The CEO gets an email: "your site is red on Google". The agency proposes migrating to a "Core Web Vitals Premium" host. Two months later LCP stays at 3.1 s: the homepage image still weighs 2.4 MB, served without lazy load or WebP. The server was innocent; the quote was not.
Core Web Vitals (LCP, INP, CLS) affect SEO and conversion. They aggregate network, server and front-end signals. Confusing "slow hosting" and "poorly optimised site" leads to useless migrations — or vague commercial promises.
The three metrics — who does what?
| Metric | Measures | Host lever | Non-host lever |
|---|---|---|---|
| LCP (Largest Contentful Paint) | Time to show largest visible element | TTFB, edge HTML cache, fast disk | Image compression, preload, CDN assets |
| INP (Interaction to Next Paint) | Interaction responsiveness | Server stability, no 502 under load | Lighter JS, split bundles, workers |
| CLS (Cumulative Layout Shift) | Visual stability | Rare (except badly served fonts) | Image dimensions, ads, fonts |
What the host can actually improve
TTFB and stability — overloaded shared plan, misconfigured PHP-FPM (PHP-FPM), OPcache off (OPcache): fixable by tuning or higher plan. Measure with TTFB diagnosis.
Cache and included CDN — some WordPress hosts include page cache or CDN partnership. Check if cache covers dynamic HTML or assets only.
HTTP/2 / HTTP/3 — multiplexing and QUIC help asset-heavy sites with healthy origin (HTTP/3).
Geography — datacenter region close to audience cuts RTT; useful without CDN.
What the host will not fix alone
- Unresized hero images.
- Twelve WordPress plugins loading JS in head.
- Google fonts without
font-display: swap. - Late-injected ad banners (CLS).
- Heavy React SPA without code splitting (INP).
Migrating to the "best PageSpeed host" without a front audit means moving a slow site to a faster SSD.
The peak: PageSpeed score is not a hosting SLA
Google evaluates real user experience (CrUX); local Lighthouse on fibre is not enough to promise a green badge post-migration.
Decide and move forward without blind spots
- Separate origin TTFB, LCP element, INP long tasks in CrUX / Search Console.
- Fix images and JS if TTFB < 400 ms but LCP/INP red.
- Optimise PHP/cache if TTFB red — before any plan change.
- Compare hosts on TTFB proof, not CWV slogans — directory, comparator.
Frequently asked questions
Does a "CWV optimised" host guarantee green?
No — LCP and INP depend mostly on front end; check commercial scope.
High LCP — host or front?
TTFB > 600 ms → server contributes. OK TTFB, LCP > 2.5 s → images and client render.
INP and the host?
INP = mostly JS; host via TTFB/stability only.
Does CDN count?
Yes for LCP assets and sometimes TTFB; not for CLS or JS-dominated INP.
Before migrating for Core Web Vitals, ask: which metric is red and which component drives it? If the answer is "we do not know", the host is not the right lever yet.