You run Lighthouse, you get a 95, you move on. A month later your rankings slip and a “Core Web Vitals: needs improvement” warning appears in Search Console. How can both be true?
Because they measure two different things — and only one of them affects your rankings.
Lab data vs field data
Lighthouse, PageSpeed Insights' lab tab, and most speed tools run a lab test: one simulated device, on one controlled connection, in one location, at one moment. It's a clean, repeatable diagnostic — useful for debugging, but it's a laboratory, not the street.
Google ranks on field data: the Chrome User Experience Report (CrUX), which aggregates real measurements from real Chrome users on your site over a rolling 28-day window. Cheap Android phones on patchy mobile networks. Older laptops. Real-world cache states. That distribution is what determines whether you pass Core Web Vitals — not your one good run on a fast machine.
Why a 95 can still fail
- Your test device isn't your audience's device. A developer on fibre and a flagship phone sees a site most of your visitors never experience.
- INP only shows up with real interaction. Interaction to Next Paint — the responsiveness metric that replaced FID — measures how your page reacts to real taps and clicks. A lab test that doesn't interact can't catch what real users feel.
- Real layouts shift. Late-loading ads, embeds and fonts cause the cumulative layout shift that real users hit and your clean lab run doesn't.
The three metrics that decide it
Core Web Vitals in 2026 come down to three field measurements: LCP (largest contentful paint — loading, good under ~2.5s), INP (interaction to next paint — responsiveness, good under ~200ms), and CLS (cumulative layout shift — visual stability, good under ~0.1). Google grades you at the 75th percentile of your real users — so it isn't enough for the median visitor to have a good time; three out of four must.
One snapshot lies; the trend tells the truth
Because field data is a 28-day rolling aggregate, a single check tells you almost nothing. A deploy that quietly regresses INP, a new hero image that wrecks LCP, a third-party tag that adds layout shift — these show up as a drift over weeks, not a red number on one test. If you're only spot-checking with a lab tool, you find out when rankings move, which is far too late.
Track what Google actually sees
CrUX Tracker monitors your site's real-user field data continuously — LCP, INP and CLS at the 75th percentile — so you catch regressions as they happen instead of discovering them in a ranking drop.
If your lab scores look great but you're not sure your real users agree, start tracking your field data on the CrUX Tracker page.