Two tools give you two scores after each one tests your page on a different device, over a different connection, from a different place. PageSpeed Insights runs on Google’s hardware and, on its mobile tab, pretends to be a mid-range phone. GTmetrix, out of the box, tests from Seattle on a desktop with no speed limit at all. Same page, different room, different number.
The score is five numbers blended, not a stopwatch
A performance score is not how long your page took. It is five separate measurements, each scored, then averaged with weights.
Chrome for Developers publishes those weights on its Lighthouse performance scoring page. For Lighthouse 10 they are:
- Total Blocking Time, 30%
- Largest Contentful Paint, 25%
- Cumulative Layout Shift, 25%
- First Contentful Paint, 10%
- Speed Index, 10%
That page shows a last updated date of September 19, 2019, which is older than the version of Lighthouse it documents. Go read the table yourself before you trust anyone quoting it, this article included.
Look at the top line. Total Blocking Time is worth 30 of the 100 points. That is the time your page sat frozen while scripts ran. A fast test machine chews through those scripts. A slow one does not. Two tools with different processors get different Total Blocking Time on the exact same page, and the score moves with it.
That same page also says that in general only those five metrics feed the Performance score, not the opportunity and diagnostic audits listed underneath them. Hold onto that. It is the reason the GTmetrix letter looks wrong later on.
The two tools are not running the same test
Google’s About PageSpeed Insights page says Lighthouse simulates a mid-tier device, a Moto G4, on a mobile network for the mobile test, and an emulated desktop on a wired connection for the desktop test. It also says PageSpeed runs in a Google datacenter that can vary, and reports its location as North America, Europe, or Asia. That page was last updated October 21, 2024.
GTmetrix ships different defaults. Its article Set Default Analysis Options for GTmetrix Tests lists them in three lines: test location Seattle, WA, USA; browser and device Chrome, Desktop; connection speed unthrottled. Published October 5, 2021, last updated May 6, 2026.
Read that last word again. Unthrottled. By default GTmetrix is not slowing the connection down at all, while PageSpeed Insights on mobile is acting like a mid-range Android phone on a mobile network. Those two setups were never going to agree.
GTmetrix says as much itself. Its report guide lists what it expects to differ between its Performance Score and Google’s: where the test ran, the processor and memory of the test machine, the network, and Google’s own extra adjustments to processor and network speed. It ends that part by saying the two scores are not directly comparable, though they should be similar.
The GTmetrix letter grade is not a Lighthouse score
This is where most of the confusion starts. The big letter on a GTmetrix report is not the Performance score.
GTmetrix says so itself. Its article Everything You Need to Know About GTmetrix Reports, dated November 16, 2020, says the GTmetrix Grade is a weighted average of two things: the Performance Score at 60% and the Structure Score at 40%. Structure is GTmetrix’s own reading of Lighthouse opportunity and diagnostic audits, the ones Google keeps out of its score.
So a B on GTmetrix and a 94 on PageSpeed Insights are not a contradiction. They are two different recipes. GTmetrix also warns in that same article that its weightings are continually monitored and tweaked and will evolve over time, so the grade you got last year and the grade you get today may not be built the same way.
The GTmetrix FAQ answers the comparison question head on. It says PageSpeed Insights carries Chrome User Experience Report data, collected from real Chrome users, while GTmetrix is a lab test run in a controlled setup. That page carries no update date. Whether Google shows you that real user data at all depends on your traffic, and that is a separate question with its own answer.
Why one tool gives you two answers ten minutes apart
Before you blame the other tool, check that either tool repeats itself. Chrome for Developers lists the usual causes on the same scoring page: A/B tests or changes in the ads being served, internet traffic routing changes, testing on different devices such as a high-performance desktop against a low-performance laptop, browser extensions that inject JavaScript, and antivirus software.
Google’s About PageSpeed Insights page adds three more: local network availability, client hardware availability, and client resource contention. In plain terms, your Wi-Fi, your machine, and whatever your machine was already doing.
One run is not a reading. Run the same test three times and use the middle result. Some of the gap people bring to this question closes on its own once they do that.
Which number to act on
Google puts the limit in writing on the PageSpeed Insights page: good lab data does not necessarily mean real user experiences will also be good. A lab score is a simulation of one device on one connection. Your customer is on a five-year-old phone in a truck with two bars.
The Chrome scoring page adds another warning worth reading before you spend money. A perfect 100 is extremely challenging to achieve and not expected. Taking a score from 99 to 100 needs about the same amount of metric improvement that would take a 90 to 94.
So do not pay anyone to close a gap between two tools that were never measuring the same thing, and a monthly retainer should list the work it did. The color bands on that same page show where the line sits: 0 to 49 is red, 50 to 89 is orange, 90 to 100 is green. Once your PageSpeed Insights mobile test is green, the score has told you most of what it knows. Anything still wrong after that will not show up in the number.
What to do next
Five steps, in order.
- Pick one tool and stay with it. Comparing a GTmetrix letter to a PageSpeed number is comparing two different measurements.
- Run the same test three times with the same settings, and use the middle result. Throw out the odd one.
- In GTmetrix, change the defaults. Set the test location nearest your service area, pick a mobile device instead of desktop, and turn on a connection speed limit. Now it is testing something closer to your actual customer.
- On PageSpeed Insights, read the mobile tab first if people find you on a phone. The desktop tab flatters almost every site.
- If both tests are green and the page still feels slow, stop tuning the score and go look at what loads after the page appears, and at how fast the page answers a tap.
If you want the fix list ranked by what actually moves the number, start with what makes a page load faster, in order. A site built lean from the first day avoids most of this argument, and the Galloping website build is the one piece of speed proof this company owns: its own pages scored 100 on Google’s PageSpeed test, on desktop and on mobile, measured on the day it launched.
Speed is only one input into whether a customer finds you. If you also want to see which companies AI assistants name when someone asks for your trade in your town, the free scan asks them and writes down the answers.