Just implemented semantic clustering and saw 3x improvement in AI citations - here's exactly what we did

Love seeing semantic clustering applied to content strategy. Let me add the technical perspective.

Why this works:

AI systems understand content through:

1. Vector representations - Content becomes mathematical points in space
2. Similarity calculations - Cosine similarity finds related content
3. Entity recognition - Named entities are connected
4. Contextual understanding - Surrounding content provides meaning

When your content is semantically clustered:

The AI sees: “This site has 15 interconnected pieces on [topic], all referencing each other, with consistent entity usage.”

vs. scattered content: “This site mentions [topic] in random places, unclear expertise level.”

Technical implementation tips:

1. Use sentence transformers - Better than word-level embeddings for content
2. t-SNE for visualization - See your clusters before restructuring
3. Hierarchical clustering - Reveals sub-topics naturally
4. Silhouette score - Validates cluster quality

The math backs up the results you’re seeing.

We did this at scale. 1,200 articles, 45 clusters. Here’s the process:

Phase 1: Audit (2 weeks)

• Export all content URLs and titles
• Pull metadata (dates, authors, categories)
• Identify existing internal links

Phase 2: Clustering (3 weeks)

• Used Keyword Insights for initial grouping
• Manual review and adjustment
• Identified pillar topics

Phase 3: Restructuring (8 weeks)

• Created/updated pillar pages
• Rewrote internal links with entity-focused anchors
• Added schema markup
• URL restructuring where needed

Phase 4: Measurement (ongoing)

• Am I Cited for AI citation tracking
• GSC for ranking changes
• Traffic pattern analysis

Results at 6 months:

• 67% increase in AI citations
• 23% increase in organic traffic
• 40% increase in pages per session

The internal linking was the biggest driver. AI follows link patterns.

Enterprise perspective - semantic clustering at scale is different.

The challenges:

1. Content sprawl - Thousands of pages, multiple authors
2. Governance - Who owns cluster strategy?
3. Technical debt - Legacy URLs, redirect chains
4. Cross-team alignment - Product, marketing, support all create content

Our framework:

Entity → Cluster → Pillar → Spokes → Cross-links
   ↓         ↓        ↓         ↓         ↓
Define   Group    Create   Support   Connect

Governance model:

• Content council owns cluster strategy
• Each cluster has a designated owner
• Quarterly content audits
• Automated link suggestions via CMS

The payoff:

When AI queries our industry topics, we get cited ~35% of the time. Before clustering: ~12%.

But it took 18 months and significant investment.

Small business reality check.

We have:

• 50 blog posts
• 1 person managing content
• Zero budget for fancy tools

What actually worked:

1. Spreadsheet clustering - Listed all posts, manually grouped by topic
2. Hub pages - Created 5 main topic pages linking to relevant posts
3. Anchor text audit - Made sure links describe destination content
4. FAQ sections - Added Q&A to pillar pages

Time invested: 20 hours over 2 months Tools used: Google Sheets, WordPress, common sense

Results:

AI citations went from “almost never” to “regularly.” Not measuring exact percentages because we don’t have enterprise monitoring, but we see ourselves in ChatGPT responses now.

You don’t need BERT embeddings. You need a logical content structure.

For those who want the technical approach, here’s my Python workflow:

Tools:

• sentence-transformers (embedding)
• scikit-learn (clustering)
• matplotlib (visualization)
• pandas (data handling)

Basic process:

1. Scrape content → clean text
2. Generate embeddings (all-MiniLM-L6-v2 works well)
3. Apply k-means or HDBSCAN clustering
4. Visualize with t-SNE
5. Export cluster assignments

The insight from visualization:

When you plot your content in 2D, you see:

• Natural topic groupings
• Orphan content (unconnected pieces)
• Content gaps (sparse areas in relevant topics)

Pro tip:

Run clustering at multiple granularities:

• 5-10 clusters = high-level themes
• 20-30 clusters = sub-topics
• 50+ clusters = specific entities

The hierarchy reveals your content architecture.

Client pattern I’m seeing across industries:

Companies that succeed with semantic clustering:

1. Have genuine expertise in their topics
2. Commit to comprehensive coverage
3. Maintain content over time
4. Measure AI visibility (not just traffic)

Companies that struggle:

1. Try to game the system with thin content
2. Create clusters without substance
3. Ignore internal linking
4. Don’t measure outcomes

The uncomfortable truth:

Semantic clustering amplifies what’s already there. If your content is authoritative, clustering makes AI see that. If your content is thin, clustering exposes the gaps.

My recommendation:

Before clustering, audit content quality:

• Is each piece genuinely useful?
• Does it contain original insights?
• Would an expert consider it accurate?

Cluster good content first. Improve or remove weak content second.

Entity perspective on semantic clustering:

The entity layer matters most.

When you cluster semantically, you’re really organizing ENTITIES:

• Primary entities (your main topics)
• Supporting entities (related concepts)
• Connecting entities (relationships between topics)

Example for fitness brand:

Primary entity: “Strength Training” Supporting entities: “Progressive Overload,” “Muscle Growth,” “Recovery” Connecting entities: “Exercise Equipment,” “Nutrition,” “Sleep”

Your content cluster should:

• Define each entity clearly
• Explain relationships between entities
• Use consistent entity naming
• Include entity attributes and values

The AI connection:

AI systems build knowledge graphs of entities. Your semantic clustering feeds their understanding. The more clearly you define entities and relationships, the better AI understands your content.

Schema markup makes this explicit. Use Organization, Person, Product, and Article schemas with proper relationships.

Amazing contributions everyone. Here’s my takeaway framework:

The Semantic Clustering Pyramid:

Level 1: Content Quality (Foundation)
   ↓
Level 2: Topical Organization (Clustering)
   ↓
Level 3: Internal Linking (Connections)
   ↓
Level 4: Schema Markup (Explicit Signals)
   ↓
Level 5: AI Visibility (Outcome)

Key lessons from this thread:

1. You don’t need fancy tools - Manual clustering works for small sites
2. Quality comes first - Clustering amplifies content quality (good or bad)
3. Entities are the key - Think in terms of concepts and relationships
4. Internal linking matters most - AI follows link patterns
5. Measure what matters - Track AI citations, not just traffic

Action items for anyone starting:

1. List all content in a spreadsheet
2. Group by topic (manual or automated)
3. Identify gaps and pillar opportunities
4. Create/update pillar pages
5. Implement strategic internal linking
6. Add schema markup
7. Set up Am I Cited monitoring

The 3x improvement was real. But it took 6 months of consistent work. This isn’t a quick win - it’s infrastructure that compounds over time.

Thanks everyone for the incredible insights!