The original review found a mismatch: crawler download statistics looked reasonable, but an individual page performed poorly in a browser test. Large images were part of the investigation.
The durable lesson is to understand the measurement before deciding the site is fast. An average crawler statistic does not describe the complete experience of a customer opening a product page.
Test representative pages
Check the homepage and important content or product templates under relevant device and network conditions. Record the conditions and repeat unexpected results. Use a browser loading waterfall to see which requests and rendering work delay the page.
Don’t treat a single score as a ranking threshold. Nor should the old recording’s claim that Google downloads only text be used to interpret current crawl reporting.
Make images appropriate for their use
Inspect the dimensions and file size actually delivered to the browser. Resize and compress where appropriate, then compare the result visually. Product detail, colour and legibility still matter.
Keep originals so the change is reversible. A smaller file is useful only if it still does the job the customer needs.
Apply caching carefully
Static assets and personalised pages need different treatment. Cache reusable content appropriately, but test logged-in views, carts and changing information before enabling broad page caching.
Check compression, scripts and server response as separate possible causes. A dedicated IP address does not by itself provide more server resources or establish a speed improvement. Buy a hosting change only after diagnosing what it will solve.
Verify behaviour as well as the metric
After changes, test the customer journey and compare performance under the same conditions. If crawling controls appear inconsistent, inspect the actual robots.txt response and directives rather than assuming Google rewrote them.
See our SEO audit checklist and conversion work to connect the technical review to the site people use.
This article has been updated to focus on representative page testing, image delivery and evidence-led hosting decisions.