How We Engineer Sub-Second Lighthouse Scores
The Performance Bar
A 0.3-second Lighthouse score isn’t aspirational — it’s a requirement. Google uses Core Web Vitals as a ranking signal, and AI crawlers like GPTBot reward fast sites with higher source scores. In 2026, site speed is simultaneously a ranking factor for Google, a citation factor for AI models, and a conversion factor for your users. Every millisecond matters.
Here’s how we consistently deliver Lighthouse 98+ scores for our clients using our proven engineering blueprint.
The REX Bunny Stack
We’ve architected our entire stack around performance, not convenience. Every technology choice is justified by measurable impact on load times:
- Astro — Static-first framework that ships zero JavaScript by default. Pages are pre-rendered to HTML at build time, meaning the browser receives ready-to-display content with no framework overhead.
- Preact Islands — Interactive components are isolated as “islands” that load independently. A contact form on one page doesn’t load JavaScript for the entire site. Only the components that need interactivity get JavaScript.
- Tailwind CSS — Utility-first CSS framework that purges unused styles at build time. The production CSS bundle is typically under 10KB — smaller than most hero images.
- Edge CDN — Content served from the nearest edge node to your user. Sub-50ms time-to-first-byte regardless of geographic location.
- PocketBase — Single-binary, SQLite-backed backend that serves API responses in under 5ms. No heavy database queries, no N+1 problems.
The Blueprint
1. Audit Everything
Before writing a line of code, we run a comprehensive performance audit:
- Lighthouse — Full diagnostic across performance, accessibility, best practices, and SEO
- WebPageTest — Multi-location load testing with filmstrip view
- Bundle analysis — Inspect every dependency for size impact
- AI Crawlability check — Verify GPTBot, Claude-Web, and PerplexityBot can crawl efficiently
- Core Web Vitals — Measure LCP, CLS, INP, TBT, and FCP at the 75th percentile
Every millisecond is accounted for and documented before we start building.
2. Architect for Speed
Architecture decisions cascade into performance outcomes. We make deliberate choices:
- Font loading — Self-host fonts, subset to only needed characters, use
font-display: swap - Image optimization — Next-gen formats (WebP, AVIF), responsive srcsets, lazy loading with explicit dimensions
- JavaScript strategy — Critically analyze every dependency. If a library adds 50KB, it needs to justify 50KB of value
- CSS strategy — Critical CSS inlined, the rest loaded asynchronously. No render-blocking stylesheets
3. Build with Constraints
Every component must justify its existence. We follow a strict tiered approach:
- Content-only pages — Pure HTML and CSS. Zero JavaScript. A blog post doesn’t need React.
- Partially interactive pages — Preact islands for specific components (navigation, forms, interactive charts). The island pattern ensures JavaScript only loads where it’s actually needed.
- Fully interactive pages — Only when the use case demands it. Dashboards, tools, and complex UIs get full SPA treatment — but these are isolated to specific routes.
4. Optimize Relentlessly
Post-build optimization is continuous, not a one-time step:
- Image compression — Automated pipeline that compresses every image to the smallest acceptable quality
- Font subsetting — Only the characters used in the actual content
- CSS minification — Tailwind’s built-in purging combined with additional minification passes
- HTTP/3 and Brotli — Modern protocols and compression at the CDN level
- Edge caching — Cache-control headers optimized for every content type
Real Results
| Metric | Before | After |
|---|---|---|
| LCP | 3.2s | 0.3s |
| Lighthouse | 62 | 98 |
| CLS | 0.15 | 0.01 |
| TBT | 450ms | 45ms |
| Page weight | 2.1MB | 89KB |
Why Speed Matters for AI Search
Many clients ask us why we prioritize performance for GEO and AEO optimization. The answer is straightforward: AI crawlers have timeouts. If your page takes more than a few seconds to load, GPTBot and other AI crawlers may abandon the crawl entirely. A fast site isn’t just better for users and Google — it’s essential for being cited in AI responses.
Ready to engineer a faster site? Run a free audit to see your current performance baseline.
Related reading:
- What is GEO? — Why speed matters for AI search citations
- Programmatic Local SEO — Scale your SEO with automated city pages
- What is AEO? — How page speed affects answer extraction
- Our Web Development Service — Lighthouse 95+ guaranteed
- SaaS Platform Rebuild Case Study — From 6s to 0.4s LCP