PageSpeed Insights vs Lighthouse vs GTmetrix vs WebPageTest: which audit tool to use for client sites

Short answer: Use PageSpeed Insights first, because it is the only one of the four that shows field data from real Chrome users (when the site has enough traffic) next to a Lighthouse lab run. Use Lighthouse in Chrome DevTools while you fix things, WebPageTest when you need to see why a page is slow (waterfall, filmstrip, real locations and browsers), and GTmetrix when you want a tidy, shareable report or scheduled monitoring for a client.

Three of the four run Lighthouse under the hood, so the real differences are where the test runs, on what simulated device, and whether you see real-user data or a single lab load. This guide covers what each one measures, what the free tiers allow as of November 2026, and the order we use them in when auditing a small business site for a client.

What is the difference between field data and lab data?

Field data is collected from real visitors over time. Lab data is one synthetic page load on a simulated device and network. You need both, but they answer different questions: field data tells you what users actually experienced, lab data tells you why and lets you test a fix in minutes.

• CrUX (Chrome User Experience Report) is Google's public dataset of performance measurements from real Chrome users. PageSpeed Insights reads it and reports the 75th percentile over the previous 28-day collection period.
• Core Web Vitals are the three metrics Google treats as the main user experience signals: LCP (loading), INP (responsiveness) and CLS (visual stability).

The catch: a plumber's site with a few hundred visits a month often has no CrUX data at all. Google's PSI docs say that when a URL doesn't have enough data, PSI falls back to origin-level data for the whole site, and if the origin doesn't have enough either, it can't show real-user data. On most small sites you'll work from lab data only. That's fine, as long as you don't present a lab score as "what your customers experience."

INP is best measured in the field, because it depends on real taps and clicks. A standard Lighthouse page load doesn't interact with the page, so its performance score uses Total Blocking Time (TBT) instead. web.dev calls TBT "a reasonable proxy metric for INP" but not a substitute.

What does PageSpeed Insights actually measure?

PageSpeed Insights (PSI) shows two separate things on one page: CrUX field data at the top (if available) and a Lighthouse lab report underneath. The score in the circle is the Lighthouse lab score, not a field measurement.

Details from Google's own documentation, checked November 2026:

• Field data covers the previous 28 days and is reported at the 75th percentile. It updates daily in PSI.
• The lab run happens in a Google data center, in North America, Europe or Asia. You don't pick the location.
• The mobile lab test emulates a mid-tier phone on a throttled mobile network. Desktop uses a wired connection.
• Lab scores of 90 and up are "good," 50 to 89 "needs improvement," under 50 "poor."
• There is a free PSI API. Google says it works with or without an API key, and recommends a key for frequent automated queries.

PSI is the one we send clients to: it's Google's own tool and needs no sign-up.

What it doesn't give you: a network waterfall, a choice of location, video of the load, or any history. Each run is a fresh snapshot.

Is Lighthouse the same thing as PageSpeed Insights?

No. Lighthouse is the open-source auditing engine; PSI is a web service that runs Lighthouse on Google's servers and adds CrUX data. You can run Lighthouse yourself in Chrome DevTools (the Lighthouse panel), from the command line, or in CI.

The performance score is a weighted average of five lab metrics. As of Lighthouse 10, per Chrome's scoring docs:

So TBT, which is mostly JavaScript work on the main thread, carries the most weight. On small business sites that usually means page builder scripts, chat widgets and tracking tags.

Lighthouse's default mobile throttling, from its throttling docs, is 150 ms latency, 1.6 Mbps down and 750 Kbps up, with a 4x CPU slowdown. Locally, that slowdown is applied on top of your own machine, which is why a run on a fast laptop and a run on an old office PC can disagree.

Running it locally lets you test staging URLs that PSI can't reach:

# Mobile run, HTML report, opens in the browser when done
npx lighthouse https://staging.example.com --view

# Desktop preset, JSON output for comparing before/after
npx lighthouse https://staging.example.com --preset=desktop \
  --output=json --output-path=./before.json

Run it in an incognito window or a clean Chrome profile. Chrome's docs list browser extensions that inject JavaScript as one of the common causes of score swings, along with ad rotation, A/B tests and network routing. Run it three times and take the median rather than trusting one number.

What does GTmetrix add on top of Lighthouse?

GTmetrix runs Lighthouse too ("Lighthouse-powered reports" is their wording), then wraps it in a friendlier report, saved history, scheduled monitoring and a choice of test location and connection. It's the one that looks best when you attach it to an email to a business owner.

The free side has narrowed over the years. In GTmetrix's plan-change announcement (changes planned for December 1, 2023), guest testing ended, so you need a validated account to generate reports, and Basic (free) accounts got 5 on-demand tests that don't renew plus 5 daily API credits, with one week of data retention. We couldn't find the Basic limits restated on the current pricing pages, so treat those numbers as the last published ones and check your own account.

Mobile device testing and premium locations sit on the paid PRO plans. As of November 2026, GTmetrix's plans page lists paid plans starting at $5.99 a month billed monthly (Lite), with cheaper per-month pricing on yearly billing.

Where it earns its keep: weekly monitoring that catches the day someone uploads a 6 MB hero photo. Where it misleads: the example analysis options on the homepage show a desktop Chrome test on an unthrottled connection. A good score under those conditions says little about a phone on cellular, so check what settings a report used before quoting it.

When should you use WebPageTest instead?

Use WebPageTest when you need to see why a page is slow, not just how slow. It gives you a request-by-request waterfall, a filmstrip and video of the load, repeat-view tests, a choice of real browsers and connection speeds, and test locations around the world.

WebPageTest is run by Catchpoint, which LogicMonitor announced it was acquiring in December 2025, and the plan pages now sit on LogicMonitor's site. As of November 2026 the free Starter plan page lists 150 test runs a month and 30 test locations. An older signup page still says 300 runs, so check the number when you sign up. The free tier includes filmstrip and video, Lighthouse runs and traceroute; the API, private tests and bulk testing are paid.

The filmstrip is the most useful thing to show a client who doesn't care about metrics. A blank white frame at 4 seconds lands better than any score.

Which tool should you use for which job?

Prices and limits checked November 2026 against each vendor's own pages. Free tiers change, so re-check before you build a workflow around one.

A few opinions from doing this a lot:

1. Don't compare scores across tools. Different locations, devices and throttling give different numbers for the same page. Pick one tool for before/after and stick with it.
2. Never sell a score. "We'll get you to 100" is a promise you don't control. Sell the fixes and show the change in LCP and page weight.
3. Test on a real phone as well. If it feels slow on cellular, it is slow, whatever the number says.

What does a good client audit workflow look like?

Our order on a typical local business site:

1. PSI on the homepage and the main service page, mobile. Note whether field data exists. Screenshot the Core Web Vitals section if it does.
2. WebPageTest on the same URL from a location near the business, on a mobile connection profile. Save the filmstrip and the waterfall. This is where you find the 4 MB image and the five third-party domains.
3. Lighthouse in DevTools while you fix things, three runs each time, median result.
4. Rerun PSI (and WebPageTest if you used it) after deploy with the same settings, and put before/after side by side.
5. Optionally set up GTmetrix monitoring if the client is on a care plan, so you hear about regressions before they do.

When you report back, keep it plain. A fictional example:

I tested your homepage on a simulated mid-range phone. It took about 6 seconds for the main photo to appear, mostly because that one photo is 5 MB. After resizing it and removing two old tracking scripts, the same test shows it in under 2.5 seconds. Google's own tool, PageSpeed Insights, now rates that measurement as good. You can run it yourself at pagespeed.web.dev.

Only send something like that with your own real numbers in it.

Finding sites worth auditing

Search a trade and a town, open the top results on your phone over cellular, and run PSI on the ones that drag. That's a free, repeatable way to build a list.

If you'd rather start from a list, LeadReef finds local businesses whose websites are outdated, broken on mobile or missing, shows what's wrong with each one, and helps with a first mockup and outreach message. You choose who to contact and send it yourself, and the first 10 leads are free. You can also run it from Claude or ChatGPT over MCP and get the list straight in the chat. Full disclosure: we build LeadReef. Everything above works without it.

Frequently asked questions

Why is my PageSpeed Insights score different every time I run it?

Lab scores vary with network routing, ads or A/B tests on the page, and the test machine. Chrome's Lighthouse docs recommend treating performance as a distribution of results, so run it three to five times and use the median.

Why does PageSpeed Insights say "no data" for real users?

The site doesn't have enough Chrome traffic in CrUX over the last 28 days, at URL or origin level. That's normal for small local business sites. Use the lab results and test on a real phone.

Is GTmetrix still free?

There is still a free Basic account, but you have to sign up, and since GTmetrix's December 2023 plan change the free allowance has been small. Mobile device testing and premium locations need a paid plan, which starts at $5.99 a month as of November 2026.

Does a higher Lighthouse score improve Google rankings?

Not in any direct way you can count on. Google's Search Central docs talk about Core Web Vitals and real-world user experience, not the Lighthouse lab score, and say they align with what its core ranking systems reward alongside other page experience aspects. Treat speed as a fix for visitors, not a ranking lever.

Which tool should I show a non-technical client?

PageSpeed Insights, because it's Google's tool and they can run it themselves, plus a WebPageTest filmstrip if you need a picture of the problem. Skip the waterfall in client emails.

Sources (checked November 2026): Google for Developers, About PageSpeed Insights and Get started with the PageSpeed Insights API; Chrome for Developers, Lighthouse performance scoring; Lighthouse on GitHub, Throttling documentation; GTmetrix, homepage, plans and plan changes announcement; LogicMonitor, WebPageTest plans; WebPageTest, signup page; IT Brief, LogicMonitor acquires Catchpoint; web.dev, Web Vitals and Interaction to Next Paint; Google Search Central, Understanding Core Web Vitals and Google search results.