You've just shipped a site. It loads instantly on your machine, looks great, passes your internal checks. Then you run it through PageSpeed Insights and the mobile score comes back at 41.
The instinct is to dismiss it. "Google throttles the test." "It's not a realistic environment." "My actual users aren't seeing this."
Some of them are.
What the Test Is Actually Simulating
PageSpeed Insights uses Lighthouse under the hood, and Lighthouse doesn't test on your infrastructure or your device. It simulates a specific user profile:
- Device: Moto G Power — a mid-range Android phone, not a flagship. This represents the median globally shipped smartphone, not what developers or early adopters carry.
- Network: 1.6 Mbps download, 150ms round-trip time. That's slow 4G — the kind of connection you get in a moving vehicle, in a rural area, or in a dense urban environment with tower congestion.
- CPU: 4x slowdown multiplier applied on top of the simulated device. This accounts for thermal throttling — when a phone has been running for a while, it slows its CPU to manage heat. It also accounts for background processes competing for cycles on a device with 3GB of RAM instead of 16GB.
That 4x CPU multiplier is the one most developers don't know about. It's why JavaScript-heavy sites that feel fast on a MacBook fall apart in the test. The same 200KB of JS that your machine parses in 40ms takes 160ms on a throttled mid-range phone. Multiply that across all your scripts and you've already spent your entire performance budget before anything renders.
This isn't Google being punitive. It's Google choosing to represent the bottom half of the global mobile market instead of optimizing for the top.
Lab Score vs. Field Score, They're Different Things
PageSpeed Insights shows two kinds of data, and conflating them is one of the most common sources of confusion.
The lab score (the big number at the top) is the Lighthouse simulation. It's synthetic — the same test run in the same controlled conditions every time. Useful for diagnosing issues, but it doesn't reflect your actual users.
The field data comes from the Chrome User Experience Report (CrUX) — real sessions from real Chrome users who visited your site, aggregated over a 28-day rolling window. This is what the Core Web Vitals assessment at the top of PSI is based on.
A critical detail: CrUX data is reported at the 75th percentile. Your score isn't the average experience — it's the threshold at which 75% of your users had that experience or better. In other words, you're being graded on your worst quartile, not your typical visitor.
This is why a site can have a lab score of 45 and green field data: if most of your real users are on fast devices in fast-connection geographies, CrUX will reflect that. The simulation still reveals genuine technical problems — they're just not yet affecting your user base.
The inverse is worse: green lab score, red field data. That means something is degrading in production that the simulation doesn't catch — third-party scripts, A/B testing tools, ad networks loading after the simulated test window closes.
Both scores matter. Lab tells you what to fix. Field tells you who's already affected.
What the Metrics Are Actually Measuring
Core Web Vitals aren't arbitrary benchmarks. Each one maps to a specific moment in a user's experience.
LCP Largest Contentful Paint
When does the main content of the page appear? Not when the spinner stops, not when the page is technically "done" — when the largest visible element (usually a hero image or above-the-fold text block) finishes rendering.
LCP fails silently in ways that feel fast on a good connection: a hero image that isn't preloaded, a slow TTFB from an uncached server response, render-blocking CSS that delays paint. The user is staring at a partially loaded page while your analytics still logged a "fast" load.
CLS Cumulative Layout Shift
How much does the page jump around while loading? CLS measures the total amount of unexpected layout movement — elements shifting position as images load in without declared dimensions, as web fonts swap in, as late-loading ad scripts inject content above existing text.
This is the one that causes real user errors: a person goes to tap a button, the page shifts at the last millisecond, and they hit something else. That's not a score problem. That's a UX failure.
INP Interaction to Next Paint
Replaced FID (First Input Delay) in March 2024, and it measures something more complete. FID only captured the delay between a user's first interaction and when the browser started processing it. INP captures the full event lifecycle: input delay + processing time + the time until the browser actually paints the result.
A page can load quickly and still fail INP badly. If your main thread is blocked by long JavaScript tasks, every tap and click will feel sluggish — the browser can't respond until the current task finishes. This is what "JavaScript bloat" actually means in measurable terms.
Why This Matters Beyond the Score
Core Web Vitals are a confirmed ranking signal in Google Search. The weight is real but moderate — you won't outrank a highly authoritative competitor just by going green. But at equal authority, technical performance is a differentiator.
The more direct impact is on behavior. Google and Deloitte found that improving load time by 0.1 seconds increased conversions by up to 8% on retail sites. Bounce rate on mobile climbs sharply past the 3-second mark. These aren't PageSpeed numbers — they're revenue and retention numbers.
When a developer dismisses a low mobile score, they're often making a decision on behalf of users they never actually see: the person in a different country on a prepaid data plan, the user on a 3-year-old phone, the potential customer who left before your page finished loading and never came back.
What to Do With This
Don't dismiss a low mobile score as "the test is unrealistic." Instead:
- Check your field data first. If CrUX is red, real users are already affected. If it's green or absent (not enough data), you have time but not infinite time.
- Fix LCP first. It has the biggest perceived impact. Preload your hero image, reduce TTFB, eliminate render-blocking resources above the fold.
- Audit your JavaScript for INP. Long tasks on the main thread are the primary cause. Look for third-party scripts, large bundles, and synchronous operations running at page load.
- Declare image dimensions and reserve space for dynamic content to address CLS. It's the most mechanical fix of the three.
The score is a proxy. The real thing it's measuring is whether someone on a mid-range phone in a spotty 4G area can use your site without frustration. That person exists. They're in your analytics right now.
If you're working on a WordPress site and want LCP, CLS, and INP handled automatically — image optimization, script management, caching — BerqWP does this without requiring manual configuration for each metric.