EXP-008: Topic Cluster Traffic Efficiency — Where Are We Over-Producing?

Experiment ID: EXP-008 Status: COMPLETE Date: 2026-05-14 Data Window: 2026-04-03 to 2026-05-14 (41 days, full site lifetime) Owner: Effloow Experiment Lab Builds on: EXP-006 (content type), EXP-007 (timing × topic heat)

1. Hypothesis

Primary: Content production at Effloow is misaligned with traffic demand. Clusters that over-produce (Model Releases, AI Frameworks, AI IDE tools) generate far fewer views per article than under-produced clusters (MCP Ecosystem, Local LLM Self-Hosting, LLM Production). Reallocating 30% of production from the bottom three clusters to the top three would increase total article traffic by 50–70% within 30 days.

Null hypothesis: There is no statistically meaningful difference in traffic efficiency across topic clusters — production mix does not predict traffic.

Business question: Given that Effloow produces 3 articles per day, where should those slots go to maximize traffic and SEO authority per published piece?

2. Data Sources

3. Methodology

3.1 Topic Cluster Classification Rules

Each article was assigned to one cluster using a priority-ordered ruleset applied to its slug. First matching rule wins.

3.2 Traffic Attribution

GA4 top_pages provides the top 10 pages by monthly views. Only article paths (/articles/*) were included; homepage, /services, and index pages were excluded. Articles not in the top 10 are assumed to have < 36 views (below the lowest ranked article in the dataset) — likely near zero given the site's early stage.

Total measurable article traffic = 393 views across 7 articles in top_pages.

3.3 Efficiency Metric

Traffic Efficiency (views/article) = Total cluster views ÷ Total cluster article count

This reflects expected traffic per article produced in that cluster, given current organic search and referral signals.

4. Corpus Classification Results

4.1 Article Count by Cluster (all 168 articles)

4.2 Cluster Article Breakdown

MCP Ecosystem (14 articles)

Model Releases (40 articles — sample)

5. Traffic Performance Data

5.1 GA4 Top Pages — Article Traffic (May 2026 snapshot)

Total measurable article traffic: 393 views Coverage rate: 7 of 168 articles (4.2%) drive 100% of measured traffic

5.2 Traffic Efficiency by Cluster

6. Production vs Traffic Gap Analysis

6.1 Allocation vs Performance Matrix

Reading: Positive gap = cluster outperforms its production share. Negative = produces more than it earns in traffic. Efficiency ratio = traffic share ÷ production share.

6.2 Visual: Production vs Traffic Misalignment

Cluster              | Produced | Traffic | Direction
---------------------|----------|---------|----------
MCP Ecosystem        | ████░░░░ | ████████████████████ | Under-produced
Local LLM            | ██████░░ | █████████████████░░░ | Under-produced
LLM Production       | ███░░░░░ | ████████░░░░░░░░░░░░ | Under-produced
AI Tool Reviews      | ██░░░░░░ | ██████░░░░░░░░░░░░░░ | Under-produced
Model Releases       | █████████ | ████░░░░░░░░░░░░░░░░ | Over-produced
AI Frameworks        | ████████░ | ░░░░░░░░░░░░░░░░░░░░ | Over-produced
AI IDE / Tools       | ███████░░ | ░░░░░░░░░░░░░░░░░░░░ | Over-produced
DevOps               | ███░░░░░░ | ░░░░░░░░░░░░░░░░░░░░ | Over-produced

7. Key Findings

Finding 1: Four Clusters Absorb 78% of Production, Deliver 41% of Traffic

The bottom four clusters by efficiency (Model Releases, AI Frameworks, AI IDE, DevOps) together consume 64.3% of production but deliver only 9.2% of traffic. This is the largest production-traffic misalignment in Effloow's content history.

Finding 2: MCP Ecosystem is the Highest-ROI Cluster at 8.8 Views/Article

With only 14 articles (8.3% of corpus), MCP content captures 31.3% of article traffic. The average MCP article delivers 9.8× more traffic than a Model Release article (8.8 vs 0.9 views/article). This efficiency gap is the clearest signal in the dataset.

Why MCP over-performs:

• High search intent ("best MCP servers", "build MCP server") with still-low SERP competition
• Evergreen content — MCP adoption is growing, not contracting
• Cross-platform authority: MCP tutorials attract backlinks from tool documentation
• The top-mcp-servers article has been live for 41 days and continues accumulating organic traffic

Finding 3: Model Releases Over-Produce for Zero Long-Term SEO Value

40 articles (23.8% of corpus) cover new model releases. These articles suffer a structural flaw: they are time-sensitive but not time-durable. A "GPT-6 developer guide" written on release day competes with:

• Official OpenAI documentation
• Major tech publications (TechCrunch, The Verge)
• GitHub READMEs from official repos

This content wins only in the 24–72 hour post-release window. After that, it loses search position to authoritative sources. The 40 model release articles have generated just 36 measurable views total — 0.9 views/article — the worst efficiency of any producing cluster.

Finding 4: The April 12–17 Cohort is Now Earning Organic Traffic

Unlike EXP-006 which found only the first-48-hour launch cohort visible in GA4, the May 2026 data shows articles from April 12–17 now leading traffic. This is evidence that organic SEO is activating after ~30 days of indexing. Articles from April 3–4 (Gemma 4, Ollama) remain visible, confirming that setup/self-hosting guides have long traffic half-lives.

Critical implication: The traffic signal in this dataset is now partly organic, not purely cross-post-driven. This makes the cluster efficiency numbers more predictive of future organic performance.

Finding 5: AI IDE Articles (26 total) Have Zero GA4 Visibility

Despite being 15.5% of the corpus and covering high-intent topics (Cursor vs Windsurf, Claude Code guide), zero AI IDE articles appear in top_pages. Possible causes:

1. Extreme SERP competition — "Cursor vs Windsurf" is contested by major outlets
2. Content currency decay — Cursor 2.0 articles from April are already outdated as of May
3. No backlink structure — AI IDE comparisons require external site authority to rank

8. Traffic Reallocation Simulation

If the next 30 days of production reallocated from the bottom clusters to the top performers:

Scenario: Shift 30% of slots from Model Releases + AI Frameworks → MCP + Local LLM

Current rate: ~3 articles/day → 90 articles over 30 days

Projected additional traffic in 30 days: +206 views (+52% uplift)

Caveat: This simulation assumes constant traffic efficiency per article. As cluster saturation increases, marginal returns diminish. MCP efficiency will decay as internal competition increases (already 14 articles targeting similar keywords).

9. Recommendations

Immediate Content Mix Adjustment

Recommended Production Mix (Next 30 Days)

10. Experiment Assessment

11. Appendix: Cross-Post Gap × High-Efficiency Clusters

MCP articles with cross-post gaps (immediate action available):

Local LLM articles with cross-post gaps:

These 4 articles alone represent the highest traffic + lowest distribution coverage. Cross-posting them within 24h could add 100–150 additional referral views this month.

EXP-008 complete. Next: EXP-009 — MCP Cluster Saturation Point (run 2026-06-14 with 60-day data window and MCP article count ≥ 20).