PageSpeed Insights
PageSpeed Insights is Google's free tool that analyses a URL and shows its Core Web Vitals values – from real user data where available, and from a simulated lab test. It also provides concrete suggestions for improvement.
In practice
The lab score can differ from the real field values, because the field data reflects the actual devices and connections of real visitors. What counts for Google rankings in the end is the field data, not the lab result. That field data comes from the Chrome UX Report (CrUX): a rolling 28-day window, assessed at the 75th percentile of real page views – so an optimisation only shows up there in full after roughly a month. INP requires real user interactions and therefore appears only in the field data; in the lab section (Lighthouse), Total Blocking Time (TBT) serves as the proxy.
The lab score is a weighted sum of five measurements. TBT counts for 30 per cent, LCP and CLS for 25 per cent each, First Contentful Paint and Speed Index for 10 per cent each. The test runs on a simulated mid-range smartphone with a throttled connection. The desktop test runs separately and almost always scores better. Small fluctuations between two runs are normal and no cause for concern.
Just as important is what the score does not say. A lab score of 100 does not mean the page is fast for real visitors. Only the field data shows that. On low-traffic websites the field section is often missing entirely, because CrUX has too few measurements; then only the lab remains. Rankings are not based on any Lighthouse score, but on the Core Web Vitals thresholds in the field. At the 75th percentile these are LCP up to 2.5 seconds, INP up to 200 milliseconds and CLS up to 0.1. In practice it is worth checking your most important pages regularly, rather than relying on a single measurement.
What usually lies behind a poor score, and in which order images, fonts, scripts and hosting are best tackled, is described in our magazine article on fast websites.
The lab score is a weighted sum of five measurements. TBT counts for 30 per cent, LCP and CLS for 25 per cent each, First Contentful Paint and Speed Index for 10 per cent each. The test runs on a simulated mid-range smartphone with a throttled connection. The desktop test runs separately and almost always scores better. Small fluctuations between two runs are normal and no cause for concern.
Just as important is what the score does not say. A lab score of 100 does not mean the page is fast for real visitors. Only the field data shows that. On low-traffic websites the field section is often missing entirely, because CrUX has too few measurements; then only the lab remains. Rankings are not based on any Lighthouse score, but on the Core Web Vitals thresholds in the field. At the 75th percentile these are LCP up to 2.5 seconds, INP up to 200 milliseconds and CLS up to 0.1. In practice it is worth checking your most important pages regularly, rather than relying on a single measurement.
What usually lies behind a poor score, and in which order images, fonts, scripts and hosting are best tackled, is described in our magazine article on fast websites.