Sunsetting dynamic rendering without the traffic cliff

Sunsetting dynamic rendering without the traffic cliff

A marketplace ran Rendertron for 3 years — serving Googlebot pre-rendered HTML, users a CSR app. Google deprecated the pattern. They needed off it without dying. Here's the staged exit.

— Audit what the prerender served vs what users got. We found 14% of meta tags differed — a cloaking risk nobody noticed.
— Stand up real SSR on one route group (category pages) first. Keep prerender on the rest.
— Compare logs: Googlebot fetch frequency and render success on SSR routes vs prerendered.
— Only after SSR routes held rankings for 21 days, expand to the next group.

Turns out the prerender cache was 6 days stale on price changes — SSR fixed a freshness bug they'd blamed on Google.

We rolled SSR across all groups over 5 weeks, then killed Rendertron.

Result: zero ranking loss through the transition, crawl-to-index latency dropped from 6 days to under 24 hours, and organic ticked +7% from the freshness gain alone.