical?

HTTP status.

Owner.

Keep/Remove/Review.

Then classify.

KEEP:
Same entity, authoritative/current.

REMOVE:
Different entity.

REVIEW:
Ambiguous, stale, redirected.

Example:

LinkedIn company → KEEP.

Founder LinkedIn → REMOVE from Organization; move to Person.

Product marketplace → REMOVE from Organization.

Wikidata correct item → KEEP.

News article → REMOVE from sameAs; retain as evidence link if relevant.

Old Facebook page → REVIEW.

Now graph becomes smaller.

That's okay.

Smaller can be better.

Entity Mapping Butuh Relationship yang Tepat

Instead of sameAs everything:

Organization → founder → Person.

Organization → brand → Brand.

Product → brand → Brand.

Product → manufacturer → Organization.

Person → worksFor → Organization.

Organization → parentOrganization → Organization.

WebPage → about/mainEntity → Entity.

Article → about → Entity.

Dataset → creator → Organization.

Correct relationship gives information.

sameAs destroys information if used as universal connector.

Graph quality comes from predicate precision.

Not edge count.

Unknown Lebih Baik daripada False SameAs

If you're not sure whether directory profile is same company:

Don't add.

Investigate.

If company has identical name in two cities:

Don't assume.

If social account isn't verified:

Don't add.

Unknown is manageable.

False merge can propagate.

This is especially important for automated schema generation.

Do not auto-discover sameAs by matching names.

Name matching creates huge false-positive risk.

Need identity resolution.

Human review.

Stable IDs.

Evidence.

Automated candidate generation is fine.

Auto-publish equivalence isn't.

sameAs untuk Dataset juga Harus Presisi

Google Dataset documentation gives a useful example of proper semantics.

When the same dataset is republished, sameAs can point to the canonical version if identity really is the same.

If dataset changed significantly, Google guidance points toward `isBasedOn`, not sameAs.

This shows principle clearly.

Same data unchanged:
identity.

Derived/modified:
relationship, not equivalence.

The same logic applies broadly to entity architecture.

If thing changed into a different thing, stop calling it same.

sameAs Overload Bisa Membuat Debugging Sulit

AI output wrong.

Which reference caused?

34 sameAs URLs.

Some current.

Some stale.

Some wrong entity.

Some media.

Some products.

Now audit requires archaeology.

Curated sameAs reduces surface area.

If external identity conflict appears, trace faster.

Good graph is auditable.

Every edge has reason.

“Why is this URL here?”

Answer should be one sentence.

If answer is:

“SEO agency dulu bilang bagus.”

Remove or investigate.

Semantic Debt Is Real

Technical debt:
Old code.

Semantic debt:
Old meanings.

sameAs list is semantic code.

Bad relation may survive years.

New developers assume intentional.

Plugin copies.

Schema graph replicates.

Then rebrand adds more.

Soon no one knows.

Treat structured data as production data.

Version.

Review.

Change log.

Owner.

Schema governance should include sameAs policy.

Policy Sederhana

Eligible:

Official organization social profile.

Authoritative public identity database entry.

Official profile controlled or clearly representing same organization.

Verified canonical repository identity where applicable.

Not eligible:

Founder/person profile.

Product listing.

News article.

Customer review.

Partner article.

Search result.

Hashtag.

Competitor.

Random directory.

Unofficial account.

This policy isn't universal law.

But it's good starting discipline.

Add exceptions only with semantic reason.

Entity Mapping Menjadi Bersih Saat Kita Berani Menghapus

Teams often fear removing markup.

“Kalau sameAs turun dari 30 jadi 6, authority turun nggak?”

There is no basis to claim that.

What definitely improves:

Correctness.

Auditability.

Entity type separation.

Maintenance.

Less contradictory identity.

That is enough.

Then measure AI entity accuracy.

Brand queries.

Founder.

Product.

Location.

Legal name.

Cross-model.

Before/after.

If errors drop, good observation.

If not, continue source audit.

Don't judge by property count.

sameAs is one of the most useful Schema.org properties when used literally.

It says:

“This is the same thing.”

That precision is its power.

And that same precision is why overusing it breaks the graph.

If every related URL becomes “same”, then nothing is meaningfully distinct.

A knowledge graph without distinction isn't knowledge.

It's a pile of links.