AI Search untuk Tim Marketing: KPI Baru yang Perlu Masuk Dashboard

Marketing dashboard biasanya sudah penuh.

Organic sessions.

Paid clicks.

CTR.

CPL.

Conversion.

Revenue.

Brand search.

Share of voice.

Social engagement.

Email.

Sekarang muncul request baru:

“Tambahkan AI visibility.”

Problem pertama:

AI visibility itu metric apa?

Kalau tim langsung membuat satu angka 0 sampai 100, dashboard memang jadi rapi.

Tetapi decision belum tentu lebih baik.

AI Search menciptakan beberapa jenis exposure yang berbeda.

Brand bisa disebut tanpa citation.

Cited sebagai source.

Masuk shortlist.

Muncul di product result.

Muncul sponsored.

Dijelaskan dengan fakta salah.

Tidak disebut, tetapi website mendapat traffic dari AI referral.

Semua itu punya meaning berbeda.

Jadi KPI baru harus dimulai dari event taxonomy.

Jangan satu score.

2026 membuat measurement sedikit lebih konkret

Google pada Juni 2026 mengumumkan Search Console generative AI performance reports untuk memberikan view khusus atas impressions di generative AI features seperti AI Overviews dan AI Mode, sementara data tetap berhubungan dengan overall Search performance.

Ini penting karena marketing sekarang mulai memiliki platform-native visibility data, setidaknya untuk Google.

ChatGPT Search juga memberikan source links/citations dalam kondisi tertentu, sementara referral dari AI platform dapat terlihat di analytics sesuai setup.

Tetapi cross-platform measurement belum seragam.

Jangan berharap satu metric universal.

Dashboard harus menerima asymmetry.

Metric 1: Buyer Query Coverage

Bukan total keyword.

Ambil query yang penting.

Discovery.

Category.

Comparison.

Brand.

Price.

Product.

Trust.

Location.

Procurement.

Support.

Misalnya 50 core query.

KPI:

Percentage of query panel where brand appears at least once under defined testing protocol.

Call:

Buyer Query Coverage.

Not “market share.”

It is internal panel.

Write denominator.

50 queries.

3 runs.

Fresh session.

Date.

Platform.

This is auditable.

Metric 2: Citation Coverage

How often owned source is cited?

But don't overvalue.

If AI mentions brand accurately from regulator/media, that's okay.

Use:

Owned Citation Coverage.

Third-Party Citation Coverage.

No Source Visible.

Also:

Citation accuracy.

A citation can exist but not support claim.

Sample manually.

Don't assume citation = endorsement.

Metric 3: Answer Accuracy

This should be high priority.

Create critical claim list.

Name.

Service.

Price.

Location.

Product.

Policy.

License.

Certification.

Availability.

Leadership.

Then score.

Correct.

Partially correct.

Stale.

Incorrect.

Unverifiable.

Marketing should not celebrate mention if answer wrong.

Accuracy KPI can be:

Percentage of evaluated material claims correct.

Keep severity separate.

Don't average critical and cosmetic into one number without weighting disclosure.

Metric 4: Critical Error Count

One of the best executive metrics.

How many Level 3/4 errors?

Wrong price.

Wrong branch.

Wrong service.

Wrong legal status.

Wrong health/finance claim.

Wrong certification.

This metric can be absolute.

“2 open critical errors.”

No need percentage.

Actionable.

Owner.

SLA.

Trend.

Marketing dashboard becomes risk dashboard too.

Metric 5: Source Conflict Rate

How many critical claims have conflicting sources?

Official site says A.

Directory B.

Media old.

Partner C.

AI might choose any.

Track:

Open source conflict.

Resolved.

Persistent.

This is leading indicator.

Before hallucination appears, source conflict already exists.

Marketing can coordinate with PR, product, ops.

Metric 6: Freshness Debt

How many critical sources are overdue review?

Price page.

Location.

Policy.

Team.

Product.

Promo.

Feed.

Make:

Critical claims due for verification.

Not “articles updated this month.”

Article count is output.

Freshness debt is risk.

If 30 price claims overdue, priority.

Metric 7: Shortlist Presence

For recommendation/comparison queries.

“CRM untuk startup Indonesia.”

“Agency GEO enterprise.”

“Software payroll 500 employees.”

Does brand enter shortlist?

But define.

Top 3?

Any listed option?

Mention in explanation?

Use one rule.

Don't change monthly.

Shortlist presence is commercially useful because buyer may not search brand name.

But no guarantee.

It is observed behavior.

Metric 8: Positioning Accuracy

Brand appears.

But why?

AI says SaaS is for SME.

Company targets enterprise.

Bad.

Create categories:

Correct ICP.

Too broad.

Too narrow.

Wrong category.

Old positioning.

This metric is qualitative but valuable.

Review sample.

Marketing owns positioning source.

AI can reveal narrative lag.

Metric 9: Product/Service Availability Accuracy

For commerce and service.

Product active?

Stock?

Branch?

Schedule?

Plan?

Service area?

This can be separate from general accuracy because it changes fast.

KPI:

Availability mismatch incidents.

High business consequence.

For SaaS:

Feature plan mismatch.

For F&B:

Menu.

For property:

listing.

For automotive:

dealer/model.

Dynamic data needs dynamic KPI.

Metric 10: AI Referral Traffic

Still useful.

Session.

User.

Conversion.

Lead.

Revenue.

But attribution limitations.

AI can influence without click.

User sees answer, then types domain directly.

Dark influence.

Don't say AI contributed zero because no referral.

Use referral as measurable subset.

Metric 11: AI-Assisted Lead Signal

Ask lead form:

“How did you hear about us?”

Option:

ChatGPT/AI search.

Sales call:

“Did you research us with AI?”

CRM note.

Not perfect.

But creates qualitative evidence.

Over time:

AI-mentioned leads.

Quality.

Deal size.

Conversion.

No need over-engineer first.

Metric 12: Generative AI Search Visibility from Search Console

Google's dedicated report can become a platform-native KPI.

Impressions.

Clicks where available/appropriate in reporting.

Queries/pages depending on report capabilities.

Marketing should compare to overall Search context.

Don't blend with ChatGPT observational data.

Label:

Google Generative AI Search.

ChatGPT Observation.

Other AI platform.

Different data source.

Metric 13: Paid AI Exposure

Separate.

If running ChatGPT Ads or other sponsored AI placement:

Impression.

Click.

Spend.

Conversion.

CPA.

ROAS.

Do not include in organic shortlist KPI.

OpenAI has stated ads are separate from organic answers.

Dashboard should reflect.

Paid success is good.

Just call it paid.

Metric 14: Correction Velocity

AI error detected.

How long until canonical source corrected?

Then external source.

Then retest.

Track:

Time to owned correction.

Time to retest.

Time to resolution.

This tells operational maturity.

Not algorithm.

Marketing can improve what it controls.

Metric 15: Evidence Coverage

For top buyer claims, do we have proof?

Claim.

Source.

Independent corroboration if relevant.

Date.

Owner.

Use percentage:

Critical buyer claims with current canonical evidence.

This is leading metric.

If low, no amount of GEO hack helps.

Build evidence.

Metric 16: External Source Health

PR and marketing cross.

Top 20 third-party pages about brand.

Current?

Correct?

Positive/negative is secondary.

Factual accuracy first.

Status:

Current.

Stale.

Wrong.

Conflict.

Needs outreach.

Third-party health influences public environment.

AI may find.

Metric 17: Query Volatility

How stable are answers?

Same query, multiple runs.

Brand appears 3/3.

1/3.

0/3.

Volatility.

Not one screenshot.

Repeated runs give better sense.

KPI:

Stable presence.

Intermittent.

Absent.

But be careful.

No platform promises determinism.

This is internal observation.

Metric 18: Unknown Rate

Important.

How many answers cannot be classified because source not visible or claim not verifiable?

Teams hate unknown.

But forcing answer creates fake precision.

Track unknown.

If high, improve methodology.

Not hide.

What should not enter dashboard?

“AI Trust Score” with no formula.

“GEO Authority.”

“LLM Rank.”

“Prompt Share” without definition.

“Citation Probability” invented.

Anything that looks official but is internal.

Internal index okay.

Label:

Internal.

Method v1.

Don't present as platform metric.

Dashboard needs versioning

Query panel changes.

Model changes.

Platform changes.

Search feature changes.

Google releases new report.

Method evolves.

Version.

“AI Search Measurement v1.4.”

Change log.

Month-to-month note.

Don't show continuous line if methodology break.

Marketing already understands attribution windows.

Apply similar discipline.

Dashboard should have three layers

Executive:

Buyer Query Coverage.

Accuracy.

Critical Errors.

Organic Shortlist.

Commercial Outcome.

Operator:

Source conflict.

Freshness.

Correction.

Citation.

Positioning.

Platform detail.

Evidence.

Diagnostic:

Raw query.

Answer.

Screenshot.

Source.

Reviewer.

Run.

Method.

Leadership should not see 60 rows.

Analyst needs them.

Layering.

What cadence?

Weekly:

Critical errors.

Dynamic product/service accuracy.

Campaign/query issue.

Monthly:

Core query.

Shortlist.

Citation.

Source.

Commercial signal.

Quarterly:

Method review.

Query relevance.

Evidence audit.

Platform changes.

Don't run everything daily.

AI answers vary.

Daily over-monitoring can create noise.

Use cadence based on business risk.

Marketing KPI also needs owner

Mention/Citation: GEO/search.

Accuracy: shared with domain owner.

Source: PR.

Product: product/commerce.

Price: commercial.

Location: ops.

Critical legal: legal.

Marketing coordinates dashboard.

Does not own truth.

This is major shift.

AI Search forces dashboard to become cross-functional.

Dashboard should end with decisions

Not just trend.

This month:

2 critical errors.

Owner.

Action.

5 source conflicts.

Outreach.

Shortlist dropped on enterprise query.

Reason hypothesis.

Evidence gap.

New buyer query cluster.

Content opportunity.

Google generative AI impressions up.

Compare clicks/leads.

Paid AI campaign separately.

Every KPI needs next action.

Otherwise it's presentation.

Tim marketing tidak butuh lebih banyak angka.

Mereka butuh angka yang menunjukkan apakah mesin jawaban membantu atau merusak buyer journey.

KPI baru yang paling penting sebenarnya bukan visibility.

It is reliable visibility.

Brand appears in relevant questions.

Description is correct.

Source is defensible.

Paid is separated.

Error is corrected.

Buyer can continue.

Commercial impact is observed where possible.

That is dashboard worth building.

Not because AI Search replaces marketing metrics.

Because it adds a new layer between brand information and customer decision, and marketing needs to see whether that layer is working in their favor.

Ada satu KPI tambahan yang layak dipisahkan: Buyer Question Gap.

Ini mengukur berapa banyak pertanyaan penting yang belum memiliki source resmi yang cukup jelas.

Contoh:

“Apakah produk punya API?”

Jawabannya ada di sales deck, tidak ada di website.

“Bagaimana data dihapus setelah kontrak?”

Jawabannya hanya ada di email legal.

“Apakah cabang buka Minggu?”

Jawabannya ada di Instagram Story lama.

Gap seperti ini menjelaskan kenapa AI mencari source lain atau memberi jawaban lemah.

Track:

Total critical buyer questions.

Answered clearly in canonical public source.

Answered only in fragmented source.

Not answered.

KPI ini sangat actionable.

Marketing tidak harus menunggu AI salah.

Mereka bisa menutup information gap lebih dulu.

Dashboard juga perlu satu kolom Confidence in Measurement.

High:
platform-native data atau raw observation lengkap.

Medium:
observational sample dengan limitation.

Low:
inference atau attribution lemah.

Ini mencegah leadership membaca semua angka seolah punya reliability sama.

Contoh:

Google Generative AI impressions dari Search Console punya provenance platform.

“AI influenced 30 persen direct traffic” mungkin inference lemah.

Jangan taruh dengan style visual yang sama tanpa note.

Measurement quality adalah KPI untuk measurement system itu sendiri.

Tim yang tahu kualitas datanya akan membuat keputusan lebih baik daripada tim yang punya dashboard lebih penuh tetapi tidak tahu mana angka yang benar-benar dapat dipertanggungjawabkan.