PageSpeed Insights

PageSpeed Insights and Lighthouse are often treated as the same tool because they share the same scoring engine, but they measure two different things.

Lighthouse, run locally via Chrome DevTools, the CLI, or CI, produces lab data.

PageSpeed Insights runs that same Lighthouse engine on Google's own servers, then layers real-user field data from the Chrome UX Report (CrUX) on top, a rolling 28-day window of how actual visitors on real devices and networks experienced the page. This is reported at the 75th percentile.

Lighthouse tests can also be done on Geekflare tools or API.

That's what a report actually looks like: the field-data section up top (real Chrome users, 28-day window), with the Lighthouse lab score and diagnostics further down the same page.

That field data is the part that actually is relevant for rankings. Google has confirmed that Core Web Vitals, the ranking signal, come from CrUX field data. A page can score 95 in a local Lighthouse run and still show Poor Largest Contentful Paint in Search Console's report because real visitors on slower connections and older devices are having a worse experience than the test simulated.

This is also why PSI scores fluctuate between runs in a way that can be confusing: shared server infrastructure, varying network routing, and the CrUX rolling window all introduce variance that a fixed local Lighthouse run doesn't have.

The two tools are complementary rather than competing. PSI's field data tells you whether real users are actually having a bad experience and whether that's likely affecting rankings. Lighthouse gives you a controlled way to isolate a specific fix, verify it worked, and catch regressions before they ship, without waiting weeks for the CrUX window to update.