1. Home
  2. Guides
  3. Read a Lighthouse report
Performance guide

How to read a Lighthouse report

Lighthouse is the open-source auditing tool behind many speed reports, including the one on this site. Its output is dense, so here is how to read it: what the scores mean, which numbers matter, and what to fix first.

Key points

  • Lighthouse gives four scores from 0 to 100: Performance, Accessibility, Best Practices and SEO. 90–100 is good, 50–89 needs work, 0–49 is poor.
  • The Performance score is a weighted mix of lab metrics. Most of the weight sits on Total Blocking Time, Largest Contentful Paint and Cumulative Layout Shift.
  • A score is a summary. Fix findings by their estimated savings, not by chasing the number.

What Lighthouse is

Lighthouse is an open-source tool maintained by the Chrome team. It loads a page in headless Chrome, records what happens, and runs a set of automated checks. It is the engine behind the Lighthouse panel in Chrome DevTools and the lab section of PageSpeed Insights. WebsiteHealthLabs runs Lighthouse on its own servers, so every report shows the Lighthouse version it used.

By default Lighthouse tests mobile: it emulates a mid-range phone and simulates a slow 4G connection with CPU slowdown. Desktop runs use a faster profile. Because the test is simulated and run from one location, results vary slightly from run to run.

The four scores

CategoryWhat it covers
PerformanceHow quickly the page loads and becomes usable, from lab metrics.
AccessibilityAutomated checks such as colour contrast, alt text, form labels and ARIA use.
Best PracticesModern web practice: HTTPS, console errors, deprecated APIs, image aspect ratios.
SEOBasic crawlability and on-page checks: title, description, viewport, link text, robots.

Scores are coloured by band: 90–100 good, 50–89 needs improvement, 0–49 poor. Reaching 100 is not the goal. A score in the nineties on a real, content-rich page is excellent.

Automated is not complete. Lighthouse can only test what a machine can detect. A perfect Accessibility score does not mean a page is accessible, and a perfect SEO score does not mean it will rank.

How the Performance score is built

The Performance score is a weighted average of metric scores. In current Lighthouse versions the weights are approximately:

MetricWeightWhat it tells you
Total Blocking Time30%How long the main thread was blocked while loading.
Largest Contentful Paint25%When the main content appeared.
Cumulative Layout Shift25%How much the layout moved.
First Contentful Paint10%When anything first appeared.
Speed Index10%How quickly the visible area filled in.

Weights are revised between major versions, so check the Lighthouse documentation for the version you use. The practical lesson is stable: blocking JavaScript, a slow main element and layout shifts are what cost points.

Reading the metrics

  • First Contentful Paint (FCP): first text or image painted. A slow FCP points to a slow server or render-blocking resources.
  • Largest Contentful Paint (LCP): the big one for “does it look loaded”. See the LCP guide.
  • Total Blocking Time (TBT): a lab proxy for responsiveness. See the INP guide.
  • Cumulative Layout Shift (CLS): visual stability. See the CLS guide.
  • Speed Index: how quickly the visible content is populated during load.
  • Server response time: how long the root document took. A high value delays everything else.

Opportunities, insights and diagnostics

Below the metrics, Lighthouse lists audits that did not pass. Each has a title, an explanation and, for many, an estimate of how much time or data could be saved. Read them in this order:

  1. Largest estimated savings first. They affect the metrics that carry the most weight.
  2. Audits that mention your LCP element or render-blocking resources.
  3. Diagnostics last. They explain why a metric is what it is, but may not have a direct fix.

WebsiteHealthLabs turns these into findings with a severity: High when the estimated saving is a second or more, Medium for roughly 300 ms or more, and Low otherwise. Estimates are Lighthouse's own and do not add up neatly, because fixes overlap.

What lab data cannot tell you

  • It is one visit from one location on a simulated device. Your visitors use many.
  • It does not include interactions, so it cannot measure INP.
  • Logged-in pages, cookie banners and personalised content may look different to real users.
  • It is a snapshot. Run it a few times and compare, and use field data (Core Web Vitals reports) to confirm improvements.

See also: Accessibility basics, HTTP headers explained.

FAQ

Why does my score change between runs?

Lighthouse simulates the slow network and CPU, and real conditions on the test machine and your server vary. Differences of a few points are normal. Run the test several times and look at the trend.

Is a score of 100 necessary?

No. Chasing the last points often costs more than it earns. Aim for the green range and make sure the Core Web Vitals metrics are good.

Does Lighthouse affect my Google rankings?

The score itself is not used for ranking. Google uses real-user Core Web Vitals data, among many other signals. Lighthouse helps you find problems that affect those metrics.

WebsiteHealthLabs Editorial
Guides explain audit results in plain English. Facts are checked against public documentation from Google, W3C and the browser vendors, and we update them when guidance changes.

↑ ↓ to moveEnter to openEsc to closeAll results