Question: The “topic cluster” model assumes Google rewards comprehensive coverage of related subtopics by associating them through internal linking. But if Google actually evaluates topical authority at the entity level rather than URL structure level, pillar pages become architecturally irrelevant. What observable SERP patterns would confirm or refute whether Google’s authority signals operate at entity-graph level versus crawl-path level, and how would each scenario change your internal linking strategy?
The Assumption Worth Testing
Topic clusters became orthodoxy around 2017. Build pillar page, link to cluster content, link back. Google supposedly learns topical relationships from this architecture and rewards comprehensiveness.
Hidden assumption: Google needs your internal links to understand that “sprint planning” relates to “project management.”
Counter-hypothesis: Google’s knowledge graph already encodes entity relationships from the entire web. Your architecture teaches nothing. Internal links serve crawl efficiency and PageRank distribution. Topical authority comes from content demonstrating entity expertise, not from URL structure.
Stakes: if topic clusters work through the claimed mechanism, information architecture is an SEO investment. If they work through confounded variables (sites building clusters also produce better content), architecture is ceremonial overhead.
Why Correlation Evidence Is Insufficient
Most topic cluster advocacy cites correlation: “we built clusters and rankings improved.” This proves nothing about mechanism.
Sites that build deliberate topic clusters also tend to:
- Produce more comprehensive content
- Update more frequently
- Attract more external links through better content
- Have higher editorial standards
- Invest more in keyword research
Any of these confounds could explain ranking improvements. Architecture might be epiphenomenal: the discipline producing good architecture also produces good content independently.
Isolating architecture’s causal contribution requires controlling for these confounds. Most “case studies” don’t.
Observable Predictions
The two models make different predictions:
Crawl-path model predicts:
- Orphan pages (minimal internal links) systematically underperform
- Architecture quality correlates with ranking after controlling for content quality
- Changing internal link structure changes rankings for linked pages
- Sites with equivalent content but better architecture outrank competitors
Entity-graph model predicts:
- Orphan pages can rank if content quality is high
- Architecture quality shows no independent correlation with ranking
- Internal link changes affect crawl efficiency but not topical authority scores
- Content quality and external links explain ranking variance; architecture adds nothing
Test Design
Test 1: Orphan page analysis
Identify pages with <5 internal links pointing to them. Compare ranking performance to well-linked pages on equivalent keywords.
Controls needed:
- Content quality (word count, depth, freshness)
- External link profile
- Page age
- Keyword difficulty
Sample minimum: 20 orphan pages across 3+ topics. Below this, individual page factors confound.
Crawl-path prediction: Orphan pages rank >40% worse on average after controlling for confounds.
Entity-graph prediction: No significant performance difference (<15% variance).
Challenge: orphan pages are often orphaned because they’re low quality. Selection bias. Filter sample to pages orphaned for structural reasons (site migration artifacts, forgotten content) rather than quality reasons.
Test 2: Architecture manipulation
Create two content groups on your own site:
Group A: Deliberate pillar/cluster architecture. Central pillar, 8-12 cluster pages, comprehensive interlinking, hierarchical URL structure.
Group B: Equivalent topic coverage, flat structure. Same pages, same depth, same external link effort. Single directory, minimal interlinking, no hierarchy.
Control variables within 15%: word count, external links, publication timeline, author signals.
Timeline: 6-9 months. Shorter tests don’t allow ranking signals to stabilize.
Crawl-path prediction: Group A outperforms by >30% average ranking improvement.
Entity-graph prediction: <15% variance between groups.
Test 3: Link structure change impact
Take a well-ranking page. Remove 80% of internal links pointing to it. Monitor rankings for 90 days.
Crawl-path prediction: Rankings drop significantly as topical authority signals decrease.
Entity-graph prediction: Rankings hold (minor fluctuation) because authority exists in content/external signals, not internal link topology.
Control: don’t change anything else about the page or site during test period.
The Knowledge Graph Recognition Test
Entity-graph model implies Google associates domains with entities independently of site architecture. Test for this association:
Create content targeting an entity Google has in its knowledge graph. Do not build cluster architecture. Single comprehensive page, minimal internal linking.
Monitor:
- Does page rank for entity-related queries?
- Does Google’s knowledge panel ever cite your content?
- Do “People also ask” boxes include your content for entity questions?
- Does Google autocomplete incorporate your terminology or framing?
If entity-graph operates: single well-optimized pages achieve entity association without architectural support.
These proxy signals (knowledge panel citation, PAA inclusion, autocomplete influence) indicate direct entity-graph association, not just ranking factor accumulation.
Second-Order Considerations
Content production correlation: Sites building topic clusters produce more content. More content means more ranking opportunities, more internal link targets, more external link magnets. Separating “architecture effect” from “content volume effect” is difficult.
Control by matching content volume across test groups. If Group B (flat structure) has same page count as Group A (clustered), you isolate architecture.
Editorial attention correlation: Deliberately architected content often receives more editorial polish. Writers working within cluster frameworks think more systematically about coverage gaps.
This is a feature, not a confounder. If topic cluster methodology improves content quality as a side effect, that’s valuable even if architecture itself does nothing for Google.
PageRank distribution effect: Internal links do distribute PageRank. A cluster architecture might concentrate PageRank on pillar pages, helping them rank for competitive head terms.
This isn’t the claimed “topical authority” mechanism. It’s standard PageRank mechanics. If your tests show architecture effects, determine if they’re PageRank distribution (affects specific linked pages) or topical authority (affects all pages on topic).
Test: compare ranking changes for linked pillar pages vs unlinked cluster pages after architecture changes. If only linked pages change, PageRank distribution. If all topically-related pages change, topical authority signal exists.
Strategic Implications by Outcome
If entity-graph model confirmed:
Stop: Building elaborate architecture primarily for SEO. The overhead doesn’t produce ranking advantages through topical authority mechanisms.
Continue: Internal linking for user navigation, crawl efficiency, and deliberate PageRank distribution to priority pages. These functions remain valuable.
Redirect effort: Focus on comprehensive entity coverage regardless of URL structure. If you want authority on “project management,” cover the entities Google associates with that concept. Coverage matters, organization doesn’t.
If crawl-path model confirmed:
Invest: In deliberate information architecture. Pillar/cluster structures produce measurable ranking advantages.
But verify mechanism: Is it topical authority or PageRank concentration? If purely PageRank, you can achieve same effect with simpler link structures. Full cluster architecture only justified if topical authority signal exists independently.
If hybrid (likely):
Internal links affect PageRank distribution (measurable, mechanical). Topical authority comes from content and external signals (entity-graph). Architecture influences rankings through PageRank, not through teaching Google topical relationships.
Optimal strategy: use internal linking strategically for PageRank distribution to priority pages. Don’t build elaborate clusters expecting Google to learn relationships from your architecture.
Falsification Criteria
Entity-graph model fails if:
- Architecture changes produce ranking changes for pages not directly linked (suggests topical authority propagation)
- Orphan pages systematically underperform after controlling for all confounds
- Competitor analysis shows architecture quality correlates with ranking after controlling for content/links
Crawl-path model fails if:
- Orphan pages rank comparably to well-linked pages with equivalent content
- Architecture manipulation produces no measurable ranking change
- Knowledge panel/PAA inclusion occurs without cluster architecture
Run falsification tests before committing to architecture investment. Most SEO teams assume topic clusters work without testing. That assumption may be costing you resources for ceremonial structure.