How to Read a Lighthouse Report Without Chasing 100
You run Lighthouse, get a performance score of 62, fix three things, run it again and get 71. Run it a third time without changing anything and it says 58.

You run Lighthouse, get a performance score of 62, fix three things, run it again and get 71. Run it a third time without changing anything and it says 58. At that point it is fair to wonder whether the tool is measuring your site or the weather.
A Lighthouse report is useful, but only if you read it in the right order and know which numbers move for reasons that have nothing to do with your code. This walkthrough covers how I read one, what to fix first, and why chasing a perfect 100 is usually a waste of an afternoon.
Run the Lighthouse report under steady conditions
Most confusing results come from the way the test was run, not from the page. Before reading anything, control the setup:
- Use an incognito window or a fresh browser profile. Extensions inject scripts and can add hundreds of milliseconds to main thread work.
- Close other heavy tabs. Lighthouse in DevTools uses your own CPU, and a video call in the next tab will skew it.
- Pick mobile, and stick with it. Mobile mode applies simulated throttling, roughly a mid range phone on a slow 4G connection. Desktop scores are usually much higher, so do not compare one with the other.
- Run it three to five times and take the median. Single runs can swing by 10 points or more.
- Test the same URL each time. A homepage and an article page are different tests.
For anything you plan to track over weeks, the command line version or a CI job is more stable than clicking the button in DevTools.
Read the metrics, not the score
The big coloured number at the top is a weighted blend of five lab metrics. The weights in recent versions put most of it on three:
| Metric | What it measures | Good mobile target |
|---|---|---|
| Largest Contentful Paint (LCP) | When the main content, usually a hero image or heading, appears | 2.5 s or less |
| Total Blocking Time (TBT) | How long the main thread is busy and cannot respond to input | 200 ms or less |
| Cumulative Layout Shift (CLS) | How much the page jumps around while loading | 0.1 or less |
| First Contentful Paint (FCP) | When anything first appears | 1.8 s or less |
| Speed Index | How quickly the visible area fills in | 3.4 s or less |
Read these five values, not the score. A page can lose 20 points from TBT alone while LCP is excellent, and the fix for each is completely different. If LCP is the problem, you look at images, fonts and server response. If TBT is the problem, you look at JavaScript.
Use the filmstrip and the "view treemap" button
Just below the metrics, the filmstrip shows screenshots of the page loading. It tells you in two seconds whether the page is blank for a long time or renders quickly and then stalls. The treemap shows every JavaScript file by size and how much of it went unused. On many sites, one analytics or chat bundle accounts for a large share of the unused code.
Opportunities and diagnostics, in the right order
Below the metrics, Lighthouse lists Opportunities with estimated savings and Diagnostics without them. Sort your work like this:
- Find which metric is worst relative to its target.
- Click "Show audits relevant to" and pick that metric. The list shrinks to what actually affects it.
- Fix the top item with the largest estimated saving. Common winners are "Eliminate render-blocking resources", covered in what render blocking resources are and how to fix them, and "Properly size images".
- Rerun three times and compare medians.
Treat estimated savings as rough. They are calculated against the simulated connection, so a 1.2 second saving in the report might be 300 milliseconds for a real visitor on fibre. The direction is reliable. The size is not.
When an audit points at a slow request, cross check it in the Network panel. Our guide on reading the DevTools network waterfall shows how to see which file is really holding up the page.
Why chasing a 100 score is the wrong goal
A lot of performance advice treats 100 as the finish line. I think that is a mistake for most real sites, for three reasons.
First, the scoring curve is steep at the top. Going from 90 to 100 often means shaving a few hundred milliseconds of blocking time that users on decent hardware will never notice, while costing days of work removing features people use.
Second, the lab score is a single simulated visit. What actually matters for users, and what search engines use as a page experience signal, is field data: real visits measured over 28 days, shown as Core Web Vitals in the "Discover what your real users are experiencing" section at the top of PageSpeed Insights when enough traffic exists. A page can score 75 in the lab and pass every field metric comfortably, or score 95 and fail INP, a field metric that Lighthouse's default navigation run cannot measure at all.
Third, a fixed score target encourages gaming. Lazy loading the hero image to raise one number, for example, usually makes LCP worse for real people.
A better goal: every lab metric in the green on mobile, and field data passing. Past that point, spend the time on bugs or accessibility.
Do not skip the other three categories
Lighthouse also scores Accessibility, Best Practices and SEO. These are mostly pass or fail checks and are often quicker to fix than anything in Performance. Missing form labels, low contrast text and images without alt attributes show up here. They are not a full accessibility audit, since automated checks catch only a part of real problems, but they are a cheap first pass.
If you want to go further with the tools you already have open, our list of DevTools features most developers never open includes the performance panel features that pick up where Lighthouse stops.
Frequently asked questions
Why does my Lighthouse score change every time I run it?
Lighthouse measures a single load on your machine, so CPU load, extensions, network jitter and server response time all affect it. Variation of 5 to 10 points is normal. Run it several times in a clean profile and use the median.
What is a good Lighthouse performance score?
A score of 90 or above is rated good, 50 to 89 needs improvement, and below 50 is poor. In practice, aim for all individual metrics in the green on mobile rather than a specific total. Field data from real users matters more than the lab number.
Is Lighthouse the same as PageSpeed Insights?
PageSpeed Insights runs Lighthouse on a remote server and also shows field data from real visitors when it is available. Lighthouse in DevTools runs on your own computer and shows lab data only. Use PageSpeed Insights for real user data and DevTools for quick checks while developing.
More in Dev Tools
Dev Tools
How to Use the DevTools Coverage Tab to Find Dead CSS
A stylesheet on a project I inherited was 184KB minified, one file for the whole site.
Dev Tools
How to Use curl Timing to Find Why an API Request Is Slow
"The API is slow" is one of the least useful bug reports, and also one of the most common.
Dev Tools
Git Rebase vs Merge and How to Choose for Your Team
Your feature branch is three days old, main has moved on by 40 commits, and the pull request page says "This branch is out of...