EXP-010: Search Demand Is the Missing Variable — Why Pipeline Articles Get Zero Traction

Experiment ID: EXP-010
Status: COMPLETE
Date: 2026-05-28
Data Window: 2026-04-03 to 2026-05-28 (55 days, full site lifetime)
Owner: Effloow Experiment Lab

1. Hypothesis

Primary: Organic traffic at Effloow is driven by topic search demand — not by publication age, word count, or content track. Articles covering topics with established organic search queries will outperform articles covering novel/cutting-edge topics regardless of content quality.

Secondary: The evidence-led pipeline tracks (sandbox-poc, paper-poc, tool-scout) are systematically targeting low search-demand topics, creating a structural gap between content production volume and traffic generation.

Null hypothesis: After controlling for publication age, legacy and pipeline articles show no significant traffic difference.

Business question: What is the single biggest optimization lever for Effloow's traffic growth — and can we quantify the opportunity cost of the current pipeline topic selection?

2. Data Sources

3. Methodology

3.1 Content Track Classification

Articles were classified by content_track frontmatter field:

3.2 Age-Adjusted Traffic Rate

To control for publication age, monthly views were converted to views-per-day for each top-performing article:

age_adjusted_rate = total_views / days_since_publication

3.3 Named Entity (NE) Scoring

Titles were scored for named technology/product entities (tools, protocols, model names, brand names). This tests whether keyword specificity in titles correlates with traffic.

4. Results

4.1 Traffic by Content Track

Finding: 100% of measurable traffic comes from legacy articles.

Note: Views represent presence in GA4 top_pages (threshold ~25 views). Pipeline articles may have sub-threshold traffic.

4.2 Top 7 Articles: Age-Adjusted Analysis

The most important table in this experiment. Age-adjusted rate controls for indexing time.

Critical observation: how-we-built is the oldest article (55 days) but ranks 6th in v/day. llm-fine-tuning is the newest of the top 7 (41 days) but leads in v/day by 36%. Age is not the primary driver.

4.3 Named Entity Count vs Traffic

Finding: Named entity count is a weak predictor. best-ai-code-review has NE=4 but only 0.61 v/day. how-we-built has NE=1 but 0.56 v/day. The top performer llm-fine-tuning has NE=2.

NE count correlates slightly with word count (NE=2 articles average 3,164 words) but does NOT directly predict traffic.

4.4 The Search Demand Hypothesis

Analysis of top-7 articles by estimated organic search intent:

Versus pipeline article topics:

5. Interpretation

5.1 The Core Finding

Effloow's pipeline is optimizing for content novelty while organic traffic is driven by content demand.

The three pipeline tracks have an inherent structural disadvantage:

• paper-poc: Reproduces academic research → readers are researchers, not developers doing Google searches
• sandbox-poc: Tests brand-new tools/techniques → zero established search volume when published
• tool-scout: Covers tools < 2 weeks old → the tool's user base is too small to drive organic search

Legacy articles covered tools with existing, established user bases: Ollama (~2M GitHub stars), LoRA/QLoRA (standard ML technique), MCP (97M monthly installs), Gamma (millions of active users). These are topics with organic search demand that existed before the article was written.

5.2 Age Cannot Explain the Gap

If age were the sole variable, older pipeline articles should be gaining on legacy articles. The oldest pipeline articles (published late April) are ~28 days old. The lowest-performing legacy article (how-we-built, 0.56 v/day) was published April 3. By May 28, pipeline articles have had sufficient indexing time to appear in search results if they had search demand — they simply have no demand to capture.

5.3 Word Count Is a Necessary But Insufficient Condition

EXP-009 proved that 3,000+ words = 3-4x traffic. This experiment shows that word count acts as a floor (too short = no chance), but topic demand is the ceiling. A 5,000-word article about a niche paper-poc will not outrank a 3,000-word article about a mainstream tool.

6. Traffic Opportunity Sizing

Current state:

• 226 articles published
• 303 measured views from 7 articles (top_pages)
• Remaining ~219 articles: collectively < 25 views each
• Monthly total visitors: ~1,979

If the pipeline achieved legacy-equivalent performance (3.37 views/article):

• 136 pipeline articles × 3.37 = 458 additional monthly views
• Total: 303 + 458 = 761 views (2.5x current traffic)

If the pipeline targeted high-demand topics (1.0+ v/day top-tier):

• Even 20 articles at 1.0 v/day = 600 additional monthly views
• Total: ~900 views (3x current traffic)

This represents the opportunity cost of the current topic selection strategy.

7. Recommendations

Immediate (this sprint)

1. Add search demand gate to topic selection: Before a topic enters the pipeline, require evidence of existing organic demand. Acceptable signals:

   • Google Trends: upward slope in the past 30 days
   • Tool has ≥ 1,000 GitHub stars (established user base)
   • Keyword appears in developer forums/communities with > 50 discussions
   • Topic is a question people actually Google ("how to X with Y", "best X for Y")

2. Reclassify paper-poc track: Most paper-poc articles cover academic innovations with near-zero developer search volume. Pivot this track to cover papers that validate or explain high-demand concepts (e.g., "Why LoRA works — the paper behind the technique").

3. Extend sandbox-poc cool-down period: Tools < 4 weeks old have insufficient search volume. Implement a 4-week embargo before writing a tool-scout article (let the tool accumulate organic mentions first).

Short-term (next 2 sprints)

4. Create a demand-first content calendar: Identify 20 topics with high established demand that Effloow does not currently cover well. Use these to fill the pipeline slots that would otherwise go to low-demand paper-pocs.

5. Retroactively expand top legacy articles: llm-fine-tuning (2,770 words → target 4,000+) and ollama-open-webui (3,262 words → target 4,500+) are already ranking — expanding them can compound existing traffic.

Strategic (next quarter)

6. Two-tier pipeline architecture:
   • Tier 1 (Demand-Led, 70%): Topics with proven search demand → full 3,500+ word treatment → primary traffic engine
   • Tier 2 (Innovation-Led, 30%): Current paper-poc/sandbox-poc coverage → shorter (1,500 words), framed as lab notes, not expected to rank organically

8. Limitations

• GA4 top_pages only shows top 10 entries. Pipeline articles may have sub-25-view traffic not captured in this analysis.
• "Search demand" is inferred qualitatively — this experiment does not use keyword volume tools (SEMrush, Ahrefs). A follow-up experiment with actual search volume data would strengthen the hypothesis.
• The legacy vs pipeline comparison is confounded by both age and author intent (legacy articles were often written with more considered topic selection).

9. Next Experiment

EXP-011 (proposed): Keyword Volume Validation — Pull actual Google Search Console impression data for the top 20 legacy articles. Confirm whether high-traffic articles have significantly more impressions (search appearances) than low-traffic articles, and at what keyword positions. This would convert EXP-010's qualitative search demand hypothesis into a quantitative, falsifiable claim.

Experiment conducted 2026-05-28. Data sources: data/metrics.json, content/articles/*.md. Analysis: automated Python pipeline. No external keyword tools used.