Entity Drift: Saat Informasi Brand Berubah tapi AI Masih Mengingat Versi Lama

Brand sudah berubah.

Website sudah berubah.

Logo sudah berubah.

CEO sudah berubah.

Tapi ketika seseorang bertanya ke AI, jawaban yang keluar masih versi lama.

“Bukannya kita sudah update tiga bulan lalu?”

Sudah.

Masalahnya, internet tidak punya tombol “hapus ingatan lama”.

Informasi lama bisa tetap hidup di media, direktori, cache, PDF, marketplace, social profile, press release, partner page, dan halaman lama yang belum ditutup.

AI system juga tidak selalu menggunakan source yang sama pada setiap jawaban.

Akibatnya, perubahan brand tidak otomatis menghasilkan perubahan serentak di seluruh answer ecosystem.

Fenomena ini bisa disebut entity drift: ketika representasi publik tentang entity berubah dari kondisi sebenarnya, atau beberapa versi identity hidup bersamaan sehingga system eksternal masih menghasilkan state lama.

Entity drift bukan satu bug.

Ia biasanya hasil dari time lag, source conflict, dan governance yang kurang rapi.

Contoh Paling Sederhana: CEO Berganti

Januari:
CEO A.

April:
CEO B appointed.

Website About update.

LinkedIn organization update.

Press release baru.

Tapi ada 200 artikel lama yang masih menyebut “CEO A”.

Itu historical fact.

Tidak salah jika article bertanggal 2024.

Masalah muncul kalau halaman evergreen:

“Leadership Team”

masih CEO A.

Directory business profile masih A.

Conference bio bulan lalu masih A.

Wikipedia belum update.

AI bisa melihat mix.

One answer:
CEO A.

Another:
CEO B.

Third:
“CEO is A, while B serves another role.”

Now entity drift visible.

Tidak selalu hallucination murni.

Source ecosystem sendiri conflict.

Rebranding adalah Drift Factory

Old Brand → New Brand.

Perusahaan berharap:

Friday:
Launch.

Monday:
AI understands.

Realistically, old brand may have years of web footprint.

Backlinks.

Reviews.

Media.

Directory.

YouTube.

Marketplace.

PDF.

Contracts.

Job posts.

Old domain.

Historical knowledge.

If new site simply deletes old name, relation between old and new becomes harder, not easier.

Better transition:

“New Brand, formerly Old Brand.”

Effective date.

Legal entity relation.

Redirect.

Official announcement.

Updated profiles.

Archive historical pages.

Now machine has bridge.

Entity drift can be managed.

Not instantly erased.

Product Discontinuation juga Sering Drift

Product X stopped July 2026.

Website removes product page.

But retailer still lists.

Review article still current in search.

Comparison sites still mention.

AI recommends Product X.

Customer clicks official site.

404.

Bad experience.

Better:

Keep legacy page if references still strong.

“Product X was discontinued in July 2026. Current replacement: Product Y.”

Redirect only if replacement truly equivalent.

Do not silently redirect unrelated.

Now old references lead to current truth.

This is entity lifecycle design.

Alamat Berubah, Internet Tetap Punya Gedung Lama

Office moves.

Website contact update.

Maps update.

But old directories still list.

PDF company profile old.

Media article old.

AI answer may tell user wrong office.

Physical consequence.

Someone actually goes there.

This is why location drift deserves high severity.

Entity governance should classify change by risk.

Address.

Phone.

Safety.

Legal status.

Product availability.

Leadership.

Higher urgency.

Brand tagline?

Lower.

Not all facts equal.

Entity Drift Bisa Datang dari Website Sendiri

Common cause:

Homepage updated.

About old.

Footer old.

Schema old.

Contact old.

PDF old.

Blog author bio old.

English site old.

Regional site old.

AMP/cache old.

Now official domain itself sends multiple truths.

Before blaming AI, crawl your own site.

Search old:

Brand name.

CEO name.

Address.

Product.

Legal name.

Pricing.

Slogan.

Use site search and search engine `site:` queries.

Find fragments.

Entity drift often begins internally.

Structured Data Bisa Menjadi Zombie

Visible page says CEO B.

JSON-LD still points Person A.

Or Organization name old.

sameAs old social profile.

Product status active.

Schema validation still green.

Semantically stale.

This is dangerous because team assumes:

“Schema already handled.”

No.

Structured data needs lifecycle.

When critical entity changes, schema update in same release.

Treat entity change as cross-layer deployment.

Not content-only.

Third-Party Drift Harus Diprioritaskan, Bukan Dikejar Semua

You cannot update entire web.

Prioritize sources by impact.

Tier 1:
Official profiles.

Regulator.

Major directories.

Major platforms.

High-visibility media profile.

Marketplace official listing.

Partner pages with strong visibility.

Tier 2:
Industry directories.

Conference profiles.

Review sites.

Tier 3:
Low-quality scraped pages.

Focus.

Some old articles should not be edited because they correctly represent historical state at publication time.

Instead, ensure current canonical truth is stronger and clearly dated.

Historical truth vs current truth.

Do not erase history.

Timestamp adalah Obat Penting

“CEO A memimpin Company X.”

Ambiguous forever.

“Pada 2024, CEO A memimpin Company X.”

Now historical.

“Sejak April 2026, CEO B menjabat CEO.”

Current.

Dates turn contradiction into timeline.

This is especially valuable for:

Leadership.

Ownership.

Funding.

Product status.

Pricing.

Policy.

Office.

Certification.

Partnership.

Brand identity.

AI may still synthesize incorrectly, but source clarity improves.

Entity pages need temporal facts where relevant.

Skenario di Tim Brand

PR:
“Media lama masih sebut nama lama. Kita minta edit semua?”

Legal:
“Artikel 2023 memang benar saat itu.”

SEO:
“Kalau dibiarkan AI bisa ambil.”

Brand:
“Terus gimana?”

Answer:

Do not rewrite history unless source factually wrong.

Create clear current identity page.

Publish rebrand timeline.

Update high-impact evergreen third-party profiles.

Use redirects.

Maintain former-name relation.

Monitor AI.

This is mature approach.

Entity drift is not solved by historical deletion campaign.

Sometimes deletion removes evidence connecting identities.

Entity Drift dan Model Drift Berbeda

Model drift:
System behavior changes.

Entity drift:
Public entity state changes relative to representation.

They can interact.

If model update begins preferring fresher sources, old entity error may fall.

If model uses old source, error rises.

Measurement must separate.

Track:

Entity truth version.

Website change.

Source ecosystem.

Model/surface.

Date.

If answer changes, inspect both.

Don't attribute everything to content.

Canonical Entity Registry Membantu

Internal source of truth:

Entity:
Organization.

Current name.

Former names.

Current CEO.

Former CEO.

Current address.

Former address.

Products.

Status.

Effective dates.

Canonical URLs.

Every public page can reference.

Change management:

1. Update registry.
2. Update canonical page.
3. Update schema.
4. Update related pages.
5. Update official profiles.
6. Notify critical partners.
7. Archive/redirect legacy.
8. Retest.

This is entity operations.

Not just SEO.

Drift Detection Query Set

Build 20-50 branded queries.

“Who is CEO of X?”

“Where is X located?”

“What does X sell?”

“Is X formerly Y?”

“Who owns X?”

“Is Product A still available?”

“Official website?”

Repeat across models.

Track.

If wrong fact appears:

Which source cited?

How often?

Which model?

When first seen?

Is it persistent?

Then classify.

One error:
Observation.

Persistent:
Issue.

Severity:
Action priority.

Hallucination vs Drift

If AI says CEO A and old source says CEO A, output is outdated relative to present.

Drift.

If AI says CEO Z who never existed, likely unsupported/hallucinated.

Different remediation.

Drift:
Source cleanup and timeline clarity.

Fabrication:
Entity ambiguity, synthesis error, source mismatch.

Do not use “hallucination” for every old fact.

Precision helps diagnosis.

Public Evidence Can Freeze Old Identity

Case study from 2022.

Award page.

Partner listing.

Press release.

These are valuable evidence but can contain old brand name.

Don't delete.

Add:

“Brand formerly known as X.”

Or current note where appropriate.

Partner may update profile title while preserving historical content.

The goal is bridge.

If old evidence disappears, new brand loses history.

Entity continuity is asset.

Website Migration Bisa Mempercepat Drift kalau Buruk

New domain.

Old domain 404.

No redirects.

About rewritten.

Old PDFs indexed.

Social profiles mix domains.

Now old and new entity can split.

Migration plan must include identity mapping.

URL mapping.

Redirects.

Structured data IDs.

sameAs.

Canonical pages.

Sitemap.

Public profiles.

Press announcement.

Search Console.

AI monitoring.

Not guarantee AI merge.

But reduces fragmentation.

Entity Drift pada Multi-Country Brand

Global brand changes product availability in Indonesia.

Global page current.

Indonesia page stale.

AI user Indonesia gets global answer.

Product not available locally.

Entity/market drift.

Need local truth.

Availability by market.

Operator.

Distributor.

Location.

Language.

Temporal.

Multi-country drift is more complex because facts can be simultaneously true in one country and false in another.

Context must be explicit.

Automation Bisa Membantu Drift Detection

Daily/weekly crawl:

Critical facts.

Schema values.

Page mentions.

Old brand names.

Old leadership.

Broken redirects.

Third-party API where permitted.

Alerts.

But external web monitoring must follow legal/terms.

Human review still.

Automation can flag.

Not decide truth.

Example:

Old CEO name found on 17 official pages.

Queue.

Editor reviews.

Some historical article.

Leave.

Some About author bio.

Update.

Risk-based.

Entity Drift SLA

Set internal response target.

Critical legal/safety:
Immediate.

Leadership:
1-3 business days.

Location/contact:
Immediate/fast.

Product availability:
Same day.

Brand description:
Within release.

Third-party correction:
Best effort, track.

AI observation:
Retest weekly/monthly.

No universal SLA.

Define based on business.

Without SLA, drift persists because everyone assumes someone else will fix.

Public Truth Needs Owner

Who owns company facts?

Corporate communications.

Legal.

Brand.

Product.

Operations.

Create one governance owner.

Not necessarily one person edits all.

One authority approves.

If PR says A and website says B, escalation path clear.

Entity drift is often organizational drift.

Teams have different truths.

Machine just reflects conflict.

Don't Expect Instant Memory Update

External systems crawl and refresh on their own cadence.

Even after website correct, old answer can persist.

Do not panic same day.

Track over time.

Document remediation.

Run controlled tests.

If error remains persistent and platform has feedback mechanism, use it.

But avoid spamming.

Measurement discipline.

Before:

Wrong CEO 8/10.

After update + four weeks:

3/10.

Later:

1/10.

Observed improvement.

Not proof all users see correct answer.

But signal.

Entity Drift Tidak Bisa Dihapus Sepenuhnya

Internet is historical.

Old information remains.

That's okay.

The goal isn't zero old mentions.

The goal is clear temporal relationship.

Current canonical truth should be easy to find.

Historical sources should be dated.

High-impact stale profiles updated.

Structured data current.

Redirects work.

AI error monitored.

If organization manages that, drift becomes manageable.

When Brand Changes, Graph Must Change Too

Entity system should encode lifecycle.

FormerName.

Successor.

Predecessor.

validFrom.

validThrough.

Current status.

Archived product.

Former CEO.

New owner.

Don't model world as static.

Brands evolve.

AI-era identity architecture needs time.

Without time, every change looks like contradiction.

With time, it becomes history.

So when AI “still remembers” old version, ask two questions:

Is the AI wrong?

Or is the web still telling the old story without a date?

Often, the answer is both.

And fixing it starts with making the current story clearer, not pretending the old one never existed.