Schema untuk Research Page: Report, ScholarlyArticle, atau WebPage

Research page adalah salah satu halaman yang paling mudah over-schema.

Ada kata “research”.

Ada methodology.

Ada chart.

Ada PDF.

Lalu tim mulai debat:

“Report?”

“ScholarlyArticle?”

“WebPage?”

“Article?”

“Dataset?”

Kadang jawaban terbaik justru bukan memilih satu type.

Satu research URL bisa menjadi WebPage yang menerbitkan atau menjelaskan sebuah Report. Di kasus lain, main entity-nya ScholarlyArticle. Di kasus lain lagi, halaman cuma landing page untuk kumpulan research dan bukan research work itu sendiri.

Schema.org type harus mengikuti object nyata.

Bukan label menu website.

Kalau navigation bernama “Research”, semua halaman di bawahnya tidak otomatis ScholarlyArticle.

Mulai dari Pertanyaan: Karya Apa yang Sedang Diterbitkan?

Contoh 1:

“AI Search Visibility Indonesia Q3 2026.”

Isinya:

Executive summary.

Method.

Sampling.

Findings.

Limitations.

Appendix.

Produced by GEO.or.id.

PDF download.

Ini sangat natural sebagai Report.

Schema.org defines Report as a report generated by governmental or non-governmental organization.

Report is subtype of Article/CreativeWork.

Institutional report fits.

Contoh 2:

“Retrieval Source Stability in Generative Search Systems.”

Format:

Abstract.

Authors.

Affiliations.

Methods.

References.

Peer-review or scholarly publication context.

DOI.

Journal/conference context maybe.

This is ScholarlyArticle.

Schema.org defines ScholarlyArticle simply as a scholarly article.

Academic character matters.

Not just “serious writing”.

Contoh 3:

`/research/`

It lists:

Reports.

Datasets.

Research articles.

Methods.

This is probably CollectionPage/WebPage or DataCatalog-like structure depending exact purpose.

Not Report.

Not ScholarlyArticle.

It's a hub.

Research Hub Bukan Research Work

This distinction prevents a lot of bad markup.

`/research/ai-search/` could be topic hub.

Contains five articles.

Two reports.

One dataset.

The page itself:

CollectionPage or WebPage.

Each linked work:
separate entity/page.

Don't mark hub as one giant Report.

Otherwise publisher/author/date confuse.

If page has no singular study, no singular research object.

Hub is navigation.

Research work is object.

WebPage Tetap Ada di Semua Kasus

Technically, the URL is a WebPage.

The published work can be mainEntity.

Conceptually:

WebPage
→ mainEntity
→ Report

or

WebPage
→ mainEntity
→ ScholarlyArticle

This modeling is useful because page-level properties and work-level properties can differ.

WebPage:
URL.
breadcrumb.
site relationship.

Report:
name.
author.
publisher.
datePublished.
reportNumber.
about.
citation.

ScholarlyArticle:
headline/name.
authors.
publisher.
date.
pagination if relevant.
citation.

You don't need complex graph on small site, but semantic separation is clean.

Report Tidak Harus Punya PDF

Report is content type, not file format.

HTML report can be Report.

PDF can be distribution/encoding of same work.

Don't say:

“PDF = Report.”

A PDF can be:

Brochure.

Whitepaper.

Invoice.

Manual.

Report.

Article.

Dataset documentation.

Format doesn't define semantics.

If HTML page presents full report and PDF mirrors it, model one Report with associated encoding/download as appropriate.

Don't create duplicate Report entities unless they are distinct versions.

Whitepaper Itu Apa?

“Whitepaper” is marketing/editorial label.

Could semantically be:

Report.

TechArticle.

Article.

CreativeWork.

Depends content.

Whitepaper with:

Research method.

Industry data.

Findings.

Institutional publisher.

Report reasonable.

Whitepaper that is:

Product thought leadership.

No method.

No original data.

Mostly argument.

Article or TechArticle depending content.

Don't upgrade whitepaper to Report because PDF has cover page.

Substance matters.

ScholarlyArticle Butuh Scholarly Context

A company blog post with 50 references isn't automatically ScholarlyArticle.

Ask:

Was it written as scholarly publication?

Does it have academic authorship/affiliation?

Formal method?

Citations/references?

Publication context?

Peer review, journal, conference, preprint, institutional scholarly repository?

No single checklist universally defines scholarship, but context matters.

For normal corporate research, Report often more honest.

Don't use ScholarlyArticle as prestige schema.

Schema type doesn't confer academic legitimacy.

If research is not scholarly work, don't cosplay academia through JSON-LD.

Author vs Research Team

Report can have Organization author.

Person authors.

Research team.

Publisher Organization.

Be accurate.

If methodology was developed by team, but CEO wrote foreword, CEO isn't automatically author of whole report.

If anonymous institutional report:
Organization can be author/publisher.

If named analysts:
Person objects.

No need invent researcher profiles solely for schema.

Authorship should reflect actual work.

This is provenance.

Research Page with Dataset

Very common.

Report:
“AI Visibility Study 2026.”

Dataset:
CSV observations.

Methodology page.

The relationship:

Report uses/analyzes Dataset.

Dataset may cite Report as related publication.

Landing page may link both.

Do not make Dataset sameAs Report.

Different objects.

If dataset altered from original, use provenance relationships appropriately.

Research architecture becomes much stronger when work and data are separated.

User can read findings.

Researcher can inspect data.

AI system can potentially retrieve both as sources.

But don't promise citation effect.

Research Methodology Bisa Menjadi CreativeWork Sendiri

Large recurring study:

Methodology v1.

Reports quarterly.

Datasets monthly.

Method page stable.

Could model method as CreativeWork/TechArticle depending format.

Internal graph:

Method → usedBy → Report.

Report → basedOn → Dataset.

Dataset → createdUsing → Method, if internal ontology.

Public Schema.org vocabulary may not map every relation cleanly.

Don't invent properties.

Natural links and citations can explain.

Machine semantics can be partial.

Correct partial graph > fake complete graph.

ResearchProject Juga Ada, tetapi Jangan Campur Project dengan Output

A research project is activity/initiative.

Report is output.

ScholarlyArticle is output.

Dataset is output.

Project page can describe:

Team.

Goal.

Period.

Funding.

Outputs.

Don't mark project itself as Report.

Schema.org has types related to Project/ResearchProject in evolving vocabulary, but platform support varies.

If using less common types, treat as general semantic markup, not Google feature guarantee.

Always check current consumer docs.

Research Organization Bisa Punya Hierarchy

Organization:
GEO.or.id.

Research unit:
AI Search Observatory.

Project:
Indonesia AI Visibility Study.

Report:
Q3 Report.

Dataset:
Observation Dataset.

Person:
Researcher.

Now each node has role.

Website architecture can reflect:

/research/
/research/ai-visibility/
/research/ai-visibility/q3-report/
/research/ai-visibility/dataset/

This is entity-first.

No need all nodes public if organization small.

But role clarity helps.

Tanggal Research Harus Dipisahkan

Study period:
1–31 July 2026.

Analysis date:
August 1–5.

Publish date:
August 9.

Modified date:
August 20.

These are not same.

Report page should show.

If dataset rolling:
version.

If paper revised:
version.

AI search is time-sensitive.

Research without date context becomes stale quickly.

Schema can encode publication/modification, but visible page should state study period.

Don't rely on datePublished to represent data period.

Very common mistake.

Citation dan Reference Harus Bisa Diverifikasi

ScholarlyArticle typically has references.

Report too.

Schema.org `citation` can link/reference other CreativeWorks.

But don't mark random related articles as citations unless report actually cites them.

Visible references should align.

If report uses 20 sources, maybe structured data doesn't need all 20.

No need maximalism.

But selected citations should be real.

Don't manufacture academic appearance through citation properties.

Research integrity first.

ReportNumber Berguna untuk Institutional Reports

Schema.org Report has `reportNumber`.

If organization has publication numbering:

GEO-RPT-2026-Q3-01.

Useful.

Stable identifier.

PDF and HTML share.

Citation easier.

If no report numbering system, don't invent fake number only to fill property.

Could use internal `identifier`.

But design publication system deliberately.

Stable identifiers are valuable for research archive.

No universal requirement.

WebPage-Only Bisa Jadi Pilihan yang Benar

Not every “research” page has enough information to classify work confidently.

Maybe preliminary notes.

Interactive exploratory tool.

Method teaser.

Announcement.

If main object unclear:
WebPage plus visible Organization/author as appropriate.

Don't force specific type.

Semantic humility is good schema strategy.

A generic correct type is better than specific wrong type.

Google's general structured-data guidelines explicitly encourage the most specific applicable type, but “applicable” is key.

Specificity without truth is misclassification.

Google Search Support dan Schema.org Harus Dipisahkan

Google's Article structured-data documentation supports specific Article types for its Search feature.

Schema.org includes many Article subtypes such as ScholarlyArticle and Report.

These semantic types can still be valid Schema.org.

But do not assume every subtype receives identical Google Search treatment.

If target is a Google rich result, follow current Google documentation.

If target is semantic interoperability, Schema.org type choice can be broader.

One implementation can sometimes model both levels.

But test.

Don't claim feature eligibility beyond docs.

Research Page dan AI Search

Research content is attractive for GEO because it can provide original evidence.

But don't make leap:

“Report schema increases AI citation.”

No universal proof.

A well-structured research page may have attributes that humans and retrieval systems find useful:

Clear title.

Date.

Method.

Data.

Findings.

Limitations.

Author.

Sources.

Stable URL.

Dataset.

These are strong publishing practices.

Structured data can describe them.

Citation selection remains external.

Measure.

Don't use schema as causal explanation.

How to Choose: A Practical Decision

Question 1:
Is this URL primarily a hub/interface?
WebPage/CollectionPage.

Question 2:
Is the main work an institutional research report?
Report.

Question 3:
Is the main work a scholarly article in academic context?
ScholarlyArticle.

Question 4:
Is it a regular editorial article discussing research?
Article/BlogPosting.

Question 5:
Is there a separate body of structured data?
Dataset as separate object.

Question 6:
Are multiple works presented?
WebPage with multiple entities/links.

This decision tree is enough for most sites.

Don't multi-type everything because you're unsure.

If one object is genuinely a Report, remember Report already sits under Article in Schema.org hierarchy.

You don't need to add `Article` just for prestige unless consumer requirements specifically demand modeling choices.

One primary semantic type can be clearer.

Research Page Quality Gate

Before markup, confirm visible page has:

Clear research title.

Publisher.

Author/team.

Purpose.

Method.

Period.

Sample.

Findings.

Limitations.

References.

Version/date.

Related dataset if any.

Download.

Contact.

If none:
maybe it's not report.

If just announcement:
WebPage/Article.

Schema shouldn't create research characteristics absent from content.

Skenario yang Menyelamatkan dari Over-Schema

Marketing:
“Page ini research, pakai ScholarlyArticle biar AI anggap akademik.”

Research lead:
“Ini market survey internal, bukan scholarly publication.”

Developer:
“Report?”

“Ya. Ada method, sample, findings, limitations.”

SEO:
“Google support?”

Developer:
“Schema.org valid, tapi kita tetap cek feature docs. Jangan klaim rich result khusus.”

Now everyone uses same language.

That's maturity.

Schema type becomes description.

Not persuasion.

The best research schema feels boring because it matches what the reader already sees.

A real report is marked Report.

A real scholarly paper is marked ScholarlyArticle.

A hub remains WebPage.

A dataset remains Dataset.

No inflation.

No AI theater.

Just a clean representation of research that actually exists.