A law firm in Hackensack proudly showed us their PageSpeed Insights score: 96 on mobile. Their developer had spent $3,400 optimizing for it. The site was, on paper, fast. Bounce rate was 71%, average mobile session 38 seconds. Real visitors were leaving.
The 96 was a lie. The PageSpeed score measured a single cold-cache load from a Google data center in Iowa, on a simulated mid-range Android. The actual visitors were on real iPhones in Bergen County, on imperfect cell connections, often with three other tabs open. Those visitors saw a 4.8-second LCP and quit.
Lab data versus field data
Lab tools — PageSpeed Insights, GTmetrix, WebPageTest — simulate a single load under controlled conditions. The score is consistent, reproducible, and tells you almost nothing about what your real visitors experience.
Field data — Real User Monitoring or RUM — measures actual loads from actual visitors on their actual devices. Google’s free version is the Chrome User Experience Report (CrUX). It is what Google uses to rank you. It is not what your developer is looking at.
Why the 95 score is misleading
Real visitors hit your site with a partially-warm cache, variable cellular signal, an older iPhone or a mid-tier Android two generations behind the simulated device, background apps eating RAM, and intermittent DNS hiccups. A 96 lab score might be a 62 in the field. We see that gap regularly.
Where to find your CrUX data
PageSpeed Insights shows you both — most people just do not scroll. At the top is the lab score everyone fixates on. Below, in “Discover what your real users are experiencing,” is the 28-day rolling CrUX data. That is the data that matters for ranking. The lab score is a diagnostic tool, not a verdict.
If CrUX says “does not have sufficient real-world speed data,” your site lacks the traffic for Google to score it on field metrics. The lab data is then your best proxy — but understand it is a proxy.
The 3 metrics that actually predict ranking impact
- Largest Contentful Paint (LCP). The time to render the largest visible element — usually a hero image or headline. Google wants under 2.5 seconds on the 75th percentile of real visits. If your CrUX shows 3.1 seconds at the 75th, you have a ranking problem regardless of your lab score.
- Interaction to Next Paint (INP). Replaced FID in March 2024. Measures how long the page takes to respond to taps, clicks, and key presses. Under 200ms is good, over 500ms is bad. Heavy JavaScript — most page builders, most third-party scripts, most chat widgets — kills this metric.
- Cumulative Layout Shift (CLS). Measures how much the page jumps around as it loads. Ads loading in, web fonts swapping, images without dimensions reserved — each one shifts the layout and burns CLS. Under 0.1 is good. Most WordPress sites with ad units score 0.25 or worse.
How to read both correctly
Lab data for diagnosis, field data for verdict. If your CrUX shows a 3.5-second LCP at the 75th percentile, your real visitors are having a bad time. Open the lab diagnostics — they tell you which assets are slow, which scripts are blocking, which images are oversized. Fix those, then watch the CrUX numbers move over the next 28 days.
Chasing a 100 lab score while your field LCP is 4 seconds is theater. Chasing a 75th-percentile LCP under 2.5 seconds, even if your lab score is 82, is the actual work.
The Hackensack law firm: what we found
Lab score 96. Field LCP 4.7 seconds at the 75th percentile. The gap was three things their developer had not touched: a chat widget loading 340KB of JavaScript synchronously, a Google Fonts call that blocked rendering for 600ms, and an oversized hero image served full-resolution to mobile because the responsive image tags were missing.
Fixed all three over two weeks. Field LCP dropped to 1.9 seconds. Mobile bounce rate fell from 71% to 44%. The lab score actually dropped to 91 — and they did not care, because the field data was finally good.
How AJD handles this
Every performance audit we run starts with the CrUX data, not the lab score. We pull the 75th-percentile LCP, INP, and CLS from PageSpeed Insights and the Search Console Core Web Vitals report. If field data exists, that is the verdict. If not, lab data is a working hypothesis to re-measure in 30 days.
If your developer keeps showing you a 95 PageSpeed score and you cannot understand why your bounce rate is still bad, you are looking at the wrong report. Whether you work with us or not, open PageSpeed Insights right now, scroll past the big number, and find the CrUX section. That is the number that pays your bills.
