A Lighthouse score of 100 is not a ranking factor

A green 100 in Lighthouse feels like an SEO win. It isn't one. The Core Web Vitals that Search Console reports come from real visitors in Chrome, and Lighthouse never meets a real visitor. You can have a perfect lab score and failing field data, or the other way round.

Two kinds of data

web.dev splits performance data into lab data and field data. Lab data comes from loading a page in a controlled environment: one device, one network setting, one location. Lighthouse is a lab tool. Field data comes from the people who actually visit, on whatever phone and connection they have. Chrome's tools generally get their field data from the Chrome User Experience Report (CrUX).

Search Console's Core Web Vitals report is based on real world usage data, drawn from CrUX, and it's the report Google's own Core Web Vitals page tells you to check. If a group of URLs doesn't have enough data, Search Console leaves it out or falls back to a wider group of pages.

What the Lighthouse score actually is

The Lighthouse performance score is a weighted average of five lab metrics. In Lighthouse 10's weighting, Total Blocking Time carries 30 percent, Largest Contentful Paint and Cumulative Layout Shift 25 percent each, and First Contentful Paint and Speed Index 10 percent each. Only two of those five are Core Web Vitals. The same page says much of the run-to-run variation comes from things outside Lighthouse, like ads, A/B tests and browser extensions.

The third Core Web Vital, Interaction to Next Paint (INP), replaced First Input Delay on March 12, 2024. INP measures responsiveness. Lab tools like Lighthouse load pages in a simulated environment without a user, so they can't measure it. Total Blocking Time is the lab stand-in, and web.dev calls it a proxy for INP. A perfect Lighthouse score tells you nothing direct about the metric that replaced FID.

What Google says about chasing scores

Google's page experience documentation says Core Web Vitals are used by its ranking systems, but good scores in Search Console or third-party tools don't guarantee top rankings, and trying to get a perfect score just for SEO reasons "may not be the best use of your time." It also says Google tries to show the most relevant content even when the page experience is poor. A good experience helps when many pages are similarly helpful. It doesn't beat relevance.

The target isn't 100 anyway. Google's Core Web Vitals page asks for LCP within 2.5 seconds, INP under 200 milliseconds and CLS under 0.1. Field tools judge those at the 75th percentile of visits, so the goal is that most real visits pass, not that one test run looks perfect.

Why the two disagree

PageSpeed Insights shows both kinds of data on one screen. Its documentation explains that the lab section is a simulated load on a single device and a fixed set of network conditions, while the field section is CrUX data from the previous 28 days, so the values may differ. Neither number is wrong. Lab data is good for reproducing and debugging a problem. web.dev's advice is that field data is what you should use to decide what to work on.

The fix

  1. Start with field data. Open the Core Web Vitals report in Search Console, or the field section at the top of PageSpeed Insights. If a metric already passes there, leave it alone. No field data yet? Use lab numbers to find problems, not as a score.
  2. Find the real bottleneck. For a slow LCP, web.dev's LCP guide says to concentrate on real user data first, then use PageSpeed Insights or DevTools to identify the LCP element and break its time into parts. If the element is an image, the guide's basics apply: make it discoverable in the HTML source, never lazy-load it, and give it fetchpriority="high". Shrinking the file may help less than you expect; the guide notes that the load duration part tends not to be the big bottleneck for most sites.
  3. Check the server side. If time to first byte is the slow part, the same guide names multiple server redirects and distance from the server as causes, and suggests a CDN for the distance. Redirect chains are an easy one to fix: see Your homepage has four addresses.
  4. Use Lighthouse as a debugger, not a scoreboard. Run it to reproduce a problem and confirm a fix, then let the field data catch up over its rolling 28 days.
  5. Then stop. Once the field metrics pass, the next hour is better spent on the content that makes the page relevant.