Your Core Web Vitals Score Is Green. That's the Problem.

The agency installed a caching plugin. The badge turned green. Three months later you're having the same conversation with a different agency, except now there's a 40-slide deck involved and the badge is still green and the conversion rate is still flat.

Here's the admission that should come before any argument: both of those agencies were optimizing for the wrong thing. And the metric they were chasing was designed, structurally, to let them get away with it.

What a Green Badge Actually Tells You

PageSpeed Insights scores your site at the 75th percentile of real user data pulled from the Chrome UX Report. Pass means 75% of your users had an experience at or above the threshold. The remaining 25% are not in the score. They never were.

According to Google's own CrUX data published in the HTTP Archive's 2024 Web Almanac, roughly 43% of origins globally still fail at least one Core Web Vital in field data, even when lab scores look clean. The divergence between lab and field is not a rounding error. It is the whole story. Lab tests run on a controlled device, a controlled connection, with no other tabs open and no third-party scripts misbehaving. Your actual users do not live in that environment.

The uncomfortable implication: an agency that reports green PageSpeed scores as evidence of progress may be reporting on a test environment that has nothing to do with your users' real experience.

The CrUX Divergence Nobody Puts in the Report

The Chrome UX Report is publicly queryable. It contains field data, real interactions from real Chrome users on real devices and connections, segmented by origin, country, and connection type. It is free. It is the data source Google actually uses for ranking signals.

Most CWV audits delivered by agencies are built on PageSpeed Insights lab scores or Lighthouse runs. Those tools are useful for diagnosis. They are not what determines your search ranking, and they are not what your worst-served users experience.

The gap between the two is where the problem lives. A site can score 90+ in Lighthouse and still have 30% of its real user population recording poor LCP, because those users are on mid-range Android devices on congested mobile connections, and the lab test was run on a simulated Moto G4 with a fixed throttling profile that hasn't been updated to reflect actual 2024 network conditions. The HTTP Archive's analysis of CrUX data for Q4 2024 found that e-commerce origins specifically show the widest lab-to-field divergence of any category, frequently by 800ms or more on LCP alone.

If your agency's reporting never included a CrUX query alongside the Lighthouse score, they were showing you the easier number.

CLS Is the Metric Most Audits Get Wrong Methodologically

Cumulative Layout Shift is particularly vulnerable to lab-versus-field divergence because layout shift is often triggered by content that loads differently under real conditions: ads that take variable time to render, fonts that flash before subsetting kicks in, embeds that resize after the initial paint.

A Lighthouse run in a clean browser with no ad scripts active will record a CLS of 0. The same page with a live ad slot loading 400ms late will record a CLS that fails the threshold for every user who sees that ad. The Chromium team documented this specifically in their 2023 guidance on CLS measurement, noting that lab CLS and field CLS diverge most severely on pages with dynamic content insertion, which describes most commercial sites.

The fix for CLS is not suppression. It is reservation: allocate the space before the content arrives, so the layout does not shift when it does. That requires knowing which elements are actually causing shift in field conditions, not in a clean lab run. You find that in CrUX and in real user monitoring, not in PageSpeed Insights.

Why the 25% Stays Invisible

The 75th-percentile threshold is a deliberate design choice by Google. It is meant to represent a meaningful majority of users, not every user. That is a reasonable position for a ranking signal. It is a poor position for a business metric.

For most e-commerce sites, the users most likely to be in that bottom 25% are also the users on the worst devices and the slowest connections. They are disproportionately likely to be first-time visitors who haven't cached anything. They are, in other words, the users with the highest acquisition cost and the lowest tolerance for a bad experience, because they have no prior loyalty to offset the friction.

Chasing the green badge does not address them. It statistically excludes them from the measurement by design. An agency that celebrates the green badge without pulling the CrUX percentile distribution is telling you that three-quarters of your users are fine, which is true, and ignoring the question of who the other quarter is and what they're worth to you.

What a Diagnostic Process Actually Looks Like

Start with the CrUX data for your origin, segmented by device type and effective connection type. Compare it to your Lighthouse scores. If they agree, your lab setup is reasonably representative. If they diverge significantly, the lab setup is flattering you and you should stop reporting from it.

Then look at the distribution, not just the 75th percentile. CrUX exposes the full histogram: what percentage of your users are in the good, needs improvement, and poor buckets for each metric. A site where 74% of users are in the good bucket and 26% are in the poor bucket looks identical at the 75th percentile to a site where 90% are good and 10% are poor. They are not the same site from a revenue perspective.

From there, the conversation changes. You're no longer asking how to get the badge. You're asking which users are having poor experiences, what those experiences look like on their actual devices, and whether the cost of fixing that segment is justified by what that segment is worth. That is a business question. It requires real data. It is also the question that almost never gets asked in a standard CWV audit, because the badge is easier to deliver than the answer.

If the next agency you talk to opens with their PageSpeed methodology rather than with a CrUX pull for your domain, you already know something about how much of your 25% they intend to think about.