A single PageSpeed run is not a measurement

Someone tells you your website is slow. You run Google’s PageSpeed Insights, get a number in a coloured circle, and it feels like a verdict. Then you run it again ten minutes later and the number has moved by nine points, on a site nobody has touched.

Both numbers are real. Neither of them is the measurement. That distinction is worth understanding before you spend money on the basis of one, because the score is not a property of your website in the way a file size is. It is the outcome of a simulated page load on a simulated phone on a simulated network, and simulations vary.

What follows is what the number is actually made of, why it moves, and the part of the report that most people scroll straight past even though it is the part that matters.

What the score is made of

The performance score is a weighted average of five measurements, not a judgement. The report links its own scoring calculator, and it is worth opening once, because the weightings explain most of what looks like randomness.

How quickly something appears

The first measurement is the moment anything at all is drawn on screen. The second, and much more important one, is the moment the largest thing is drawn — usually your main heading, your hero image, or whatever occupies the most space above the fold.

That second one carries a quarter of the score on its own, and it is the one most sites fail. It is also the one with a trap in it: it is not “the biggest thing at first paint”. It is the biggest thing painted at any point before someone interacts, and it updates every time that element is redrawn. Delaying a big element does not take it out of the running. It just gives it a later timestamp, which scores worse than if you had never delayed it.

Whether the page responds when touched

The next measurement is how long the browser spends locked up running code instead of listening for taps. This is where third-party scripts land: chat widgets, tracking tags, cookie tools, social embeds. Each one is small on its own and the total is rarely small.

It carries the largest single share of the score, which is why an otherwise light page can score badly for reasons that have nothing to do with how it looks.

Whether things move about while loading

The last one you will recognise immediately as a reader, even if you have never heard it named: you go to tap a link, an advert or an image finishes loading above it, everything shifts down, and you tap the wrong thing.

This is the cheapest of the three to fix and the most irritating to experience. It is usually images and embeds without their dimensions declared, so the browser does not know how much room to leave.

Why the number moves when nothing changed

The test is a single load, run once, on a shared machine in a data centre, over a throttled connection, on a simulated mid-range phone. Any of those can be having a slightly worse moment than the last time you asked.

Add to that the third parties. Your analytics tag, your fonts, your chat widget — none of those are served by you, and their response times move independently of anything you control. One slow response from someone else’s server can move your score by several points.

The practical rule that falls out of this: run it at least three times and take the middle result. Do not act on a single run in either direction. A single run will happily confirm that a change you made worked when it did nothing, or condemn a change that genuinely helped. And do not treat a difference of a few points as a result at all — the noise is wider than that.

The box almost everybody scrolls past

At the very top of the report, above the coloured circle, there is a section about what real users are experiencing. If your site has enough traffic, it will show data collected from actual Chrome users visiting your actual site on their actual phones over the previous month.

That is the real measurement. The score underneath it is a laboratory estimate designed to be actionable; the box at the top is the truth. They frequently disagree, and when they do, the box at the top wins.

It is also the one Google itself uses when page experience factors into ranking. So if the box says your real visitors are fine and the circle says 71, your problem is smaller than the circle suggests. If it says “no data”, your site does not yet get enough traffic for Google to have collected any, which is information about your traffic rather than about your speed.

When the score is genuinely telling you something

None of this means the number is worthless. It means it is a smoke alarm rather than a diagnosis. Take it seriously when:

• It is consistently bad. Three runs in the forties is a real finding. One run in the forties among three in the eighties is weather.
• The same item appears in the list of opportunities every time. The list beneath the score is far more useful than the score, and it is stable even when the number is not.
• The real-user box is amber or red. That is not a simulation and it does not need a median.
• It got dramatically worse after a specific change. Then you have a before and an after, which is the only comparison that means anything.

Equally, be sceptical of anyone selling you a rebuild on the strength of one screenshot of a bad score. It is the easiest number in this industry to produce on demand, and the work that actually makes a page faster is rarely a redesign.

What to do first

In order, and none of it needs a developer until step four.

1. Run the test three times. On mobile, and write down all three. That is your baseline, and you now know your own noise floor.
2. Read the real-user box first. It sits above the score. If it says you are fine, breathe out.
3. Ignore the number, read the opportunities. Note anything that appears in all three runs.
4. Take the shortlist, not the score. Give it to whoever maintains your site and ask which items are cheap and which are structural.
5. Re-test three times after each change. Only believe a difference of more than a few points.

One more thing worth saying, since it is the honest version. A fast site that nobody enquires through is still a site nobody enquires through. Speed removes a reason to leave; it does not supply a reason to stay, and the place enquiries actually leak is usually further down the page than the homepage. Fix the score when it is genuinely bad, then go back to the thing the site is for.

If you would rather not read the report yourself, send us the address and we will run it properly and tell you which findings are worth your money — including the ones where the answer is that you should leave it alone.

Run it three times before you do anything else. Most of what people believe about their site’s speed comes from having run it once.