Why does my site score differently in Lighthouse than in PageSpeed Insights?

Because the two tests run under different conditions: Lighthouse in your own browser measures the page on your hardware, your network, and your browser state, while PageSpeed Insights runs Lighthouse on Google's servers against a simulated mid-range phone with a throttled connection — and PSI additionally shows field data from the Chrome UX Report, which is not a test at all but the recorded experience of your real visitors over the previous 28 days (Google, "About PageSpeed Insights," verified live August 18, 2026).

The qualifier: none of these numbers is wrong; they answer different questions. Google's own web.dev guidance ("Why lab and field data can be different") is explicit that lab tools control conditions to make results reproducible, while field data reflects the messy spread of real devices, networks, and behavior — and that you should use field data to measure and lab tools to diagnose. The number Google's ranking systems assess is the field data, at the 75th percentile.

One URL, three legitimate numbers

One trap from our own build logs is worth naming, because nothing in the report warns you: Lighthouse will happily score the wrong page. Password-protected preview URLs — Vercel's deployment protection is the case we hit — answer with a 302 redirect to a login screen, and Lighthouse follows the redirect and scores the login page. The login page is tiny, so the score comes back high, on a page none of your visitors will ever see. We now measure production URLs or a local build of the site, never a protected preview (SAA build process, 2026). If a staging score looks suspiciously clean, check what HTML the test actually loaded before believing it — a real 99+ has to be earned on the real pages.