I put Lighthouse scores on my case studies, so it may seem odd for me to argue that a perfect 100 matters less than people think. I'll make the case anyway.
Lighthouse scores have become a sales prop. An agency screenshots a green circle and calls it proof. Sometimes it is proof. Often it's a desktop score, taken on a stripped-down test page, from the one run out of five that came back highest. It helps to know which kind you're looking at.
What Lighthouse measures#
Lighthouse is a lab test. A program loads your page on a simulated device with a throttled connection and grades the result from 0 to 100. Run it several times on the same page with the same tool and you can get different scores, which is normal. It's a controlled experiment on one simulated visit, and it doesn't record anything your visitors go through.
When I read a score, I keep two things in mind:
- A low score almost always points at problems on the page. A 40 on mobile means the people using your site are having a bad time.
- A perfect score means the lab run was clean. On its own, it doesn't tell you whether the site makes money.
The score is a composite of several metrics with different weights. Three of them are worth your attention, because Google also measures them on your visitors.
The three numbers worth watching#
These are the Core Web Vitals. Google collects them from Chrome users on your site and uses them as one signal among many in ranking.
| Metric | What it measures | What your customer feels |
|---|---|---|
| LCP | Time until the main content is visible | "Is this thing loading or not?" |
| INP | How fast the page reacts to a tap or click | "Did my tap even register?" |
| CLS | How much the layout jumps around | "The button moved and I hit the wrong thing" |
LCP covers the first impression, and with it the money you spent getting that visitor to the page. INP covers the moment someone starts using your quote form. CLS decides whether they end up rage-tapping the wrong link. These three measure what people experience on your site. The composite score is a summary of one lab run.
Where score-chasing goes wrong#
Lab data and field data can disagree. The score comes from one simulated load. Field data comes from thousands of visitors on their own phones and networks. A site can score high in the lab and still feel slow to the people using it, and when the two disagree, I trust the field data. Google publishes field data through the Chrome UX Report, and PageSpeed Insights shows it at the top of the results once a site has enough traffic. It's free to check, and because it comes from Chrome users' own visits, it's hard to fake.
Inflated scores. There are ways to raise a score without improving anything for a person using the site: deferring scripts so the test finishes before the heavy code arrives, lazy-loading content the test never scrolls to, or stripping out features that made the business money. The circle turns green, and visitors get the same site or a worse one.
Desktop screenshots. Desktop scores run far higher than mobile. If someone shows you a 100 without saying which device, which page and which data, you're looking at the most flattering number they could generate.
How to check a performance claim#
This works for claims from anyone, including me.
- Ask for the mobile score, since desktop numbers run higher.
- Ask whether it's lab data or field data from real users.
- Ask for the URL and run the test yourself. It's free.
- Ask what LCP, INP and CLS look like, in plain terms.
I publish my scores with the device and the date they were measured, so anyone can check them. What I'm aiming for is a site that loads before your customer's patience runs out, reacts when they tap and holds still while they read. Get those right and the green circle tends to show up on its own. If you chase the circle first, you can end up with a perfect score on a site that still loses customers.
If someone pitches you with a 100 and can't say which device or whose data it came from, be careful with the rest of the pitch.