How to Test Website Speed: Reading PageSpeed Insights, GTmetrix and WebPageTest Scores

by | Nov 19, 2026 | Uncategorized | 0 comments

Almost everyone knows how to test website speed in the literal sense: paste a URL into a tool, hit the button, wait. The hard part comes next, when the screen fills with scores, grades, coloured bars and acronyms, and you have no idea which number deserves your attention and which one is just noise.

This guide walks through running a real test on the three tools that matter most, Google PageSpeed Insights, GTmetrix and WebPageTest, and then decodes the results in plain language. The goal is not a green 100. The goal is knowing which two or three fixes will actually make your pages feel faster for real visitors.

Before You Test: Three Rules That Save You Hours

Most confusing speed test results come from bad test conditions, not from a bad website. Set these three rules before you run anything.

  1. Test the same URL three times, keep the median. A single run can be skewed by a cold cache, a slow third party or a busy server. One run is an anecdote, three runs is data.
  2. Test the pages that matter, not just the homepage. Your homepage is often your lightest page. Test a product page, a long blog article and a landing page running ad traffic.
  3. Decide whether you are measuring lab or field data. These are two different things and mixing them up is the single biggest source of confusion. More on this below.

Lab Data vs Field Data in One Minute

Type What it is Where you see it Use it for
Lab data A simulated visit from one machine, one connection, one moment in time Lighthouse score in PageSpeed Insights, GTmetrix Performance tab, WebPageTest runs Debugging, finding the cause, comparing before and after a fix
Field data Anonymised measurements from real Chrome users over the last 28 days The “Discover what your real users are experiencing” block at the top of PageSpeed Insights, Search Console Core Web Vitals report Judging whether you have a real problem, and whether Google sees it

The short version: field data tells you if you have a problem. Lab data tells you why. Never panic over a lab score when your field data is green, and never celebrate a lab score when your field data is red.

website speed test

Tool 1: Google PageSpeed Insights

PageSpeed Insights (PSI) is the tool most people start with, because it is free, it is Google’s own, and its metrics map directly to what Google uses as a ranking signal.

How to Run the Test

  1. Go to pagespeed.web.dev.
  2. Paste the full URL including https:// and click Analyse.
  3. When the report loads, look at the Mobile tab first. Mobile is the default for Google’s indexing and it is almost always the weaker of the two.
  4. Scroll past the big coloured circle. Yes, really.

Reading the Field Data Block (The Part That Counts)

If your URL gets enough Chrome traffic, the top of the report shows real user data from the Chrome User Experience Report. You will see four or five metrics with a coloured bar showing the distribution of your visitors.

The number displayed is the 75th percentile, meaning 75 percent of your visitors had an experience at least that good. Google grades it like this:

Metric Plain language Good Needs work Poor
LCP (Largest Contentful Paint) How long until the biggest thing on screen, usually a hero image or headline, is visible Under 2.5 s 2.5 s to 4 s Over 4 s
INP (Interaction to Next Paint) How long the page takes to visually respond after a tap or click Under 200 ms 200 ms to 500 ms Over 500 ms
CLS (Cumulative Layout Shift) How much the content jumps around while loading Under 0.1 0.1 to 0.25 Over 0.25
TTFB (Time to First Byte) How long your server takes to start answering Under 800 ms 800 ms to 1.8 s Over 1.8 s
FCP (First Contentful Paint) When the first pixel of content appears Under 1.8 s 1.8 s to 3 s Over 3 s

LCP, INP and CLS are the Core Web Vitals. TTFB and FCP are diagnostics: they explain why LCP is slow but they are not graded by Google as vitals themselves.

One detail nearly everyone misses: PSI shows both this URL and origin data when it has enough samples. If your specific page has no field data, you will only see the origin summary, which is an average of your whole domain. Do not treat an origin-level result as proof about one page.

Reading the Lighthouse Performance Score (0 to 100)

The big circle is a simulated lab test on a throttled mid-range mobile device and a slow 4G connection. It is deliberately pessimistic. It is also a weighted average of five lab metrics:

Lab metric Weight in the score What it reflects
Total Blocking Time (TBT) 30% JavaScript hogging the main thread, the lab stand-in for INP
Largest Contentful Paint 25% Main content render speed
Cumulative Layout Shift 25% Visual stability
First Contentful Paint 10% First sign of life
Speed Index 10% How quickly the viewport fills visually

Two consequences follow from that table:

  • A score of 48 with a green LCP and green CLS usually means one thing: too much JavaScript. TBT alone can drag you down 30 points.
  • Scores bounce. A 10 point swing between two identical runs is normal, because the simulation adds variance. Never report a single Lighthouse score as a KPI.

Opportunities and Diagnostics: What to Act On

Under the score you get a list of suggestions with estimated savings in seconds. Treat these as hints, not as a to-do list in order. The estimated savings are theoretical maximums that assume everything else stays the same, which it never does.

The item that actually matters is further down, in the diagnostics: “Largest Contentful Paint element”. Expand it. PSI breaks LCP into four phases:

  • TTFB: server thinking time. If this is more than about 40 percent of LCP, your problem is hosting, database queries or missing page caching, not images.
  • Resource load delay: the gap between the HTML arriving and the browser starting to fetch the LCP image. Caused by lazy-loading the hero image, images injected by JavaScript, or slow CSS discovery.
  • Resource load duration: the download itself. This is where image size, format and CDN matter.
  • Element render delay: the image is downloaded but nothing shows yet, usually blocked by fonts, CSS or scripts.

That single breakdown tells you more than the whole Opportunities list, because it tells you which department owns the problem.

website speed test

Tool 2: GTmetrix

GTmetrix is the best tool for a quick, readable health check and for showing non-technical stakeholders what is going on. It runs on real Chrome instances in real locations.

How to Run the Test

  1. Go to gtmetrix.com and create a free account. Without logging in you are stuck with a default test location and desktop settings that will not match your audience.
  2. Click the test options and set the test server location to the country closest to your real visitors.
  3. Set the connection and, if your traffic is mostly mobile, the device to a mobile profile.
  4. Run the test, then run it twice more and compare.

Decoding the GTmetrix Grade

GTmetrix shows three things at the top, and people constantly confuse them:

  • GTmetrix Grade (A to F): a blend of the Performance score and the Structure score. It is a summary, not a diagnosis.
  • Performance score: essentially the Lighthouse performance score described above, with the same weightings. Same caveats, same variance.
  • Structure score: GTmetrix’s own audit of best practices. This is the interesting one, because a page can load fast today and still have a weak structure that will break as soon as the page gets heavier.

Below that you get three raw numbers that are often the most useful line on the whole page:

Number What it tells you Rough target
Largest Contentful Paint Perceived load speed in that location Under 2.5 s
Total Page Size Total weight of everything downloaded Under 1.5 MB for a content page is comfortable
Total Requests Number of files the browser had to fetch Under 70 to 80 on most sites

The GTmetrix Tabs Worth Opening

Structure

Each audit has an Impact rating. Sort by impact and ignore everything marked low. “Serve images in next-gen formats” with high impact on a page with a 900 KB hero JPEG is worth your afternoon. “Avoid an excessive DOM size” on a page with 1,200 nodes is not.

Waterfall

This is the heart of GTmetrix. See the waterfall section below for how to read it.

Video and Filmstrip

Enable video capture in the test settings. Watching a 4x slow motion replay of your own page loading does more to convince a client, a boss or yourself than any score. You will see exactly when the text appears, when the image pops in and when the layout jumps.

Tool 3: WebPageTest

WebPageTest is the deepest of the three. It is where you go when PSI says “something is slow” and you need to know which exact request, on which exact connection, from which exact city.

How to Run a Useful Test

  1. Go to webpagetest.org and choose the Site Performance test type.
  2. Pick a test location that matches your audience and a browser, normally Chrome.
  3. Choose a connection profile. 4G or Mobile LTE is a fair real-world baseline. Cable is flattering and mostly useless for decision making.
  4. Under advanced settings, set the number of tests to 3 or more, so you get a median run automatically.
  5. Run the test and open the median run, not the fastest one.

Reading the WebPageTest Summary

The top of the report gives letter grades for things like time to first byte, keep-alive, compression, caching and CDN usage. These are structural hygiene checks. An F on compression or caching is a cheap fix with a real payoff, so start there.

Then look at the first view vs repeat view columns. First view is a brand new visitor with an empty cache. Repeat view is someone coming back. If repeat view is barely faster than first view, your caching headers are wrong and you are making returning visitors re-download everything.

The Filmstrip

WebPageTest’s filmstrip shows screenshots at fixed intervals with the waterfall aligned underneath. This is the best explanation tool in existence: you can point at the exact frame where the page is still blank and then point at the request directly below it that is responsible. Set the interval to 0.1 s for detailed analysis.

website speed test

How to Read a Waterfall Chart in Plain Language

Both GTmetrix and WebPageTest give you a waterfall: one horizontal bar per file, stacked in the order the browser requested them. Here is what the coloured segments inside each bar mean.

Segment Plain language If it is long
DNS lookup Translating a domain name into an IP address Too many separate third party domains, consider preconnect
Initial connection Opening the TCP connection Server is geographically far away, a CDN would help
SSL / TLS Securing the connection Old TLS config or an extra hop, check HTTP/2 or HTTP/3 support
Waiting / TTFB Your server building the page Backend problem: caching, plugins, database, hosting plan
Content download Transferring the file File is too big: compress, resize, convert to WebP or AVIF

The Four Waterfall Patterns Worth Recognising

  1. A long first bar. The very first request is the HTML document. If its waiting segment is huge, nothing else matters until you fix the server. Everything downstream is just waiting.
  2. A staircase. Requests starting one after another in a descending diagonal instead of in parallel means a dependency chain: a script loads a script that loads a stylesheet. Each step costs a full round trip.
  3. A wall of bars starting at the same moment. That is fine, that is parallelism working. Bandwidth contention can still slow the important file down, so check whether your LCP image is competing with 30 other files.
  4. Long bars after the page already looks finished. Analytics, chat widgets, heat maps, pixels. They do not usually hurt LCP, but they are prime suspects for poor INP because they occupy the main thread.

Also look for the vertical lines drawn across the waterfall. They mark Start Render, First Contentful Paint, LCP and onLoad. Anything to the right of the LCP line is, by definition, not delaying your main content.

website speed test

Which Metrics to Act On First

Here is the priority order we use when auditing a site. Work top to bottom and stop when the field data turns green.

Priority Signal Typical fix
1 TTFB over 800 ms in field data Full page caching, better hosting, CDN, fewer blocking plugins
2 LCP over 2.5 s in field data Preload the hero image, never lazy-load it, serve modern formats, correct dimensions
3 CLS over 0.1 Width and height on images, reserved space for ads and banners, font-display settings
4 INP over 200 ms, or TBT over 300 ms in lab Defer non-critical JavaScript, remove duplicate tag manager scripts, break up long tasks
5 Page weight over 2 MB Image compression, font subsetting, removing unused CSS and plugins
6 Everything else Only after the five above are green

Which Numbers You Can Safely Ignore

  • The last 10 points of a Lighthouse score. Going from 90 to 100 is usually invisible to humans and expensive in engineering hours.
  • “Reduce unused JavaScript” on third party scripts you cannot edit. If it is your consent manager or your payment provider, you can defer it, you cannot rewrite it.
  • Fully Loaded Time. It keeps counting until network activity stops, so a chat widget polling every few seconds inflates it forever. It does not reflect what a user feels.
  • Low impact Structure audits in GTmetrix. They are nice to know, not worth a sprint.
  • Any lab test run on a page that is behind a cache warm-up. The first visit after a deploy or a cache purge is always the worst. Run it again.
  • “Eliminate render-blocking resources” when the saving is under 150 ms. The cure is often riskier than the disease.

A Repeatable Testing Workflow

Put these steps in your calendar once a month and you will never be surprised by a Core Web Vitals warning in Search Console again.

  1. Start in Search Console. Open the Core Web Vitals report to see which URL groups are failing for real users. This is your source of truth.
  2. Pick one representative URL per failing group.
  3. Run PageSpeed Insights on that URL, mobile tab, and read the field data plus the LCP phase breakdown.
  4. Run GTmetrix from your audience’s location to get page weight, request count and a visual filmstrip.
  5. Run WebPageTest with 3 runs on a 4G profile and open the median waterfall to identify the exact offending requests.
  6. Fix one thing. Just one. Then re-test all three tools and record the before and after.
  7. Wait 28 days for the field data to catch up before declaring victory, because CrUX uses a rolling 28 day window.
website speed test

Common Misreadings We See Every Week

  • “My score dropped from 92 to 74 overnight, we are penalised.” Lab variance plus a slightly heavier ad call. Check field data before escalating.
  • “GTmetrix says A but PageSpeed says 54.” Different device, different connection, different location. Compare like with like or do not compare at all.
  • “We hit 100 so speed is done.” A 100 on a stripped landing page tells you nothing about your checkout, which is where the money is.
  • “The tool says save 3.2 s on images.” Those savings do not add up linearly. Fixing four 1 second opportunities rarely saves 4 seconds.
  • “Desktop is green so we are fine.” Google evaluates mobile and desktop separately, and most traffic is mobile.

Need a Second Pair of Eyes?

Running the tests is the easy part. Deciding which three of forty suggestions actually move the needle on your specific stack is the part that takes experience. If your Core Web Vitals are red and you want a prioritised, plain-language action plan rather than a screenshot of a red circle, the team at custom-web-promotions.com can audit your key pages and tell you exactly what to fix first.

FAQ: Testing Website Speed

What is a good website speed score?

For lab scores, 90 or above is excellent, 50 to 89 is acceptable and under 50 needs work. But the score that really matters is your field data: LCP under 2.5 seconds, INP under 200 milliseconds and CLS under 0.1 at the 75th percentile. A site with a lab score of 68 and all three vitals green is in better shape than one with a 95 and a failing INP.

Why do PageSpeed Insights and GTmetrix give different results?

They test under different conditions. PageSpeed Insights simulates a throttled mid-range mobile device on a slow connection from Google’s infrastructure. GTmetrix uses a real browser, in the location and on the connection profile you chose. Different hardware, different network and a different moment in time will always produce different numbers. Use each tool against itself over time rather than against the others.

How often should I test my website speed?

Monthly for a stable site, and always immediately after a theme update, a plugin addition, a new tracking script or a redesign. Those four events cause the large majority of sudden speed regressions. Setting up GTmetrix or WebPageTest monitoring on three or four key URLs automates this.

Does website speed really affect SEO rankings?

Yes, but as a tie-breaker rather than a main driver. Core Web Vitals are part of Google’s page experience signals, so between two pages of comparable relevance and authority, the faster one has the edge. The bigger business impact is usually on conversion: slow pages lose visitors before they ever see your offer.

Can I test website speed for free?

Yes. PageSpeed Insights is completely free, GTmetrix offers free tests with a free account, and WebPageTest allows a number of free tests per month. Chrome’s built-in Lighthouse panel in DevTools is also free and runs locally, which is handy for testing staging sites or pages behind a login.

Should I test the mobile or desktop version?

Test both, but prioritise mobile. Google indexes mobile-first, mobile usually accounts for the majority of traffic, and mobile devices have weaker processors, which exposes JavaScript problems that desktop testing hides completely.

What is the difference between LCP and fully loaded time?

LCP measures when the main content becomes visible, which is what a user perceives as “the page has loaded”. Fully loaded time measures when all network activity stops, including analytics beacons, chat widgets and third party pixels that run long after the user is already reading. LCP is the metric to optimise, fully loaded time is mostly vanity.

Search Keywords

Recent News

Subscribe Now!