<b>Google was indexing a version of the site that no longer existed</b>

<b>Google was indexing a version of the site that no longer existed</b>

A fintech blog republished 80 articles with updated 2025 data. Three weeks later, Google still showed 2023 figures in the snippets. The cache wouldn't budge.

We assumed slow recrawl. But Search Console showed Googlebot fetching the URLs daily — and getting <code>304 Not Modified</code> every time.

Their SSR layer sat behind a CDN that generated <code>ETag</code> headers from the raw HTML template hash, not the rendered output. The template never changed; only the injected data did. So the CDN told Googlebot 'nothing new here' on every visit, and Google kept the old rendered snapshot.

Real users got fresh content because their requests had query strings that dodged the cache key. Googlebot hit the clean canonical URL and got the stale 304 forever.

The trap: conditional request headers derived from the wrong layer of a rendering stack are invisible until you read the response headers, not the body.

Fix: ETag computed from final rendered HTML including data payload.

Result: snippets refreshed within 9 days, and the updated articles regained featured-snippet slots they'd lost — clicks +34% on the affected set.