PageSpeed Insights Now Uses Lighthouse

A different number this week

Somebody ran their home page through PageSpeed Insights yesterday and found the score had dropped sharply. Nothing on the site had changed.

The tool had. On Monday, Google switched PageSpeed Insights to Lighthouse, the testing engine already built into Chrome's developer tools, and the score now measures something different.

Some pages scored better, plenty scored worse, and almost every number moved.

Before anyone rebuilds a site over it, it is worth knowing what the new number measures.

What the old score measured

The old optimisation score checked a page against a list of rules: images compressed, files cached, code trimmed, nothing holding up the first drawing of the page.

A page could follow most of the rules and still be slow, and a quick page could lose points for missing one. It measured good habits, not time.

Since January the tool had also shown real-world speed from Chrome users where there was enough data, but most small sites never had enough.

What the new report shows

• A performance score from 0 to 100, from a simulated phone load.
• Field data from real Chrome users, where the page has enough visitors.
• Lab data: timings such as first contentful paint and time to interactive.
• Opportunities: changes, each with an estimate of the time it would save.
• Diagnostics: more detail on how the page is built.
• Passed audits: the things the page already does well.

The fourth is where the work is. The score says how the page did; the opportunities say why, and roughly what each fix is worth in seconds.

How the score is measured

Lighthouse loads the page as a mid-range phone on a slow mobile connection would, and times how long it takes to show content and to become usable.

First contentful paint is when anything from the page first appears. Time to interactive is when the page can reliably respond to a tap. The score combines several timings like these.

Ninety and above shows in green; lower scores turn orange, then red.

The phone test is deliberately harsh. A page that feels fine on the office Wi-Fi can score badly, because many customers are not on office Wi-Fi.

A desktop tab scores more kindly, but the mobile result is closer to how most people now browse.

Why the score moves between runs

Run the same page three times and the score may differ each time. Server response varies from moment to moment, and ads, chat widgets and other scripts load differently on each visit.

Treat a single run as an estimate. Run important pages several times, note the range, and compare like with like: the same page, the same tab, before and after a change.

Reading the opportunities

Each opportunity names a change and estimates the saving. Start at the top, where the estimates are largest, and leave anything worth a fraction of a second.

The usual ones on small business sites are images larger than they are shown, images below the first screen loading too early, and scripts that hold up the page. A slow server response shows up too, often among the diagnostics.

One needs care for now: serving images in next-generation formats. That mostly means WebP, which works in Chrome and the newest Edge but not in Safari or Firefox, so JPEG and PNG copies still have to be kept.

Diagnostics add detail, such as how much the page asks a phone to download, and are mostly for whoever maintains the site.

A worked example

A luggage store with two locations sells suitcases, carry-ons and travel accessories, and its website lists the range with prices and stock at each store.

Its home page had sat in the green on the old tool for years. Yesterday the owner ran it again, as he does every month, and it came back red.

The old score rewarded the compressed images and caching the site already had. The new one timed a phone waiting on a five-photograph slider, two web fonts, a reviews widget and a chat script.

The opportunities list said the same in its own terms: images larger than shown, and scripts delaying the first content.

The slider is becoming one photograph sized for phones, the chat script will load after the page, and the reviews widget is moving below the first screen.

The aim is a page that appears quickly on an older phone, not a particular number.

Not chasing a perfect number

A score of 100 is rare for a real business page with analytics, a map, a booking widget and photographs that sell. Chasing it usually means removing things customers use.

Google has said its Speed Update affects only the slowest pages, and that no tool reports whether a page is affected. The score is a guide to slowness, not a ranking.

The test that settles it is still the real one: your page, on an ordinary phone, on mobile data, timed by someone counting.

Lab data and field data

Where a page has enough Chrome visitors, the report shows real-world figures above the lab test, and the two often differ.

Field data covers your actual visitors, their phones and their connections, over roughly the past month. The lab test is one simulated visit on a fixed setup.

When the two disagree, trust the field data for what customers actually go through, and use the lab test to see whether a change helped.

The counter-case

A page that appears quickly on a real phone does not need rebuilding because a lab score turned orange.

Some low scores come from things worth keeping, such as a map on the contact page or a booking widget, and the cost is worth paying where those pages earn it.

But a page in the red deserves an hour with the opportunities list. Work from the top, test on a phone, and stop when the page is quick, not when the number is perfect.

This week

1. Run your main pages three times each.
2. Note the range, not a single score.
3. Read the mobile tab before the desktop one.
4. Work through opportunities from the largest saving.
5. Keep JPEG copies if you try WebP.
6. Test on a real phone over mobile data.
7. Rerun after each change, on the same tab.

Step four is where an hour goes furthest. The estimates are rough, but they rank the work, and the top items on most small business pages are images and scripts.

Where a test runs from changes its result too, which is covered in speed tests from Canada and from Virginia.

Frequently asked questions

West Coast Media Solutions Inc. provides web design, web development, hosting, digital marketing, and business consulting to organisations across Canada, drawing on more than twenty-five years in the field.