GEO Blog Rebuild Case Study: Running the Five-Layer Model on Real Data

The answer first: perfect tech, stuck at “impressions without being chosen”

I dug through cubxxw.com’s real data, and the conclusion is blunt: technical SEO is near-perfect (Lighthouse SEO 100), yet the last 3 months turned 878K impressions into only 852 clicks — a 0.1% CTR at average position 13.2 (page two). The problem isn’t the technical layer; it’s “impressions without being chosen.” This chapter runs the previous four chapters’ five-layer model against those real numbers, diagnosis and rebuild schedule included.

The first four chapters were method; this one is live fire. Every number comes from real Google Search Console and PageSpeed Insights measurements, not a demo.

This is Chapter 5 (Blog Rebuild Case Study) of the Generative Engine Optimization series. It lands the pillar’s five-layer model on my own site’s real data.

1. The real baseline: numbers on the table

First, PageSpeed Insights mobile (real-browser scores):

Then Google Search Console (old domain nsddd.top, last 3 months):

One-line read: the technical base is perfect (the model’s L1), but rankings sit on page two and CTR is abnormally low. Tech isn’t the bottleneck — this is the classic profile of a site with “great SEO, no traffic.”

2. The truth about 878K impressions: demystifying a vanity metric

Sort queries by impressions and the truth appears — the highest-impression queries are all noise unrelated to the blog:

These are impressions scraped from low positions (page two and beyond) — zero clicks, zero value — yet they dilute site CTR to 0.1%. The hard lesson: don’t let “878K impressions” give you a false sense of achievement. Impressions mean standing in the crowd; clicks/citations mean being called on.

Echoing Chapter 2 : these noise queries are “lexical coincidences” (the BM25 lane), but the pages’ passages don’t survive being lifted out — hence impressions with no citation and no clicks.

3. The gold is in clicks: finding the clusters to invest in

Switch the lens to clicks, and the genuinely valuable technical clusters surface:

The two extremes are the most instructive: my-hugo turns 337 impressions into 35 clicks (10.4% CTR) — a model of “content matching intent + extractable passages”; while a thought-notes post turns 87K impressions into 27 clicks (0.03%) — the cautionary opposite.

Strategic conclusion: invest in validated-demand technical clusters — Hugo, AI tools (markitdown/mem0/langgraph/notebooklm/gpt-researcher), Go & engineering practice, TDD — and spend no content on noise queries. This is Chapter 3’s “topic cluster” made concrete.

4. Shining the five-layer model on it, layer by layer

Aligning the real state with the previous chapters’ five-layer model makes the gaps obvious:

Key insight: a perfect L1 fools you into thinking “SEO is great,” but GEO is decided at L2–L5. A perfect technical base is only the entry ticket; the real moat is structure, evidence, and endorsement.

5. A priority-ranked rebuild checklist

Turning diagnosis into a schedule (aligned with the pillar’s 30/60/90):

🔴 P0 · Protect the migration (1 week)

• Keep old-domain nsddd.top 301s for 180+ days; keep both GSC properties and compare curves weekly to confirm equity transfer.
• Re-submit sitemap.xml and news-sitemap.xml on cubxxw.com; “request indexing” for core pages.
• Verify all 813 old-domain traffic pages 301 to the same path on the new domain (a page like markitdown must never 404).

🟠 P1 · Win page one + lift CTR (2–4 weeks)

• Filter GSC for “average position 8–20”; add depth, internal links, and Answer-First openers to those pages (Chapter 3’s craft). Prioritize markitdown, mem0, langgraph.
• Rewrite titles/descriptions of high-impression, low-CTR pages (with numbers/outcome promises), benchmarking my-hugo.

🟡 P2 · Evidence + Schema + clusters (1–2 months)

• Add statistics and external citations to core posts (L3, +25–40% visibility).
• Add FAQPage/HowTo schema to how-to/comparison posts (L4).
• Build “pillar + child + internal-link” clusters around Hugo/AI tools/Go (L4/L5) and distribute core posts to Zhihu/Juejin/HN (Chapter 4’s endorsement play).
• Fix Accessibility (contrast, heading levels, tooltip aria-label) and LCP along the way.

6. The domain migration: don’t let this one step undo everything

An easily-overlooked, veto-power item: cubxxw.com was migrated from the old domain nsddd.top.

• ✅ Change of Address is set in GSC; 301s preserve paths and are verified (nsddd.top/projects/markitdown/ → 301 → cubxxw.com/projects/markitdown/, canonical correct).
• ⚠️ The new-domain property is recent, so Search data is still backfilling — read history from the old domain now, and watch the new domain absorb the equity over the next 1–3 months.
• Key actions: keep old-domain 301s for 180+ days, monitor both properties, verify every redirect. Any 404 or broken link during migration pours all your prior GEO effort down the drain.

7. FAQ

Q: Lighthouse SEO is 100 — why still no traffic? A: Because Lighthouse only tests the L1 technical base. GEO traffic depends on L2–L5 (structure, evidence, quotability, endorsement) and real rankings. A perfect technical score is necessary, not sufficient.

Q: Isn’t 878K impressions a lot — why call it vanity? A: Because most of it comes from off-topic long-tail noise at low positions with zero clicks. What matters is “precise technical queries + clicks,” not total impressions.

Q: What should I watch most during migration? A: Three things — all old-domain 301s valid (no 404s), the two GSC properties’ click curves falling/rising, and core traffic pages (e.g., markitdown) indexing normally on the new domain.

Q: Can I copy this review method to my own site? A: Yes. The flow: PSI for the technical base → GSC sorted by both clicks and impressions to find clusters and noise → the five-layer model for gaps → schedule by P0/P1/P2. Chapter 6 gives low-cost measurement and monitoring tools.

Summary and what’s next

This chapter turned the five-layer model from “how you should do it” into “here’s exactly how my site is, and what to change next.” The core, in one line: a perfect technical score is only the start; the real upside is in structure, evidence, endorsement — and protecting the domain migration.

But once rebuilt, how do you know it worked? The classic “rank + click” fails in the GEO era. The final chapter builds a low-cost “citation rate” measurement system.

• Previous: GEO Trust & Endorsement — E-E-A-T and off-site distribution
• Next (Chapter 6 · Measurement & Tools): prompt testing, AI referral traffic, GSC cross-check, Profound/Peec, and a DIY monitor built on this repo’s own geo:audit/gsc/psi scripts.

Data source: my real measurements of cubxxw.com (and old domain nsddd.top) via Google Search Console and PageSpeed Insights (July 2026). Methodology in the first four chapters of this series.