Skip to content
Permo Docs

PageSpeed Insights and Lighthouse

PageSpeed Insights is Google’s tool for testing how a web page loads. It opens the page on a simulated phone and a simulated desktop computer in a Google data center, measures what happens, and lists what slowed it down. When enough Chrome users have visited the page, it adds how the page performed for them.

The test itself is run by Lighthouse, Google’s open-source testing tool. It’s the same tool that’s built into Chrome’s developer tools.

Why it matters

Section titled “Why it matters”

A test shows what a first-time visitor on a typical phone experiences: how long they stare at a blank screen, when the main content appears, whether the page jumps around while it loads. It also points to the files and scripts behind the delays. That makes it a good starting point, even though it’s one load of one page under fixed conditions.

The report has two parts that are easy to mix up. The lab test is the simulated load Google runs on demand. Real users is data collected from Chrome visitors over the last 28 days. They answer different questions, and they often show different numbers; see Lab data and field data.

How the lab test works

Section titled “How the lab test works”

Test device

Section titled “Test device”

The lab test doesn’t use your phone or your computer. Lighthouse emulates a device: a mid-range Android phone for the mobile test and a desktop computer for the desktop test, each with its own screen size. The same page can do well on desktop and poorly on mobile, because the simulated phone has less processing power and a slower connection.

Connection

Section titled “Connection”

The mobile test simulates a slow mobile connection, which Google calls slow 4G. The desktop test assumes a fast wired connection. Neither depends on the connection of the data center or of your own office.

Google doesn’t actually slow down the test. It loads the page once at full speed and then calculates how long each step would have taken on the slower device and connection. That’s why the filmstrip and the waterfall, which come from the fast load, show shorter times than the metrics.

User agent

Section titled “User agent”

Every browser tells the server who’s asking, in a line of text called the user agent. The lab test sends a user agent that describes Chrome on the emulated phone or desktop. Your server sees that text, but the request comes from Google’s data center, over Google’s network, from a test computer that’s neither the phone nor the desktop it describes.

Lighthouse version

Section titled “Lighthouse version”

Lighthouse is updated regularly, and PageSpeed Insights switches to new releases over time. A new release can change how a metric is calculated, which checks run or how much they count. So a score can move even though the page stayed exactly the same. When you compare two tests, check that they ran on the same Lighthouse version.

What the report shows

Section titled “What the report shows”

Filmstrip

Section titled “Filmstrip”

A row of screenshots taken while the page loaded. They show what a visitor would have seen at each moment: a blank screen, the first text, the main image, the finished layout.

Waterfall

Section titled “Waterfall”

The test records every file the browser loaded for the page: when each request started, how long it took and how large it was. Drawn as a waterfall, with one bar per file in the order Google recorded them, it shows what loaded first and what had to wait. The main-thread work adds when the browser was too busy to do anything else.

Findings

Section titled “Findings”

Checks the page didn’t pass, such as images that are larger than needed or scripts that hold up the first content. Many come with an estimate of the time or data they might save. A finding isn’t an error; it’s a condition the test noticed. See Audits and estimated savings.

Other categories

Section titled “Other categories”

Besides Performance, the same test produces three more scores: Accessibility, Best Practices and SEO. They come from the same load of the page, but they don’t count toward the Performance score.

Why results vary

Section titled “Why results vary”

Run the same test twice and the numbers rarely match. Google’s test machines differ slightly from run to run, and so does the path through the network. On a WordPress site, the page itself often changes too: an expired page cache means the next test waits for WordPress to build the page again, an ad or embed loads different content, or a plugin update adds a script. Small swings are normal. A change that holds over several tests says more than a single result.

In Permo

Section titled “In Permo”

Permo has Google run this test for each monitored page, once for mobile and once for desktop, and shows the results on its PageSpeed page: the scores and lab metrics per device, below them the Real users section when Google has data, then the findings in Google’s order, the filmstrip, the waterfall and details about the test machine. The filmstrip marks the screenshots where the first content and the largest content appeared. Above the waterfall, a strip shows when the browser’s main thread was busy; tall blocks are tasks longer than 50 milliseconds.

Reference values

Section titled “Reference values”
  • Lighthouse’s mobile test simulates a connection with 150 ms latency and 1.6 Mbps download speed, and slows the processor down four times (source: Lighthouse, Network throttling, accessed October 16, 2026).
  • PageSpeed Insights rates a score of 90 or above as good, 50 to 89 as needing improvement and below 50 as poor (source: About PageSpeed Insights, accessed October 16, 2026).
Section titled “Related”

Sources