The short verdict
- Lighthouse is a lab test: one simulated visit. Field data is what real visitors experienced.
- Google’s ranking signal uses field data. Lighthouse is for finding causes.
- Chase passing Core Web Vitals for real users, not a perfect 100.
01Two kinds of numbers
Run PageSpeed Insights on any page and you see two sections. At the top, if your site has enough traffic, is “Discover what your real users are experiencing”. Below that is a Lighthouse score out of 100 and a list of diagnostics. They often disagree, and people tend to worry about the wrong one.
02Lab data: Lighthouse
Lighthouse loads your page once, in a controlled setup that simulates a mid-range phone on a slowed connection, and measures what happens. The same test runs in Chrome DevTools.
- Good for: finding causes, comparing before and after a change, testing pages that are not live yet.
- Not good for: knowing what visitors actually see. It is one visit, on one simulated device, and scores move by several points between runs.
03Field data: Core Web Vitals from real visitors
Chrome collects anonymous timings from real visits by users who allow it. This is published as the Chrome User Experience Report, and shown in PageSpeed Insights and the Core Web Vitals report in Search Console. It covers the last 28 days and reports the 75th percentile: the experience of the slower quarter of visits.
- Good for: knowing whether real people have a fast experience, and what Google sees for ranking.
- Not good for: quick feedback, since it lags by weeks, or small sites, which may not have enough traffic to show any data.
04Side by side
| Factor | Lab (Lighthouse) | Field (CrUX) |
|---|---|---|
| Source | One simulated load | Real Chrome visits |
| Speed of feedback | Instant | 28-day rolling window |
| Used by Google ranking | No | Yes |
| Measures responsiveness | Indirectly, via Total Blocking Time | Directly, via INP |
| Best use | Diagnosing and testing fixes | Judging the result |
05Why they disagree
Your real visitors might be on faster phones and better networks than the simulation, so field data looks better. Or many might be on older phones far away from your server, so it looks worse. Real visitors also have cached files on return visits, which the lab never does. And lab tests cannot measure Interaction to Next Paint at all, since nobody taps anything.
“Lighthouse tells you why. Real-user data tells you whether it matters.”
06How we use both
Start with field data. If Core Web Vitals pass for real users, speed is not your biggest problem, whatever the score says. If they fail, look at which one fails, then use Lighthouse to find out why and to check that each fix helps before you publish it.
For a site with too little traffic for field data, use Lighthouse with a pinch of salt: run it three times and take the middle result, on mobile settings.
07Questions people ask
Do I need a Lighthouse score of 100?
No. It is satisfying, but past the 90s the gains for visitors are small. Passing field data matters more.
Why does my score change every time I test?
Network and server timing vary slightly on each run, and third-party scripts behave differently. A swing of a few points is normal.
My site has no field data. What should I do?
Use lab tests for now, and add a real-user monitoring script if you want your own field data before Google has enough.
Can you read my numbers for me?
Yes. Send us the link and we will tell you what the numbers mean and what to fix first.
