Website speed test: same page, five runs, a 12-point difference

Sebastian Rantala

The speed test score appears to be accurate: a single figure between 0 and 100. We ran the same page five times in a row using the same settings and obtained 78, 80, 90, 90 and 85. No changes were made to the site in between. This article explains what the test actually measures, which figures are reliable, and what the difference is between laboratory measurements and real visitors.

First of all: you might be looking for the wrong test

When searched for in Finnish, the term speed test refers to two completely different things, and the search results are a mix of both. We measured the top result on 14 September 2026: of the nine results, four measure the speed of your internet connection, not your website.

 Connection speed testWebsite speed test
Measurethe speed of your own connectionthe site is loading
Resultmegabits per secondseconds and score
Will change whenswitch mobile network provideryou’re updating the website

Jos tulokseksi tulee “245 Mbit/s”, olet väärässä testissä. Sivuston nopeutta mittaavat esimerkiksi Googlen PageSpeed Insights, GTmetrix ja WebPageTest.

The score won’t be repeated — we’ve measured it

This is the most important point of the article, and it’s hardly mentioned anywhere. We ran our own homepage through the tool five times in a row, using the same tool, the same settings and the same computer. No changes were made to the site between runs.

DrivingPointsMain content elementSpeed IndexBlocking period
1783,01 s8,17 s0 ms
2803,21 s5,41 s0 ms
3902,16 s3,40 s0 ms
4902,69 s3,21 s0 ms
5852,77 s4,55 s0 ms

The difference between the best and worst scores is 12 points. The Speed Index fluctuated between 3.21 and 8.17 seconds — more than double the difference for the same page.

The practical consequence is straightforward: if you make a change and the score goes up by five points, you don’t know whether the change helped. It could just be down to random noise. And if it goes down by five points, you won’t know that either.

Why do the figures fluctuate?

The test does not measure the site in a vacuum. It loads the page on a simulated mobile phone using a simulated slow connection and measures what happens. Each run is a separate event, and is influenced by:

  • Load on the measuring machine. The same computer is doing other things at the moment.
  • Current status of the server. Cache: warm or cold.
  • Network fluctuations between the meter and the server.
  • Animaatiot ja myöhään latautuvat osat. Speed Index katsoo, milloin ruutu näyttää valmiilta — jatkuva animaatio voi pitää sen “keskeneräisenä”.

Note from the table that the blocking time was zero on every occasion. So the figures do not fluctuate — and that is useful information.

Which figures are reliable?

ChapterWhat does this tell us?How stable
Blocking period (TBT)How long does it take for the page to respond to touch?The most stable — we have 0 ms on every run
Content Layout Shift (CLS)Does the content jump whilst loading?fairly stable
Largest Contentful Paint (LCP)When will the main content be displayed?varies, but the order of magnitude is roughly the same
Speed IndexWhen does the screen look finished?the most unstable
Pointsa combination of everythingcovers all variations

This leads to a practical tip: focus on the individual figures, not the total score. The total score is a summary that condenses all the noise from the individual sections into a single number.

Here’s how to test so that the results actually mean something

  1. Run the test at least three times and take the middle result. A single run does not constitute a measurement.
  2. Always test the same URL. The homepage and the article page are different pages, and their figures are not comparable.
  3. Compare mobile to mobile. The mobile test uses a slow connection and a low-spec device, whereas the desktop test does not. The difference is tens of points.
  4. Make one change at a time and measure the results separately. Otherwise, you won’t know what made the difference.
  5. Finally, take a look at the actual visitors. The Search Console Core Web Vitals report is based on real-world page loads. It’s slower but more accurate.

A test site or real visitors?

These two measurement methods address different questions, and therefore yield different results.

 Laboratory testReal visitors
Answer the questionWhat’s wrong with the site?Do visitors find the site slow?
Instant resultsYesNo, it takes weeks
Reproducibilitypoor, as measured aboveGood, because the selection is extensive
Use whenyou’re fixing somethingPlease assess whether there is a problem

The best approach is as follows: first, check the figures for genuine visitors and use them to determine whether there is a problem. Only then should you run a lab test to see what is causing it. If you do it the other way round, you often end up fixing things that weren’t causing anyone any trouble.

Frequently asked questions

Does the score have to be 100?

No. Google does not use a score system for rankings — it uses measured Core Web Vitals values from real visitors. Striving for a score of 100 is often work that goes unnoticed by everyone. The point of return on investment is reached once the measured values for visitors are in order.

Why did I get a different result to yesterday?

You’ve probably not changed a thing — the test is just playing up. We measured a 12-point difference in five consecutive runs on the same page. If the difference is of this magnitude, it’s just noise. Only a significantly larger change would indicate something meaningful.

Does speed affect search engine rankings?

It does have an impact, but less than is generally thought, and only when the site is noticeably slow. The relevance of the content is the primary factor. Speed is a distinguishing factor mainly when two pages are otherwise equally good — and even then, it is the metrics measured by visitors that matter, not the test score.

What slows down a WordPress site the most?

In order of priority: images that are too large, third-party scripts (analytics, chat, embeds, fonts) and the number of plugins. The first two can usually be fixed quickly. The third issue is structural and cannot be resolved with a caching plugin.

A measurement that tells you what needs fixing

During the audit, the site is run through several times to distinguish noise from genuine faults. The result is a list of issues to be rectified, prioritised by importance — not a single score.

Sources

  • Reproducibility test: Lighthouse, mobile profile and slow 4G, five consecutive runs from the same address on 14 September 2026. Scores: 78, 80, 90, 90, 85. Speed Index: 3.21–8.17 s, largest content element: 2.16–3.21 s, blocking time: 0 ms in all runs. The figures for each test and the settings used have been saved, and the measurement can be repeated using the same settings.
  • Top results analysed on 14 September 2026 (DataForSEO, Finland, desktop, site speed test search): of the nine organic results, four measure internet connection speed, not the site itself.
  • Core Web Vitals from real visitors: the Core Web Vitals report in Google Search Console.
  • Last checked on 14 September 2026.