WordPress speed in 2026: the changes that genuinely paid off

1. Server-level page caching. Nothing comes close to it. A cached page never touches PHP or MySQL, and it leaves the server inside a millisecond or two. Where your host runs LiteSpeed, install the free LiteSpeed Cache plugin and turn page caching on. That single toggle regularly beats every other item in this article added together. Our walkthrough runs about ten minutes.

2. Image discipline. Nothing wrecks LCP more often than an oversized picture, in our experience. Serve WebP, cut each file to the size of the box it sits in, and lazy-load anything below the fold, a job WordPress now does on its own. Our image optimisation guide walks you through the workflow.

3. Plugin triage. How many plugins you run matters far less than what each one weighs. One page builder dragging in 400 assets costs you more than twenty tidy little utilities ever will. Query Monitor is free, and it will name the culprits inside an afternoon. Usual suspects: page builders, related-posts engines, anything phoning home on every request.

4. PHP version. Each major release since 8.0 has run measurably faster than the one before, so a site parked on 8.1 is leaving genuine speed on the table. Upgrading amounts to a dropdown in the control panel, plus a five-minute compatibility check.

Now and again a ticket points at our own trade rather than at anybody's plugin list. The tells: TTFB above roughly 600ms on a page already cached, respectable scores that collapse the moment a handful of visitors arrive at once, and afternoon sluggishness nobody can account for, which usually means a noisy neighbour on an oversold machine. No plugin repairs any of that.

That answer is structural, and it lies in the stack underneath. NVMe storage hands back database queries in microseconds where SATA needed milliseconds, while LiteSpeed's architecture keeps its shape under concurrency in a way stock Apache never manages. There you have the entire case for our WordPress hosting: we look after the platform layer, which leaves you the four fixes listed above.

Hosted somewhere else and suspicious of the platform? Run a TTFB test against a cached page during a quiet hour, then repeat it at a busy one. A wide gap between the two says too many sites are crammed onto too small a server, and no amount of image compression outruns that. Migration here costs nothing, so the experiment is cheap.