PageSpeed Insights explained: why you don't need 100

Most people take the wrong message from this tool. They get a red 47 and lose their head, or they get a green 92 and decide the website’s sorted. Both reactions miss the point, because the number on its own doesn’t tell you much.

What is PageSpeed Insights?

It’s a free tool from Google at pagespeed.web.dev. You paste your web address in, it tests the page, and it gives you a score out of 100 with a coloured circle round it:

• Red, 0 to 49. Your website needs some serious work.
• Orange, 50 to 89. Needs work. Not the end of the world.
• Green, 90 to 100. Good. Leave it alone.

You get two scores, one for mobile and one for desktop, and they’ll be different. Look at mobile first. On most of the websites we look after, about 80% of visits come from phones, and the mobile test is the harsher one. Desktop is the easy exam. (If your desktop score is lower than your mobile score, something’s gone very wrong, because it should be the other way round.)

Under the bonnet, the score comes from a Google tool called Lighthouse, which you can also run inside Chrome if you fancy it. PageSpeed Insights runs it for you online and adds a load of extra jargon on top, just to keep things interesting.

The big score is a lab test

Hardly anyone explains this part, and it changes how you read the whole page.

That number out of 100 is a simulation. Google loads your page on a pretend phone over a pretend slow connection and times it. No real people are involved.

Near the top of the results there’s a separate section headed “Discover what your real users are experiencing”. That’s data from actual people visiting your website in Chrome, on their own phones, over roughly the last month. Google collects it in something called the Chrome UX Report. If that section and the big number disagree, believe the real people.

The gap can be huge. When I ran Cannon Hall Farm’s website through it for a video I was planning (we built their ticketing system, and it’s a busy attraction with proper traffic), the real-user data said the main content was showing up in 1.6 seconds. The lab test on the same page, the same day, said 6 seconds. The pretend phone reckoned the site was nearly four times slower than what thousands of actual visitors were getting.

One catch. If your website doesn’t get much traffic, the real-user section can be empty, because Google hasn’t got enough visitors to go on. Then the lab number is what you go by. It’s deliberately harsh, so a small, perfectly quick website can look slower than it is. Get it green and you’re fine.

What do all the letters mean?

PageSpeed Insights loves an abbreviation. Strip them out and it’s asking five questions about your page:

• FCP (First Contentful Paint). How long until I see anything at all?
• LCP (Largest Contentful Paint). How long until the biggest thing shows up, like the main photo or headline? Google counts under 2.5 seconds as good and over 4 seconds as poor.
• TBT (Total Blocking Time). Is the page frozen, and for how long? You want this as close to zero as you can get it.
• Speed Index. How quickly does the page look finished?
• CLS (Cumulative Layout Shift). Does stuff jump about while it loads? You start scrolling, an image or a font loads in late, and you lose your place. News websites are terrible for it.

In the real-user section, TBT gets swapped for something called INP, which measures how fast the page reacts when you tap something. You know when you tap a button, nothing happens, so you tap it again and it does it twice? That’s bad INP.

When I recorded the video, our own website scored 87 (orange) and our CLS was 0.17, in red. I should have fixed it before I made a video telling other people to.

▶Play the video: Google PageSpeed Insights explained, and why 100 is a trap

The video walks through all of this on screen.

Why you don’t need 100

Everyone treats the score like an exam where you need full marks. Once it’s green, put it to one side. Keep checking it, but don’t spend money getting from 94 to 96, because nobody visiting your website can tell the difference.

We built an ecommerce site for a client that sat at about 76 on mobile. For a shop stuffed with products and photos, that’s fine. Then he found PageSpeed Insights, and every week it was “can we get it higher?” So we chased it. Hours of fiddling with fonts and lazy-loading things, and we got it up to around 82. It did nothing for a single customer. Then he pasted some marketing tracking code into the site himself and it tanked the score in an afternoon.

Those hours would have been far better spent on the stuff that brings in work, like clearer words or putting your prices on the website.

Does your PageSpeed score affect your Google ranking?

Yes. Google uses page speed as a ranking factor. Strictly speaking, what it ranks you on is Core Web Vitals (LCP, INP and CLS) measured on real visitors, which is the data in that real-user section. The score out of 100 tests the same sort of things in the lab, so a high score is a good sign your Core Web Vitals are healthy too. That’s why I keep it simple and say get it green.

If your website doesn’t get enough traffic to have any real-user data, the score out of 100 is the only measure you’ve got. Use it, and get it green.

A slow website gets hit twice. People leave because they’re fed up of waiting, and Google ranks it lower. That’s why green is the target.

So why care about speed at all?

Because people leave. A slow website loses visitors before they’ve seen a thing, and the score is a decent warning light for that.

Wolf Laundry were spending £3,000 a month on Google Ads and getting zero leads. Their website took ten seconds to load, and over half the people they were paying Google to send gave up and went back before the page had finished loading. They paid for those clicks anyway. We rebuilt the site, it loads in about a second now, and it brings in 150 to 200 leads a month. Rucklidge Law went from over six seconds to around two as part of their redesign, and conversion rates went up by as much as 35%.

What’s dragging your score down?

I’d put money on it being the images. Photos straight off a camera, four thousand pixels wide, sat on a homepage that shows them at 600. Compress them and size them properly, and your score goes up and the site gets faster for real people as well.

After images, it’s a pile of plugins and scripts you forgot you installed (chat widgets, popups, tracking pixels, fancy animations), too many fonts, no caching, or cheap £2-a-month hosting.

PageSpeed Insights lists what to fix in order, biggest first. Fix the top one or two, run it again, and don’t bother with the rest unless you’re still in the red.

The score also moves about from one day to the next. We check it every day, on mobile and desktop, for every website we look after, and plot it on a graph over the month. One bad run tells you very little.

A fast website is brilliant, but if the people who land on it can’t work out what you do in about four seconds, the speed doesn’t matter much. I’ve written about what should go on a homepage for that bit.

And if you’ve got somebody in the business who likes pasting tracking code into the website, hide the login.