Client resources
Why your speed score keeps changing.
You test your website and get a 64. Ten minutes later it says 81. Another tool gives you an A, and a competitor scores 95. Here is what those numbers mean, why they move around, and what actually affects your visitors and your search results.
Start here
Five things worth knowing first.
If you only read one part of this page, read this part.
01
One test is a snapshot
The same page can score 20 points apart on two runs a few minutes apart. Run a test three to five times and go by the middle result.
02
Phones score lower, on purpose
The phone test pretends to be a mid-range phone on a slow mobile connection. Desktop scores will almost always be higher.
03
Google ranks on real visits
Search uses data from real visitors, collected over about four weeks. Not the number in the circle.
04
Some features cost points
Animation, video, chat, booking calendars and cookie banners all add weight. That can be a perfectly good trade.
05
Watch the time, not the grade
What matters is how quickly real people see your page and can use it.
Lab tests and real visitors
Two different kinds of measurement.
Most speed tools load your page once, in a test browser on a computer in a data center, and time it. That is a lab test: one visit, from one place, on one machine, at one moment.
The other kind is real visitor data (Google calls it field data). Chrome browsers that have opted in report how fast your pages actually loaded for real people, on their own phones and connections. Google keeps a rolling 28 days of it, and that is what it looks at for page experience in search.
Reading PageSpeed Insights
- “Discover what your real users are experiencing” is the real visitor data. On a new or quiet site it says “No Data”. That is normal.
- “Diagnose performance issues” is the lab test. The big score circle lives here.
Why it moves
Why the score swings.
None of these mean anything is wrong with your site.
01
The test machine is busy
Free tools run many tests at once on shared machines. If yours runs while the machine is under load, the page looks slower than it is. Total Blocking Time suffers most, and it carries a lot of weight in the score.
02
The phone score is an estimate
For the phone test, PageSpeed Insights records one load and then estimates how it would go on a mid-range phone over slow 4G. A small change in the order files arrive can move that estimate by seconds.
03
Caching
Fast hosting keeps a ready-made copy of your pages close to visitors. The first visit after a change, or the first from a new region, can be slower while that copy is rebuilt.
04
Outside services
Cookie banners, chat widgets, maps, booking calendars, video players and analytics tags come from other companies’ servers, and their speed changes from minute to minute.
05
Location
A test run from Seattle and one run from Virginia take different routes to your site, and get different times.
06
Different tools, different scales
Each tool grades its own way. An A in one and a 79 in another can describe the same healthy page.
Core Web Vitals
The three numbers Google watches.
Google calls them Core Web Vitals. “Good” means at least 75% of real visits hit the target.
LCP · good: 2.5 s or less
How fast the main content appears
Largest Contentful Paint. Usually the big headline or the main image at the top of the page.
INP · good: 200 ms or less
How fast the page responds
Interaction to Next Paint: the delay between a tap or click and the page reacting to it.
CLS · good: 0.1 or less
How much things jump around
Cumulative Layout Shift. Text that moves just as you go to tap it is the classic example.
Why phones score lower. The phone test assumes a mid-range phone with a slower processor on a weak mobile connection: your slowest likely visitor, not your average one. A page that loads in about a second on a laptop can be estimated at four or five seconds under those conditions. On a site with real design in it, a phone score in the 70s or 80s is a good result.
Two lab numbers worth a glance: First Contentful Paint (when anything first appears) and Total Blocking Time (how long the page keeps the processor too busy to respond).
Which tools to use
The tools, and how far to trust them.
Compare a tool with itself over time, never one tool against another.
Start here
Google PageSpeed Insights
The standard check: real visitor data (once there is enough) and a lab test, with specific suggestions. Single runs vary a lot, so run it three to five times and use the middle score.
Best long-term view
Search Console: Core Web Vitals
Real visitor data across your whole site, grouped into Good, Needs Improvement and Poor. It needs traffic and about four weeks of data, so it won’t help on day one.
Visual report
GTmetrix
Loads your page in a real Chrome browser from a test location you choose, with a waterfall of every file. Results depend on that location, and its grades are not on Google’s scale.
Deep detail
WebPageTest
Many locations, devices and repeat views, for serious troubleshooting. More than most people need, and best in an expert’s hands.
Grain of salt
Pingdom
Fine for uptime monitoring, its main job. Its speed grade uses older rules, and some security services block its test robot, so it can end up grading the “are you a robot?” page instead of your site.
While building
Lighthouse in your own Chrome
Quick checks during a build. Your extensions, computer and connection all affect it, so use an incognito window and don’t compare it with anyone else’s numbers.
Testing
How to test your site properly.
Five habits that turn a jumpy number into something you can act on.
- Test the same page each time. Your homepage and one important inner page make a good pair.
- Use PageSpeed Insights and run it three to five times. Note the middle score, not the best or the worst.
- Look at both phone and desktop, and compare phone with phone and desktop with desktop.
- When something changes (a new section, plugin or widget), test again the same way so you compare like with like.
- After launch, check Search Console’s Core Web Vitals report every month or so. That is the scoreboard Google actually uses.
Trade-offs
Why a lower score can be fine.
Many of the things that make a site feel polished and personal also add weight:
- Animation and movement, like fades and slides as you scroll.
- Video, especially autoplaying video near the top.
- Chat, booking calendars, maps and social feeds, each loading its own code from another company.
- Cookie consent banners, often legally necessary, and loaded before almost anything else.
- Analytics and advertising tags.
The compromise that works
Keep the first screen simple and quick, and save the movement for further down the page. The first screen is what speed tests weigh most, and by the time a visitor scrolls, the rest has finished loading.
The goal is the best balance between how your site looks and works and how quickly it shows up. A site with thoughtful animation that scores 80 on a phone usually serves a business better than a plain page that scores 99.
On the sites we look after
What we do about speed.
Most of this comes from BrightFrame, the toolkit on the sites we host, plus some care in how each page is built. See hosting and ongoing care.
- Managed hosting with edge caching, so ready-made copies of your pages are served close to your visitors.
- Fonts served from your own site, with stand-in fonts sized to match, so text doesn’t jump while the real font loads.
- The main background at the top of each page fetched first on desktop, so the most important thing on screen appears sooner.
- Images sized for where they are shown, so phones don’t download desktop-sized pictures.
- Cookie banners that don’t hold up the page, where the site’s privacy setup allows it.
- Nothing on the first screen waiting for an animation, and animations that phones never show are never downloaded by phones.
- Speed checks of each site’s most-visited pages, so a heavy image or a new widget gets caught.
A real example
Our own site, before and after.
While building this site we ran the same lab test, five runs each, before and after the speed work (September 30 to October 1, 2026, on our staging site).
Homepage, phone
66 → 79
The middle of five lab runs, before and after.
Homepage, desktop
82 → 95
Same test, same five-run method.
Homepage weight
3.1 MB → 1.1 MB
What a phone downloads to show the page.
Over the same afternoon, single PageSpeed Insights runs of the finished page on a phone ranged from 52 to 79, with no changes to the site at all. Same page, same hour. That is why we go by the middle of several runs, and put more weight on real visitor data than on any one score.
One more thing we learned: making the page faster briefly made the text jump as the fonts arrived, because the text now showed up before they did. Serving the fonts from the site itself, with stand-in fonts sized to match, took that back to zero.
Questions
Common questions.
The ones clients ask us most, once they have run a test.
My competitor’s site scores 95. Why doesn’t mine?
Compare what is on the page. A simple, mostly-text page with no animation, video, chat or booking tools will usually outscore a richer one. Check that you are comparing the same tool, the same device type and several runs. What matters is whether your site does its job for your visitors.
My score dropped and we didn’t change anything. Is something broken?
Probably not. Run the test a few more times. If the middle score is still clearly lower, look for recent additions: a new plugin, a new tracking tag, a large image, or a change to an outside service. That is when it is worth asking us to take a look.
Should we remove the animations to get a better score?
Usually not. Keeping the very top of the page still gets most of the benefit while keeping the movement everywhere else. We can show you the difference on your own site.
Will a perfect score get us to the top of Google?
No. Speed helps, mostly by keeping visitors from leaving, but rankings depend far more on clear, useful content that answers what people are searching for.
Want to know how your site measures up?
Ask for a free check-up. We’ll test your key pages properly, explain what the numbers mean for your site, and tell you which fixes are worth making and which aren’t.
