Lighthouse and PageSpeed

Why getting faster barely moves your Lighthouse score

You made the page faster. The Lighthouse performance score and PageSpeed Insights grade barely moved. Here is why, and what to track instead.

9 September 2026 · Lab scores, curves, and honest reporting

You made the page faster. Visitors should feel it. You open Lighthouse or PageSpeed Insights, run a mobile check, and the performance score barely moves.

That feels unfair. It is also how the Lighthouse performance score is built.

This guide explains, in plain words:

• why a real wait can drop while the grade stays almost flat
• how Largest Contentful Paint (the big picture time) pays points on a curve
• why long PageSpeed tip lists often do not raise the score
• how Core Web Vitals in Search Console differ from one lab run
• what to write in a pull request so nobody argues from one screenshot

Chrome is enough. I also run a free three run check at pafcore.site when you want the quiet swing and what to fix first without babysitting Lighthouse by hand.

Lab score vs what Google ranks

People mix three different things:

1. Lighthouse performance score in Chrome DevTools (one lab run on your machine or a tool’s machine)
2. PageSpeed Insights (Lighthouse lab plus Chrome UX Report field data when Google has enough visits)
3. Core Web Vitals in Search Console (mostly real visitor data over weeks, the part that ties to search)

A fix can help people today and still look small on the lab grade. Field Core Web Vitals can take weeks to move. Do not expect one afternoon of Lighthouse screenshots to match Search Console.

The score is a curve, not a straight line

The performance score is not “each millisecond equals one point.” Key lab metrics, including Largest Contentful Paint (how long until the main content shows), sit on a curve. Far out on a slow page, big time wins buy almost no points. Closer to a good target, the same kind of win pays more.

A useful picture:

• Around 7.3 seconds for the biggest thing on screen, you are near the flat end. Marks out of 25 for that metric can sit around 1.
• Around 2.5 seconds, those marks sit near the top of that bucket.

So cutting from 7.3s to 6s can feel better for visitors and still look like almost nothing on the grade. That is not a broken Lighthouse report. That is the curve.

PageSpeed Insights uses the same lab maths for the performance score. Same curve. Same surprise. For the full mark tables, see How the Lighthouse score is actually calculated.

Why the tip list fools teams

PageSpeed and Lighthouse show a long list of opportunities. Caching, unused JavaScript, image formats, and more. Most rows are useful cleanup. Most rows are not direct inputs to the performance score.

The grade comes from a small set of lab metrics (Largest Contentful Paint, Total Blocking Time, Cumulative Layout Shift, First Contentful Paint, Speed Index, and similar). A tip only raises the score if fixing it improves one of those metrics.

Before you put a tip on the board, say out loud:

“If this works, which score metric gets better?”

No answer means cleanup. Cleanup is fine. Cleanup is not a score promise.

When the number moves with no deploy, read Why your Lighthouse score changes when you did nothing.

Mobile vs desktop on the same URL

Always label the device. Mobile Lighthouse and desktop Lighthouse are different tests. Throttling is harsher on phone mode. The same site can look fine on desktop PageSpeed and weak on mobile. Search care is closer to the hard phone path for many sites.

When you compare before and after: same URL, same mode (mobile or desktop), more than one run. One desktop 94 does not cancel a mobile problem.

What to track instead of only the grade

Keep the performance score. Add the wait you meant to fix, in seconds:

Now you can say something honest: visitors should feel a clearer first picture. The score is still noisy and may stay flat while you sit on the slow end of the curve. That is better than claiming “we gained two points” as if two points were the whole story.

A short habit after each speed fix

1. Pick mobile or desktop and keep it.
2. Run Lighthouse or PageSpeed three times before and three times after (or use pafcore.site).
3. Compare middle scores to the quiet gap, not to a single lucky run.
4. Write the target wait in seconds next to the grade.
5. Only call it a clear score win when the middle rise beats the quiet gap.

For the quiet gap steps in full, use the earlier article. This piece is about why a real speed up can still look small on the Lighthouse performance score and on PageSpeed Insights.

Pull request note you can paste

Tool: Lighthouse / PageSpeed Insights, mobile (or desktop), three runs before and after.
Quiet gap (no deploy):
Middle performance score before / after:
Target wait (e.g. LCP) before / after:
Curve note: slow pages pay few points until the wait gets much better.
Clear score win? yes / not sure / no

Conclusion

Getting faster for people and getting a higher Lighthouse score are related. They are not the same job. The performance score sits on a curve. PageSpeed tip lists mix scored metrics with cleanup. Core Web Vitals in search move on field data over time, not on one lab screenshot.

Measure the wait you care about. Run more than once. Talk about both the grade and the seconds. That is how you stay honest with clients and with yourself.

I built pafcore.site so the three run spread and “what to fix first” sit in one free report. Use Chrome when you want full control. Use the site when you want that habit done in one paste.