Entity-based content clustering for GEO is outperforming keyword strategy by 4x - anyone else seeing this?

I can explain why clustering works so well for AI.

How AI systems process your content:

1. Indexing - AI crawls and stores your content
2. Entity extraction - Identifies people, places, concepts, brands
3. Relationship mapping - Understands how entities connect
4. Authority scoring - Evaluates depth and breadth of coverage
5. Citation decision - Selects sources for responses

Why clusters win:

With individual pages:

• AI sees scattered mentions
• No clear relationship map
• Authority signal is weak

With entity clusters:

• AI builds a knowledge graph of your content
• Relationships are explicit
• Authority signal is strong

The corroboration effect:

AI systems seek multiple confirmations before citing. A cluster provides internal corroboration:

• Pillar confirms spoke content
• Spokes confirm pillar content
• Cross-links create verification network

It’s like having multiple witnesses telling the same story. AI trusts that more.

The schema markup question is critical. Here’s what the data shows:

With schema vs without:

We tested clusters with and without structured data:

• Without schema: 25% AI citation rate
• With schema: 41% AI citation rate

Why schema matters:

Schema makes entity relationships EXPLICIT. AI doesn’t have to guess.

Essential schema for clusters:

On pillar pages:

{
  "@type": "Article",
  "mainEntity": {...},
  "hasPart": [
    {"@type": "WebPage", "url": "spoke-1"},
    {"@type": "WebPage", "url": "spoke-2"}
  ]
}

On spoke pages:

{
  "@type": "Article",
  "isPartOf": {"@id": "pillar-page-url"}
}

The insight:

Content structure is necessary but not sufficient. Schema markup is the metadata layer that helps AI understand your structure.

Both matter. Together they’re multiplicative.

I’ve implemented clusters for 20+ clients. Here’s the pattern:

Optimal cluster size:

• Minimum: 5 pages (pillar + 4 spokes)
• Ideal: 8-15 pages
• Maximum useful: 25-30 pages

Beyond 30, diminishing returns. Sub-cluster instead.

Cluster depth matters:

Shallow: Pillar → Spokes (one level) Deep: Pillar → Spokes → Sub-spokes (two levels)

For competitive topics, go deep. AI favors comprehensive coverage.

The internal linking rule:

Every spoke links to:

• The pillar (required)
• 2-3 related spokes (contextual)

Pillar links to:

• All spokes (in organized sections)

What kills cluster performance:

• Orphan pages (not linked to cluster)
• Conflicting information between pages
• Inconsistent entity naming
• Poor pillar content

Enterprise perspective on scaling cluster strategy:

The governance challenge:

We have 50+ clusters across 3,000 pages. Managing this requires:

• Cluster ownership (who’s responsible?)
• Content calendars per cluster
• Quality standards
• Regular audits

Our cluster management system:

1. Cluster scorecards - Metrics per cluster
2. Gap analysis - Missing spoke types identified
3. Freshness tracking - When was each piece updated?
4. AI visibility - Am I Cited monitoring per cluster

What we measure:

The insight:

Cluster strategy at scale is an ongoing program, not a project. Budget for continuous maintenance.

SaaS perspective on cluster strategy:

Our cluster map:

Product Category (Pillar)
├── What is [Category]? (Definition)
├── [Category] Benefits (Value prop)
├── How to Choose [Category] (Buyer's guide)
├── [Category] Best Practices (How-to)
├── [Our Product] vs Competitors (Comparison)
├── [Category] for [Use Case] (Segment)
└── [Category] FAQ (Questions)

The competitive edge:

When someone asks ChatGPT “[Category] recommendations,” we get cited because:

• Comprehensive coverage signals authority
• Multiple perspectives on same topic
• Clear expertise demonstration

Real numbers:

Before clusters: Mentioned in 5% of relevant AI queries After clusters: Mentioned in 38% of relevant AI queries

The sales impact:

Demos now frequently mention “I saw you recommended by ChatGPT.” That wasn’t happening before.

The entity layer is what makes clustering work for AI. Here’s why:

Entities vs Keywords:

Keywords: “strength training exercises” Entities: “Strength Training” (concept) → “Exercises” (type) → “Barbell Squat” (instance)

AI understands entities natively.

Knowledge graphs are entity-based. When your content is entity-organized, it maps directly to how AI stores knowledge.

Entity relationship types:

• is-a: Barbell Squat is-a Compound Exercise
• part-of: Compound Exercises part-of Strength Training
• related-to: Strength Training related-to Muscle Growth
• used-for: Barbell used-for Compound Exercises

Your cluster structure should mirror these relationships.

Pillar: Primary entity (Strength Training) Spokes: Related entities and their connections

The naming consistency rule:

Use EXACT same entity names across cluster. “Strength Training” not sometimes “Weight Training” or “Resistance Training.”

Inconsistent naming fragments the entity in AI’s understanding.

Clusters work for local businesses too:

Local cluster structure:

[Service] in [City] (Pillar)
├── What is [Service]? (Definition)
├── [Service] Process (How it works)
├── [Service] Cost in [City] (Pricing)
├── Best [Service] Providers in [City] (Industry page)
├── [Service] for [Customer Type] (Segment)
├── [Service] vs [Alternative] (Comparison)
└── [Service] FAQ (Questions)

Local entity optimization:

Include location entities consistently:

• City name
• Neighborhoods
• Surrounding areas
• Local landmarks

The local AI advantage:

When people ask “best [service] in [city],” AI needs local authority signals. Your cluster provides:

• Service expertise (through comprehensive coverage)
• Local knowledge (through location entities)
• Social proof (through reviews/testimonials on pages)

Results for local client:

Before: Not mentioned in local AI queries After: Cited in 45% of “[service] in [city]” queries

Local clusters work because local queries have less competition.

Incredible insights everyone. Here’s my consolidated framework:

The Entity-Based Cluster Blueprint:

Structure:

Primary Entity (Pillar)
├── Definition Spoke (What is...)
├── Process Spoke (How to...)
├── Comparison Spoke (vs alternatives)
├── Benefits Spoke (Why it matters)
├── Segment Spokes ([Entity] for [Use Case])
└── FAQ Spoke (Questions answered)

Critical Success Factors:

1. Entity consistency - Same names everywhere
2. Comprehensive coverage - 8-15 pages per cluster
3. Strategic internal linking - Every spoke to pillar + related spokes
4. Schema markup - hasPart/isPartOf relationships
5. Ongoing maintenance - Fresh content, regular audits

Why 4x improvement happens:

• AI builds knowledge graph from your structure
• Corroboration effect strengthens authority
• Multiple query patterns matched
• Clear expertise demonstration

Measurement stack:

The bottom line:

Entity-based clustering isn’t just better for AI. It’s better content strategy period. The 4x improvement is real and reproducible.

Thanks everyone for making this thread so valuable!