Five site-speed fixes that actually move rankings
Most site-speed advice online is generic enough to be useless. Here are the fixes we actually see move the needle, in the order we'd tackle them.
Site speed affects two things at once: how Google's Core Web Vitals score your pages, and whether a visitor actually sticks around long enough to convert. Both matter, but they don't always need the same fix. Here's where we start.
1. Compress and correctly size images
This is still the single biggest win on most sites. An uncompressed 4MB hero image loaded at display size of 800px is pure waste — convert to WebP or AVIF, and serve an image sized for where it's actually displayed, not the original camera resolution.
2. Defer anything that isn't needed immediately
Chat widgets, analytics scripts, and third-party embeds often load before the actual page content does. Deferring or lazy-loading these lets the page become usable sooner, even if the total download size doesn't change much.
3. Reduce server response time
A slow first response from your server delays everything downstream, regardless of how optimized the rest of the page is. Shared hosting plans and unoptimized database queries are the usual culprits — sometimes a hosting upgrade is the highest-leverage fix available.
4. Fix layout shift, not just load time
Pages that visually jump around as they load — images popping in without reserved space, fonts swapping late — hurt the Cumulative Layout Shift score even if the page technically "loads fast." Reserving space for images and ads before they load solves most of this.
5. Audit third-party scripts regularly
Plugins, tracking pixels, and embedded widgets accumulate over time and rarely get removed once installed. A periodic audit of what's actually still in use — and cutting what isn't — is one of the easiest ways to claw back speed without touching core code.
None of these require a full rebuild. Most sites we look at have two or three of these issues stacked on top of each other, which is why fixing them together tends to produce a bigger jump than fixing any single one in isolation.