Blog / How the internet works /

Measuring website speed properly

You test your homepage. The score says 94. You test again an hour later: 71. Nothing changed on the site. Which number is true?

Both are, in a way. Speed isn’t a single number, and testing it properly means knowing what each tool is actually measuring.

Two kinds of data

Lab data

A tool loads your page in a controlled setup, with a fixed device, connection and location, and records what happens. PageSpeed Insights (the Lighthouse part), WebPageTest and GTmetrix work like this.

Lab data is repeatable, great for debugging and for comparing “before” and “after”. But it’s one pretend visitor.

Field data

Real visitors’ browsers report how pages loaded for them. Google collects this from Chrome users in the Chrome User Experience Report (CrUX), and it appears in PageSpeed Insights and Search Console’s Core Web Vitals report.

Field data shows what real people experienced, across many devices, networks and countries. But it’s slower to update (it’s an average over a rolling period) and only exists if enough people visit.

Use lab data to find and fix things. Use field data to know whether it mattered.

What to measure

The Core Web Vitals are the shared language:

  • LCP (Largest Contentful Paint): how long until the main content appears.
  • INP (Interaction to Next Paint): how quickly the page reacts when someone taps or clicks.
  • CLS (Cumulative Layout Shift): how much things jump around while loading.

I explained them in Core Web Vitals in plain English and covered the interaction metric in INP is now a Core Web Vital: what to fix first.

Add two simple extras:

  • TTFB (time to first byte): how fast your server starts responding. Slow here means hosting, caching or heavy code.
  • Total page weight and number of requests: a blunt but useful sanity check.

Why scores wobble

Even in a lab, results shift because of:

  • Server load at that moment.
  • Whether a page was cached.
  • Third-party scripts (ads, chat widgets, pixels) responding slowly.
  • The test location and simulated device.
  • Random variance in how the browser runs.

So run each test several times and use the middle result. Don’t panic over a five-point swing.

A sensible measuring routine

Here’s the method I’d recommend.

  1. Pick the pages that matter. A speed score for your homepage alone is misleading. Choose 4 to 6 pages: the homepage, a blog post, a product or service page, and checkout or the contact page.
  2. Measure on mobile first. Most visitors are on phones, often on slower networks. Desktop scores flatter you.
  3. Check field data first. In PageSpeed Insights, look at the real-user results at the top. Are the Core Web Vitals passing for most visitors? If yes, a poor lab score is less urgent.
  4. Run lab tests three times per page and note the median.
  5. Read the waterfall, not just the score. In WebPageTest or your browser’s Network tab you can see which file is slowest and what blocks the page from showing.
  6. Record results in a simple table with date, page, tool, and numbers.
  7. Change one thing, then re-test. If you change five things at once you won’t know what helped.

Test from where your visitors are

A test run from a data centre in Europe says little about readers on mobile data in another continent. WebPageTest lets you choose locations and connection types. If your audience is in one region, test from there. Website speed in the Gulf: hosting near your readers and hosting a website for African readers cover why location changes results.

Don’t forget real behaviour

Lab tests load a page cold. Real visitors:

  • Have cached files from earlier visits.
  • Scroll and tap, which is when INP issues show.
  • Move across multiple pages.
  • Are sometimes logged in, which skips page caching.

For WordPress or WooCommerce, test as a logged-out visitor and, separately, a logged-in one.

Common mistakes

  • Chasing 100. The last 10 points are expensive and often make no difference to real people.
  • Testing once. One run is an anecdote.
  • Ignoring third-party scripts. A chat widget can ruin a clean site. Third-party scripts: chat widgets, pixels and speed is worth reading if your scores are hurt by outside code.
  • Comparing tools directly. Different tools use different settings. Compare a tool with itself over time.
  • Testing right after a change without clearing caches or letting field data catch up.

A few quick wins to try first

Before you dig deep, look at the obvious: oversized images, too many fonts, heavy sliders and unnecessary plugins. Page speed on a budget: what to fix first has the order I’d try.

Key takeaways

  • Lab data is for diagnosing; field data is for confirming real experience.
  • Measure several key pages, on mobile, several times.
  • Focus on LCP, INP, CLS and TTFB, and read the waterfall.
  • Change one thing at a time and keep a record.
  • Don’t chase a perfect score.

FAQ

Why is my score different in Search Console?

Search Console uses field data from real visitors, grouped by page type. PageSpeed’s lab number is a single simulated test.

How often should I test?

After meaningful changes (new plugin, new theme, new scripts) and on a regular schedule, such as monthly.

Pick your three most important pages and test each three times today. Write the numbers down. In a month, you’ll thank yourself.