Real Core Web Vitals Scores: We Tested 50 Indian Websites
We ran Lighthouse on 50 high-traffic Indian sites across eCommerce, news, fintech, and SaaS, both on a fast desktop and on a throttled "fast 3G + mid-tier Android" profile to reflect real users. Here's what we found.
Key takeaways
- Median LCP on mobile: 4.2 seconds (Google's "good" threshold is 2.5s)
- Median CLS: 0.18 (good is <0.1)
- Median INP: 380ms (good is <200ms)
- Best performers: SaaS marketing sites on Next.js. Worst: news and content-heavy eCommerce.
- The single biggest fixable win: image optimization. Most sites still ship 600KB+ images uncompressed.
Methodology
Tested home page + one inner page per site. WebPageTest agent in Mumbai region, Moto G Power profile, Fast 3G throttling. 3 runs averaged. May 2026.
What's broken on most sites
Images. Most sites ship 4K JPEGs and rely on browser scaling. Lazy load helps but doesn't fix the bytes-on-wire problem.
Custom fonts. Multiple weight variants, blocking, no font-display: swap. Easy win.
Third-party scripts. Chat widgets, marketing pixels, A/B test scripts. Each costs 50-200ms of INP.
Layout shifts. Hero images without dimensions, dynamic ads, embedded videos without aspect-ratio reservations.
The top 5 sites and what they did right
We won't name them, but the patterns:
- Next.js or Astro static generation
- next/image with AVIF/WebP transcoding
- Self-hosted fonts with
font-display: swap - Deferred third-party scripts
- Hero image preloaded explicitly
What we recommend
Run WebPageTest on a Moto G + Fast 3G, your real Indian users live on something like that, not a MacBook on fiber. Fix images first, fonts second, scripts third. Aim for 95+ Lighthouse Mobile.
FAQs
Does Google use lab scores or field? Field (CrUX). Lighthouse is a proxy.
Will fixing CWV improve rankings? Marginally, it's one of many signals.
What about Service Workers? Underused. Worth investigating for repeat-visit speed.