A supplier sends you a screenshot with a green circle and the number 98. Or a competitor sends your marketing team a link where your site scores forty two in red. In both cases somebody is showing you the output of PageSpeed Insights, and in both cases there is a fair chance you have no idea what that number means or whether it is a problem.
This article is for business owners, not developers. It explains what the tool is, who is behind it, what the individual numbers mean, which values are good and which are bad, why a hundred points need not be the goal, and what you can genuinely take from it when deciding about your website.
What PageSpeed Insights is and why this particular tool
PageSpeed Insights is a free tool from Google. You enter a web address, wait about half a minute, and get a report on how fast the page loads and how it behaves while somebody uses it.
The fact that it comes from Google is not a detail, it is the whole reason this is the tool everybody measures with. The metrics PSI shows are also used by Google's ranking systems. The reverse does not hold: a good score does not guarantee good rankings, and we will come back to that. But it is the one yardstick everybody refers to, because it is public, free and nobody can adjust it.
The practical consequence: anybody can measure your site at any time. A customer, a competitor, or the supplier you are thinking of replacing. It is not an internal metric you can keep quiet.
PageSpeed Insights is the one yardstick for website speed everybody refers to. It is Google's, it is free, and it cannot be bent
The two different numbers PSI shows
This is the most common misunderstanding and it is worth grasping before anything else. The report has two parts that measure different things and they can happily contradict each other.
| Data from real visitors | Lab measurement | |
|---|---|---|
| Where it comes from | From real visits in the Chrome browser over the last 28 days | From a single test Google runs the moment you click |
| What it runs on | On the devices and connections your visitors actually have | On a simulated mid range phone on a throttled mobile connection |
| What it is for | It tells you how the site performs in reality. This is the truth | It tells you what can be done about it. This is the repair manual |
| When it is missing | On new and low traffic sites, because there are not enough samples | Never, the lab test always runs |
| How fast it changes | Slowly, because it is a rolling 28 day calculation | Immediately after a fix is deployed |
From which follows the first practical piece of advice. When a supplier sends you a green number, ask whether it comes from the top of the report, meaning real people, or from the bottom, meaning the simulation. A lab score can be improved by pushing work behind the first render, and that does nothing for an actual visitor.
And the second: when your site shows no field data, that is not a fault. It means the site does not have enough visits for the statistics to be calculated. The lab measurement remains, and it is still useful.
The 0 to 100 score and what hides inside it
The most conspicuous part is the coloured circle with a number. There are three bands and they are fixed:
- 90 to 100, green. Good. A visitor cannot tell 92 from 100.
- 50 to 89, orange. Room for improvement, usually reachable fairly cheaply.
- 0 to 49, red. Poor. The site loses visitors measurably, especially on mobile.
What nobody sees in that number is its composition. The score is a weighted average of five measurements and the weights are not equal.
What the performance score is made of
Composition in per cent
- Time the page cannot respond (TBT)30%
- Largest content rendered (LCP)25%
- Content shifting around (CLS)25%
- First render (FCP)10%
- How fast the screen fills (SI)10%
Total 100%
| Label | Value |
|---|---|
| Time the page cannot respond (TBT) | 30% |
| Largest content rendered (LCP) | 25% |
| Content shifting around (CLS) | 25% |
| First render (FCP) | 10% |
| How fast the screen fills (SI) | 10% |
In practice this means the score is decided mainly by how long the page stays unresponsive after loading and how quickly the main content appears. When a site scores badly, nine times out of ten one of those two is the cause, and both usually come from the same thing: too many third party scripts and images nobody prepared.
The three metrics that genuinely matter
Google picked three of all the measurements and calls them Core Web Vitals. These three are what its ranking systems use. I will explain them as a visitor experiences them, not as a developer sees them.
LCP: when the main thing becomes visible
It measures the time from clicking a link to the moment the largest element on screen appears, usually the headline or the main photo. A visitor experiences it as the page having arrived.
INP: how quickly the site responds to a tap
It measures the delay between somebody clicking or tapping and something happening on screen. This is the most infuriating metric: the page looks loaded, but the first tap on a button does nothing.
CLS: whether the content jumps around
It measures how much the content moves while loading. The classic example is tapping a button just as an image loads above it, everything shifts, and you tap something else.
| Metric | Good | Needs improvement | Poor |
|---|---|---|---|
| LCP, main content visible | up to 2.5 s | 2.5 to 4 s | over 4 s |
| INP, response to a tap | up to 200 ms | 200 to 500 ms | over 500 ms |
| CLS, content shifting | up to 0.1 | 0.1 to 0.25 | over 0.25 |
That 75th percentile matters more than it looks. It means the average is not what gets judged; the experience of three quarters of visits is. You can have a lovely average and a poor rating because every fourth visit is slow, typically from an older phone or a weaker signal.
To make the proportions visible, here is LCP on a time axis. The difference between a good and a poor site is a second and a half, about the time it takes to draw a breath.
LCP: how long the main content may take to appear
Range in seconds
| Label | From | To |
|---|---|---|
| Good | 0 s | 2.5 s |
| Needs improvement | 2.5 s | 4 s |
| Poor, with no upper bound | 4 s | 6 s |
The other numbers in the report
Besides those three you will see further metrics in the report. You do not need to understand them deeply, it is enough to know what they mean.
- FCP, first render. When any content appears at all, even just a background and a logo. A signal to the visitor that something is happening.
- TBT, total blocking time. How long the page was busy and unresponsive after loading. It carries the greatest weight of all in the lab measurement.
- Speed Index. How quickly the screen visually fills up.
- TTFB, server response. How long before the server starts responding at all. This is where cheap hosting shows.
And then there are three more scores next to performance: accessibility, best practices and SEO. They are calculated differently, they are more stable, and it makes sense to have them in green too. Accessibility is more than a recommendation as well; for some sites it is a legal obligation.
How websites actually score
Before you panic about an orange number, it helps to know what the rest of the internet looks like. Measurements of real visits across the whole web look like this.
How many websites pass each metric on mobile
Values in per cent
| Label | Value |
|---|---|
| CLS, content shifting | 81% |
| INP, response to a tap | 77% |
| LCP, main content visible | 62% |
| All three at once | 48% |
Just under half of websites pass all three on mobile. The one that usually breaks it is LCP, the time before the main content is visible.
Two things follow at once. A green site is not a given, so it is a competitive advantage. And an orange site is not a disgrace, it is the average, and it can be climbed out of.
Why a hundred points need not be the goal
Now the part a supplier usually leaves out, because it sells badly.
The score fluctuates between runs. Measure a site three times in a row and you get three different numbers, easily a few points apart. The Lighthouse documentation says it outright: performance should be seen as a distribution of values rather than a single number. Insisting on exactly a hundred means chasing measurement noise.
The last five points cost more than the first forty. Getting a site from 55 to 90 tends to take days of work and shows up in how many people reach the content. Getting it from 95 to 100 tends to take days as well, and nobody notices.
Some points are worth sacrificing. An embedded video, a map, a chat widget, advertising tracking codes. Each of those costs score and some of them make money. The sensible approach is not to ban them but to know what they cost and then decide.
Wrong: the supplier removes the map, the chat and the ad tracking to show a hundred. The site is fast and the company has lost a source of enquiries it was paying for.
Right: the supplier works out how many points each item costs and leaves the decision to the owner. The map stays, the chat loads after the first tap, the tracking gets trimmed to what somebody actually reviews.
And above all: a green score is no guarantee of rankings. Google says so in its own documentation: good Core Web Vitals results do not guarantee a place at the top of the results, and Search always tries to show the most relevant content even when the page experience is suboptimal.
The practical conclusion: speed is one factor, and unlike content it is one you buy once. On its own it will not buy you first place.
What to do when your number is bad
A sequence worth following in this order, because the top steps decide the lower ones.
- 1Measure the mobile version, not the desktop one. Most people arrive from a phone and the mobile rating tends to be twenty to thirty points worse.
- 2Check whether the top of the report shows data from real visitors. If it does, go by that; the lab numbers are only the manual.
- 3Find which of the three metrics is red. You do not fix a score, you fix a specific metric.
- 4Slow main content is almost always images and the server. Ask your supplier about the size of the hero image and the server response time.
- 5For poor responsiveness, look for third party scripts: chat, tracking codes, widgets, plugins. Each one costs something and some of them nobody reviews any more.
- 6After the fix, wait a month and measure again. Field data is calculated over a rolling 28 days, so it does not improve immediately.
Other tools and when to use them
PageSpeed Insights is not the only tool and in some situations it is not the best one. Here is what is worth knowing about.
| Tool | What it is best for | Limitation |
|---|---|---|
| PageSpeed Insights | A quick check of one page and data from real visitors | It measures one URL, not the whole site |
| Search Console, Core Web Vitals section | An overview of the whole site and a list of failing pages | Only for a site you have verified, and the data lags |
| Lighthouse inside Chrome | Debugging during development, measuring pages behind a login | It runs on your machine and connection, so numbers are not comparable |
| GTmetrix | Readable breakdowns and a history of measurements | Free tests come from limited locations and it computes its own grade |
| WebPageTest | The most detailed analysis, testing from specific countries and devices | Aimed at developers, the output is dense |
| Paid tools such as DebugBear | Monitoring over time and alerts when the site slows down | A monthly cost, unnecessary for a small site |
For a business owner the rule is simple: PageSpeed Insights for checking and Search Console for the overview. The rest is work for your supplier.
How to run the measurement yourself in two minutes
You need nobody for this. The steps are these:
- 1Open pagespeed.web.dev and enter the address of your page. Individual subpages work too, not just the home page.
- 2Let the test finish, it takes about half a minute. Google really does load your site in the meantime.
- 3Switch to the Mobile tab at the top. Desktop tends to be the default and it is the less important half.
- 4Check whether the top of the report shows data from real visitors. If it does, those are the numbers that count.
- 5Only then look at the coloured circle with the score and the list of recommendations below it.
- 6Repeat it on two competitors' sites. That is where a number turns into information, because you see whether your value is ordinary in your field or exceptional.
That last step is the most useful one and almost nobody does it. A score on its own says nothing; only a comparison with the three sites your customer is weighing you against shows whether it is a problem.
Why some sites never reach a hundred
When a supplier says a hundred is out of reach, that is not necessarily an excuse. These are the common and legitimate reasons.
- The site runs on an off the shelf system with plugins. Every plugin loads its own chunk of code and their authors do not know one another. On a site assembled from twenty plugins, ninety points is a success.
- An online shop with many images. A product listing is inherently heavier than a page of text. It can be managed, but it will always be harder than a brochure page.
- Embedded things from other domains. A map, a video, a booking system, a chat, customer reviews. Each brings in code you do not control.
- Marketing measurement. Advertising and analytics tags are among the most common causes of poor responsiveness. Especially when there are five of them and nobody knows what each one does.
- Cheap hosting. When the server takes half a second to answer, you start with a handicap the browser cannot make up.
Three of those five are business decisions, not technical ones. Which is why a score should not be treated like a school grade but like a budget: what you buy with the points and whether it is worth it to you.
What PageSpeed Insights does not measure
And now the most important limitation of the whole tool, because this is where most decision making goes wrong.
- It does not measure whether the site sells. A site with ninety nine points and bad copy will not bring an enquiry.
- It does not measure content. It cannot tell that the page is missing a price, a datasheet or a named contact.
- It does not measure accessibility fully. The automated check finds only part of the problems; the rest needs a manual pass.
- It does not measure your admin. The speed of the system you work in cannot be measured from a public page, and yet it decides whether you will keep the site up to date.
- It does not measure your customers' actual devices. The lab part simulates one specific phone. Your customers have different ones.
So the healthy attitude is this: the score is the ticket in, not the destination. Below ninety you lose people before they read the first sentence. Above ninety, what is written on the page starts to decide.
How to read a quote that talks about speed
Here are a few sentences you will see in quotes and what to make of them.
- The site will be optimised for speed. Means nothing. Ask for a specific number and whether it is guaranteed.
- Speed optimisation as a paid extra. The question then is what the base price contains. Speed is decided by how the site is built, not by tuning afterwards.
- A score of 100 on desktop. Says little. Ask for the mobile number.
- We will send you a measurement after launch. Fine, but also ask for one after a month of operation, when data from real visitors exists.
On the sites I build, my price list guarantees a score above 90 on mobile and desktop. The home page of this site also has a field where you can enter your own address and run Google's measurement yourself. I am not the one measuring, Google is, so there is nothing to argue with in the result.
Frequently asked questions
What is PageSpeed Insights?
A free tool from Google that measures how quickly a page on your site loads and how it behaves in use. It shows two things: data from real visitors over the last 28 days, and a lab test run at the moment you measure. The metrics it reports are also used by Google's ranking systems.
What score is good?
Ninety and above is the green band, meaning good. Fifty to eighty nine is orange with room for improvement. Below fifty is red and the site loses visitors measurably. A visitor cannot tell 92 from 100; they can tell 45 from 90 instantly.
Do I need 100 points?
You do not, and it is usually not a sensible goal. The score fluctuates by a few points between runs, so chasing an exact hundred is chasing noise. The last points also tend to cost more work than the journey from fifty to ninety, and nobody notices them. A sensible goal is consistently above ninety on mobile.
Why did I get 98 on desktop and 51 on mobile?
Because the mobile test simulates a mid range phone on a throttled mobile connection, while the desktop one assumes a fast machine on a fast line. A gap of twenty to thirty points is normal. Go by the mobile number, because a large share of visits comes from phones.
What are LCP, INP and CLS?
The three metrics Google uses in its assessment. LCP is the time before the main content is visible, good is up to 2.5 seconds. INP is the response to a tap or click, good is up to 200 milliseconds. CLS measures content shifting during load, good is up to 0.1. Assessment is at the 75th percentile, the experience of three quarters of visits.
Why does my site show no data from real visitors?
Because it does not have enough visits for the statistics to be calculated, or it is too new. It is not a fault and says nothing about the quality of the site. In that case go by the lab measurement and remember it is a simulation.
Will a good score improve my search rankings?
Not on its own. Google states in its documentation that good core metrics do not guarantee a place at the top of the results and that relevant content wins even with a worse page experience. Speed is one factor, and it also decides how many people reach your content at all.
How long before fixes show up in the measurement?
In the lab measurement, immediately after deployment. In data from real visitors, after several weeks, because it is calculated over a rolling 28 days. If somebody promises field data will improve by tomorrow, they do not know how it is calculated.
Do I have to remove the map and the chat to get a score?
You do not. The sensible approach is to know how many points each item costs and then decide whether it pays for the business. A map can load after a click, a chat after the first tap, and tracking codes can be trimmed to what somebody actually reviews.
Summary
- PageSpeed Insights is a free tool from Google and the yardstick everybody refers to, because it is public and cannot be bent.
- The report has two parts: data from real visitors over 28 days and a lab simulation. The first says how things are, the second says what to do about it.
- The green band is 90 and above, orange is 50 to 89, red is below 50. Fifty five per cent of the score is decided by how long the page stays unresponsive and how long the main content takes to appear.
- The three metrics Google uses: main content within 2.5 seconds, response within 200 milliseconds, content shifting within 0.1.
- Only 48 per cent of websites pass all three on mobile, so a green site is a competitive advantage and an orange one is not a disgrace.
- A hundred points is not the goal. The score fluctuates, nobody notices the last five points, and some points are worth sacrificing to features that earn money.
Want to know where you stand
The home page has a field where you can enter your own address and run the measurement directly with Google. It takes half a minute and I store nothing. If the result says there is room for improvement, write to me and we will go through what to do and in what order.
Send an enquiry or see the price list.