CreatorDB · The Rebuild

From WordPress Sludge to a Cinematic, Indexable Brand Site.

A ground-up rebuild of creatordb.app — static HTML on the Cloudflare edge, AI-generated creator profiles, market-targeted SEO money pages, and a measurement layer that actually moves the needle.

Lighthouse 92 / 92 / 100 / 100 LCP 2.6s CLS 0 1,350+ URLs

Why we rebuilt · the core motivation

The Legacy Site Was Invisible to LLMs.

This wasn't a cosmetic refresh. More and more buyers now ask ChatGPT, Claude, Perplexity, or Gemini questions like "best creator analytics platform" or "Asia influencer marketing agency" — and the AI gives them a shortlist. When we tested those exact prompts, our old site never came up. The reason: AI tools only read pages they can understand instantly, and our WordPress site needed heavy scripts to even show its text — so the AIs saw a blank page. The rebuild fixes that. Every page is now fully readable the moment it loads.

Before

WordPress + Divi + RevSlider

  • 0 visible headings in raw HTML — every H1/H2 was injected by JavaScript after load
  • 30 MB raw QuickTime hero video (HEVC 10-bit, 8.7 Mbps)
  • Citation tracker reported 0/3 successful citations on "best Asia creator analytics platform"
  • Legacy plugin sprawl, slow TTFB, fragile editing workflow
  • SEO money pages (taiwan, japan, …) buried in JS-rendered Divi modules
After

Static HTML on Cloudflare Pages

  • Semantic HTML with real <h1>, <h2>, alt text, schema.org JSON-LD
  • 3.5 MB optimised hero video + 72 KB poster, served from /assets
  • Every page indexable on first crawl — no JS execution required
  • Edge functions for paywall, on-demand profile generation, search
  • Legacy SEO URLs preserved 1:1 — zero ranking loss on cutover

Google Lighthouse + Core Web Vitals · mobile

Top-Decile Performance, Out of the Box.

Measured on a live niche page. For context, most SaaS marketing sites score 35-65 on performance and 80 on SEO — we're well above that on every metric Google measures.

92
Performance
92
Accessibility
100
Best Practices
100
SEO
Largest Contentful Paint (LCP)
Largest above-the-fold element. Google's "Good" <2.5s.
2.6s
Cumulative Layout Shift (CLS)
Content jumping during load. "Good" <0.1.
0.00
Total Blocking Time (TBT)
JS blocking main thread. "Good" <200ms.
0 ms
First Contentful Paint (FCP)
Pixels first appear. "Good" <1.8s.
2.5s
Time to Interactive (TTI)
Page responsive to taps. "Good" <3.8s.
2.7s

Public profile pages · honest before vs after

Why These Pages Should Index.

Honest Claude-graded indexability score on each side. Caveat: no one can guarantee Google indexes any page — these scores reflect what's controllable about the page itself, not Google's domain-trust signals.

Before · legacy WP / Divi
3/ 10Barely indexable
Crawlable HTML (no JS dep) Headings JS-rendered
Semantic structure Divi div-soup
Schema.org coverage Yoast generic only
Unique content per page Same template everywhere · no bio · no FAQ · no news
URL structure~ Plugin paths, IDs
LCP / CLS / TBT ~5s / ~0.2 / heavy
Canonical / hreflang~ Yoast canonical only
Internal-link graph~ Menu only
Mobile-first responsive~ Divi mobile, heavy
Alt text on all images Inconsistent
Third-party trackers 12+ blocking scripts
LLM-citation friendliness 0/3 prompts cited
After · static rebuild
9/ 10High confidence
Crawlable HTML (no JS dep) 100% static
Semantic structure Real h1/h2/h3 + landmarks
Schema.org coverage 6 types per page
Unique content per page Per-creator bio + FAQs + news + sponsors
URL structure /creatorstats/<slug>/
LCP / CLS / TBT 2.6s / 0 / 0ms
Canonical / hreflang Canonical + hreflang where bilingual
Internal-link graph Niche + country + similar-creator
Mobile-first responsive Lighthouse 92/92
Alt text on all images Descriptive on every img
Third-party trackers Zero blocking scripts
LLM-citation friendliness Static HTML + llms.txt + JSON-LD Person
The big SEO change
Every page now has its own bio (~250 words, Sonnet-written), its own 10 FAQs, its own Recent News, its own sponsor history, and its own audience demographics. Nothing recycled. Google's duplicate-content classifier sees 275 genuinely different pages — not 275 cosmetic clones of a Divi template.
What we can't control
The remaining 1 point reflects what no rebuild can fix: domain-age trust signals, crawl-budget allocation for a fresh 1,350-URL footprint, and Google's discretion on whether to surface us on competitive queries. Indexing rate will tell us within 14-30 days.

Cutover without loss

Same URLs. Same Rankings. Bots Still Get Through.

A rebuild can quietly destroy years of SEO if URLs change or if a paywall blocks search engines. We avoided both.

① Rankings preserved

Every old URL still works — we kept them 1:1

All 14 market and vertical landing pages live at the exact same URLs as before. Google sees no broken links, no redirects, no ranking reset. The pages look modern, but the addresses didn't move.

/taiwan-influencer-marketing/ /japan-influencer-marketing/ /thailand-…/ /singapore-…/ /dubai-…/ /london-…/ /new-york-…/ /b2b-…/ + 6 more
② Bots get through the paywall

Search engines & AI crawlers always see the full page

The 5-view paywall only kicks in for human visitors. Google, Bing, ChatGPT, Claude, and Perplexity get a free pass — they index the full content, so the paywall protects revenue without costing us SEO or AI visibility.

Googlebot Bingbot GPTBot ClaudeBot PerplexityBot

Cost per page · all Claude models

What Each Model Actually Costs.

Real per-page math: ~10.5K input tokens (data sheet, web context, prompts) + ~7.5K output tokens (bio, FAQs, news, sponsors, niche framing). Plus 2 web search calls per page where supported.

Haiku 4.5
$1 / $5 per MTok · in / out
$0.05
per profile
100 pages: $5 1,000 pages: $50 10,000 pages: $500
Fast and cheap. Bio & FAQs are serviceable but blander; web search returns shallower context. Right for high-volume tail creators where bespoke prose is overkill.
Opus 4.7
$15 / $75 per MTok · in / out
$0.75
per profile
100 pages: $75 1,000 pages: $750 10,000 pages: $7,500
Deeper reasoning, better nuance on edge cases. Overkill for templated profile content — same SEO output, 5× the bill. Reserve for hand-curated flagship creators.

Numbers assume no prompt caching. Caching cuts input cost ~10× on repeat calls — meaningful at scale. Web search billed at $10 / 1K queries (Sonnet & Opus only).

Realistic rollout · slow and observed

Stay Small. Watch the Signal.

A couple hundred indexed pages would be a huge win on its own. The plan is to not add anything new for a while, measure what we have, and only scale once we have evidence Google is actually indexing and ranking the work.

Now · day 0
Ship the existing 275. Add nothing.
Total pages275
Sonnet spend$30 sunk
Build timeDone

Submit sitemap to Google Search Console + Bing Webmaster. Then do nothing for 30-45 days — first-batch indexing takes time on a fresh domain footprint.

Phase 1 · 60-90 days
Modest top-up. Only if signal is green.
New pages+100-150
Cost (mostly Sonnet)~$15-20
Build time~30 min

Add only high-intent creators we have data for. Submit a few dozen URLs per week. Trigger only if >40% of the original 275 are indexed and at least a handful are surfacing on long-tail queries.

Phase 2 · 6+ months
Consider scaling to ~750-1,000 — only if Phase 1 ranks.
New pages+500-600
Cost (mixed-model)~$40-60
Build time2-3 hrs

Only if there's evidence the existing 400-ish pages are bringing organic traffic. Anything bigger than 1,000-1,500 total is a separate conversation — we're not chasing volume.

Why so conservative. Programmatic SEO at our domain age is unproven — Google may decide it doesn't want most of these in the index regardless of quality. Better to learn that lesson with 275 pages than 10,000. A couple hundred well-indexed pages outranks a thousand un-indexed ones, every time.

Before we go live

Two Things to Sign Off Before Launch.

① Copy review

The wording needs the right eyes on it

  • Data API copy — who signs off?
  • Agency / consumer copy — who signs off?

Open question: who owns the final read for each side?

② Security pass

Close the open gaps before launch

  • Rate-limit /api/generate so Anthropic credits can't be burned.
  • Gate /api/debug + /api/rebuild-index behind an admin token.
  • Cost alerts on Anthropic + Cloudflare for same-day abuse detection.

Open question: do we run a 1-2 day security pass before the cutover?

CreatorDB · Rebuild Presentation 1 / 16
navigate · F fullscreen