PageSpeed Insights vs GTmetrix vs Lighthouse: which should you use?

Test the same page on PageSpeed Insights, GTmetrix, and Lighthouse and you'll often get three different scores, sometimes wildly different ones. That's not one of the tools being wrong. They test from different locations, on different simulated connections, and at least one of them mixes in real-world data the others don't have access to at all.

Once you know what each tool is actually doing, the disagreement stops being confusing and starts being useful — each one answers a slightly different question.

Lab data vs field data: the core distinction

This is the single most important concept for making sense of any speed tool. Lab data is a simulated test run once, under controlled conditions, on a specific device and connection speed profile. Field data is aggregated from real visitors actually loading your site, on their real devices and real connections, over a rolling time window.

Lab data is reproducible and great for debugging a specific change. Field data is messier but reflects what your actual visitors experience, which is also what Google uses for ranking purposes. A page can score well in a lab test and still perform poorly for real visitors on slow mobile connections, and the reverse can happen too.

What each tool actually gives you

PageSpeed Insights (PSI) is Google's own tool, and it's the only one of the three that combines both data types in one report: field data from the Chrome User Experience Report (CrUX) when your site has enough real traffic to qualify, plus a Lighthouse-powered lab test underneath. This dual view is exactly why PSI is the closest thing to 'what Google sees.'

GTmetrix runs a lab test too, but gives you far more control over test conditions: choice of test server location, simulated device and connection type, and a waterfall chart that's genuinely useful for diagnosing exactly which request is slow. It has no field data of its own — everything it reports is a single simulated run.

Lighthouse is the open-source auditing engine that powers the lab-test portion of both PSI and, indirectly, much of GTmetrix's scoring philosophy. You can run it directly from Chrome DevTools or the command line, with no server round-trip, which makes it the fastest option for iterating locally while you're actively making changes.

Why scores differ between tools, and even between runs

Beyond the lab-vs-field distinction, several other factors move the number even for identical pages. Test server location changes network latency to your host. Simulated connection speed (throttled 4G vs a faster profile) changes how long resources take to arrive. CPU throttling settings simulate a slower or faster device. And even the exact same tool, run twice in a row, can vary by a few points due to normal network jitter — this is expected, not a bug.

None of this means the scores are meaningless. It means a single number from a single tool, on a single run, is a snapshot, not a verdict. Trends across multiple runs and multiple tools matter more than any one score in isolation.

Which tool to trust for which purpose

1. 1
   Use PageSpeed Insights to know where you stand with Google

   If your site has enough traffic for CrUX field data, PSI's field report is the closest available proxy to your actual Core Web Vitals ranking signal. Check this first.
2. 2
   Use GTmetrix when you need to diagnose a specific slow request

   Its waterfall view and configurable test conditions make it easier to pinpoint exactly which image, script, or font is the bottleneck.
3. 3
   Use Lighthouse (via Chrome DevTools) while actively making changes

   It's the fastest feedback loop since it runs locally with no queue or server round-trip, ideal for a quick before/after check as you optimize.
4. 4
   Don't obsess over an exact numeric score from any single tool

   Look for consistent trends across tools and across your top pages, not a single perfect number on one run of one tool.

Where WebPageTest and browser DevTools fit in

Two other tools are worth knowing about, even though they're less commonly the first stop. WebPageTest offers even deeper configurability than GTmetrix — multiple test locations worldwide, real device testing rather than only simulated throttling, video comparison of two runs side by side, and a very detailed request-level breakdown. It rewards the extra complexity when you're chasing a specific, stubborn performance problem rather than doing a general check.

Your browser's own DevTools (the Network and Performance tabs, available in Chrome, Firefox, and Edge) skip the external tool entirely. You can throttle your own connection, record a full performance trace, and see exactly which resource is blocking rendering, without leaving the page you're debugging. It's less polished than a dedicated speed-testing product, but it's the fastest way to answer 'why is this specific thing slow, right now.'

A simple monthly routine that avoids over-testing

You don't need to run every tool on every page constantly. A practical cadence for a small site is: check PageSpeed Insights for your top few pages roughly once a month to track field-data trends over time, and reach for GTmetrix, WebPageTest, or Lighthouse specifically whenever you make a change you want to verify — a new hero image, an added script, a hosting switch. Testing more often than that mostly adds noise without adding insight, since day-to-day fluctuation is expected and not meaningful on its own.

• ✓Lab data is a controlled, repeatable simulation; field data is aggregated real-visitor experience.
• ✓PageSpeed Insights is the only one of the three with real field data built in, when your traffic qualifies.
• ✓GTmetrix offers the most configurable, detailed lab test for pinpointing specific slow requests.
• ✓Lighthouse is the shared underlying engine, and the fastest option for local iterative testing.
• ✓Score differences between tools and even between runs of the same tool are expected, not errors.
• ✓Trust trends over time and across tools more than any single score from any single run.
• ✓WebPageTest and browser DevTools are worth reaching for on stubborn, specific performance problems.