EXP-009: Word Count Is the Biggest Lever — Pipeline Articles Are 43% Too Short

Experiment ID: EXP-009
Status: COMPLETE
Date: 2026-05-21
Data Window: 2026-04-03 to 2026-05-21 (48 days, full site lifetime)
Owner: Effloow Experiment Lab

1. Hypothesis

Primary: Articles with 3,000+ words generate measurably more monthly traffic than shorter articles, even when controlling for age (indexing time). The current automated pipeline systematically produces articles averaging 2,126 words — below the 3,000-word threshold where traffic lift is observable.

Null hypothesis: Word count does not significantly correlate with monthly page views when controlling for publication age.

Business question: What single change to the content factory will yield the largest traffic increase per article published?

2. Data Sources

Methodology note: Word counts include YAML frontmatter (≈50 words uniform across all articles). All traffic figures are monthly views from the GA4 snapshot captured 2026-05-21.

3. Experiment Design

3.1 Controlling for Age (Indexing Effect)

The primary confounding variable is publication age. Articles published earlier have had more time to accumulate backlinks and index in Google. To isolate the word count effect:

• Threshold: Only articles published ≥ 20 days ago (on or before 2026-05-01) were included in the traffic correlation analysis.
• Reasoning: Google's indexing window for new content is typically 2–4 weeks for a site at Effloow's current DA level. Articles under 20 days old have near-zero chance of appearing in organic results.
• Result: 116 articles qualified (20+ days old). 83 articles were excluded as too recent.

3.2 Population Split

4. Results

4.1 Word Count vs. Monthly Traffic (Age-Controlled)

Analysis restricted to 116 early articles (≥20 days old, all with equal indexing opportunity):

Key finding: Crossing the 3,000-word threshold is associated with a 3.2× increase in average monthly views per article.

4.2 GA4 Top Articles Analysis

All 7 GA4-tracked traffic articles and their word counts:

Observation: Top traffic articles average 3,627 words. Only 1 of the 7 falls below 3,000 words (and it's the second-highest performer — suggesting topic relevance, LLM fine-tuning, also plays a role).

4.3 Pipeline Article Word Count Distribution

Current word count distribution of all 101 pipeline articles (sandbox-poc, paper-poc, tool-scout):

99% of pipeline articles are below the 3,000-word threshold where traffic lift becomes statistically observable.

4.4 Pipeline vs. No-Track Comparison

4.5 Early Pipeline Benchmark (Age-Controlled)

17 pipeline articles are 20+ days old — enough time to be indexed. Their current traffic:

All 17 early pipeline articles show zero tracked views. These articles are in the 1,568–3,051 word range; only 1 exceeds 3,000 words (warp-2, 3,051 words — barely). This is consistent with the overall pattern: articles under 3,000 words capture little to no organic search traffic in the early indexing window.

Caveat: 20-24 days may still be marginal for a domain at Effloow's authority level. The no-track articles gaining traffic are 34-47 days old. A rerun of this analysis at day 40+ on pipeline articles would provide a cleaner comparison. This experiment treats 20 days as a soft floor, not a hard equal-opportunity boundary.

5. Confounding Factors & Limitations

6. Conclusion

The data supports the primary hypothesis: article word count is the single strongest observable predictor of monthly traffic at Effloow's current scale, among the variables available for analysis.

• Articles with 3,001–4,000 words average 4.8 views/month vs. 1.5 views/month for 2,001–3,000 words (+220%).
• Articles with 4,001+ words average 6.0 views/month (+300%).
• The current pipeline outputs articles averaging 2,126 words — 43% below the minimum effective threshold.
• 99% of pipeline articles fall below 3,000 words.

If pipeline word count increased to 3,500 words average (matching no-track top articles), the model predicts monthly traffic per article to increase from ~1.5 to ~4.8 views — a 3.2× lift per article. At 3 articles/day production pace, this compounds to a meaningful monthly total.

7. Recommendations

Immediate (this week)

1. Update agent target length to 3,500 words minimum. All three tracks (sandbox-poc, paper-poc, tool-scout) currently average under 2,200 words. The agent prompts should specify a 3,500-word floor and 4,500-word target.

2. Prioritize depth over breadth for high-competition topics. For topics like "LLM fine-tuning", "self-hosting LLMs", "MCP ecosystem" — these keywords attract volume. A 4,000-word guide with working code examples will outperform a 2,000-word overview.

3. Add a word-count gate in the QA review step. Any article under 2,800 words should trigger a needs-expansion flag before publishing. The review-article agent should enforce this.

Near-term (next sprint)

4. Expand existing pipeline articles that are 20+ days old with zero traffic. The 17 early pipeline articles are already indexed — expanding them from 2,000 to 3,500 words is a lower-cost traffic intervention than publishing net-new articles. Priority candidates: warp-2-agentic-development-environment, huggingface-smolagents-mcp-bridge-guide, minimax-m2-5-cost-performance-api-guide.

5. Design EXP-010 as a controlled A/B length test. Publish 5 articles at 2,000 words and 5 at 4,000 words on the same day, with equivalent topics and keyword difficulty. Measure views at 30 and 60 days.

Structural

6. Investigate topic selection as a second variable. This experiment could not control for keyword difficulty or search volume. The next experiment should use a keyword tool (e.g., Ahrefs estimates or Google Search Console) to normalize by estimated search volume.

8. Data Appendix

Full Word Count Distribution (No-Track, Early Articles)

Category Traffic Distribution (Early Articles with Traffic)

Experiment conducted by Effloow Experiment Lab. Data source: data/metrics.json GA4 snapshot 2026-05-21, content/articles/ filesystem corpus.