Most WordPress sites now fail Core Web Vitals. Here is what changed

What actually changed in 2026?

Google moved from grading pages individually to aggregating Core Web Vitals across the whole domain, weighting LCP, INP and CLS equally.

The consequence is worth sitting with. Previously a fast landing page could rank well while the rest of the site was slow. Now the neglected sections — the old blog archive, the events pages nobody has touched since 2023 — pull down pages you actively care about.

For WordPress sites specifically this hurts, because WordPress sites accumulate. Plugins get added for one campaign and never removed. Templates get copied. The slow parts are rarely the parts anyone is looking at.

Why is INP the metric that fails?

Because it measures JavaScript, and WordPress sites accumulate JavaScript faster than anything else.

Interaction to Next Paint replaced First Input Delay in March 2024. FID measured only the delay before the first interaction was processed, which almost everything passed. INP measures the full time until the screen updates, across every interaction in the visit, and reports close to the worst one.

A typical WordPress install runs a page builder, a slider, a form plugin, a cookie banner and an analytics tag, each attaching its own handlers. None of them is individually the problem. Together they are.

What is speculative loading, and should you use it?

WordPress 6.8 added speculative loading, which fetches pages before the visitor clicks. It ships in prefetch mode by default, and switching it to prerender can produce meaningful improvements without writing any code.

The trade-off is bandwidth. Prerendering pages nobody visits wastes it, and on a large site with many outbound links that adds up. It suits sites with a clear path through them — a service page leading to a contact page — more than sprawling content archives.

It is worth trying precisely because it costs a setting change and nothing else. Measure before and after rather than assuming.

Be careful about what you measure. Lab scores from PageSpeed Insights swing several points between runs on identical code. Field data in Search Console is what Google actually ranks on, and it lags by 28 days.

Does the AI Client in WordPress 7.0 help with this?

Indirectly, and less than the marketing suggests.

WordPress 7.0 shipped a provider-agnostic AI Client in core in May 2026, with connectors for OpenAI, Anthropic and Google. It is genuinely useful infrastructure — it means plugins can offer AI features without each one bundling its own API layer.

But no AI tool removes a plugin you decided to install. The performance gap between a default WordPress install and a well-optimised one has narrowed considerably, and automated tooling closes more of it than it used to. The decisions that actually matter — which plugins, which theme, which page builder — are still made by a person.

What should you actually do?

In order of return, and none of these need a developer for the first two:

• Find out if you are failing

  Search Console shows field data from real Chrome visitors, grouped by URL pattern. That is the version that counts, and it tells you which section of the site is the problem.

• Audit your plugins

  List every active plugin and ask what would break if it were gone. On an established site, a third of them usually fail that question.

• Turn on speculative loading

  Switch from prefetch to prerender and re-measure. A setting change with real potential upside.

• Deal with the page builder

  If you run one, it is almost certainly your largest single source of JavaScript. This is the expensive fix and usually the one that matters most.