The essentials
- Two measurements in one: PageSpeed Insights shows the field data from Chrome visitors (CrUX) and a lab test produced by Lighthouse; only the first counts for Google.
- The score out of a hundred: it comes from the lab, it varies from one test to the next and it isn't a ranking factor; the Core Web Vitals thresholds are.
- The priority: fix what is failing in the field data first, starting with LCP, then INP, then CLS.
- A benchmark: a small business usually gains 20 to 40 score points from three actions: images in the right format, deferred scripts, decent hosting.
PageSpeed Insights is Google's free tool for assessing a page's performance. It combines two very different sources that it is essential to tell apart: real-user experience data, collected from Chrome users who opted into sharing (the CrUX report), and a lab test run on demand by Lighthouse, the audit engine built into Chrome. The first source feeds the Core Web Vitals Google takes into account; the second is for diagnosis.
Field or lab: the distinction that changes everything
| Criterion | Field data (CrUX) | Lab (Lighthouse) |
|---|---|---|
| Origin | Real Chrome visitors over 28 days | An on-demand simulation, with throttled machine and network |
| Availability | Only if the page gets enough traffic | Always |
| Stability | Changes slowly | Varies by 10 to 20 points from one test to the next |
| Used by Google | Yes, for page experience | No |
| Use | Knowing whether the problem exists | Knowing where it comes from |
The practical consequence is simple: judge on the field data and diagnose with the lab. A page whose field data is green doesn't need optimising because its Lighthouse score reads 62. Conversely, a page with no field data, for want of traffic, can only be assessed in the lab, with the caution that requires.
The three metrics to watch
- LCP (Largest Contentful Paint): the time it takes for the largest visible element to appear, usually the main image or the heading. Threshold: under 2.5 seconds. It is the one that fails most often and the one that is fixed fastest.
- INP (Interaction to Next Paint): the delay between a visitor's action and the page's visible response. Threshold: under 200 milliseconds. It replaced FID in March 2024 and penalises excess JavaScript.
- CLS (Cumulative Layout Shift): visual stability, meaning content shifting during load. Threshold: under 0.1. Caused by images with no dimensions, banners and fonts.
The thresholds must be met by 75% of visits for the page to count as passing. The detail of the measurements is in the Core Web Vitals.
Prioritising the fixes
- Start with the field data: if it is green, move on; performance optimisation isn't the priority.
- Deal with LCP first: header images compressed and in a modern format, dimensions declared, priority loading for the main image, hosting with a short response time.
- Then INP: reduce and defer third-party scripts, which are the primary cause; every tracking tool, chat widget and banner adds delay.
- Then CLS: declare the dimensions of every image and ad block, preload the fonts.
- Check on mobile: the mobile tab is the one that counts, since most traffic and most measurement are there.
- Wait 28 days: field data updates over a rolling window; a fix deployed today shows in the report a month later.
Using Lighthouse beyond performance
Lighthouse produces four audits, often forgotten in favour of the performance score alone. The accessibility audit spots insufficient contrast, images with no alt text and form fields with no label, useful alongside proper accessibility work. The best practices audit flags console errors, badly sized images and security problems. The SEO audit checks the basics: title tag, meta description, legible text, crawlable links. These three audits are more stable than the performance score and are worth reading at every deployment.
The traps to avoid
- Chasing a score of a hundred: the effort to go from 85 to 100 far exceeds the gain, and the score isn't a ranking factor.
- Testing only once: the lab varies a great deal; three successive tests give a fairer idea.
- Testing only the home page: product and article pages receive most of the traffic and often have different problems.
- Confusing the score with the experience: a page scoring 95 that shows a blocking banner is still unpleasant.
- Neglecting the link to conversion: every second of load time costs 5 to 15% of conversions, which makes the subject more commercial than technical; see CRO.
- Forgetting Search Console: its Core Web Vitals report gives the overall view by page group, where PageSpeed Insights tests one URL; see Google Search Console.
How GreenRed helps
Rather than juggling several tools, GreenRed's Site health module brings these metrics together in a single dashboard, compares them over time and tells you which actions come first. You can try it free, with no card, from the Pricing.
Frequently asked questions
What is the difference between PageSpeed Insights and Lighthouse?
PageSpeed Insights is the online interface showing two things: real-user experience data from Chrome visitors and a lab test. Lighthouse is the engine that produces that lab test; it is also built into Chrome's developer tools.
Does the score out of a hundred count for search?
No. Google uses real-user experience data, through the Core Web Vitals, not the lab score. That score varies by 10 to 20 points from one test to the next and is for diagnosis, not assessment.
Where should you start to improve speed?
With LCP, which fails most often and is fixed fastest: compress the header images, serve them in a modern format, declare their dimensions, load the main image with priority and check the server's response time.
Why doesn't my field data appear?
Because the page doesn't get enough visits from Chrome browsers that opted into data sharing. In that case only the lab test is available, and it should be interpreted with caution, alongside the site-level Search Console report.