Your PageSpeed score isn't the number Google uses

The steps

Run the page through PageSpeed Insights. Go to pagespeed.web.dev, paste the address and look at the Mobile tab first.
Read the top section, not the circle. "Discover what your real users are experiencing" shows the last 28 days of real Chrome visitors. Those are your Core Web Vitals. Good means Largest Contentful Paint (LCP) of 2.5 seconds or less, Interaction to Next Paint (INP) of 200 milliseconds or less and Cumulative Layout Shift (CLS) of 0.1 or less, measured at the 75th percentile.
If it says No data, use the lab figure. Small sites often don't get enough Chrome visitors for Google to report them. Scroll down to Diagnose performance issues and use the Largest Contentful Paint figure there as your guide. It comes from a single test on an emulated slow phone, so treat it as a rough guide.
Find out what your main content is. In the same lab section, the LCP breakdown insight names the element that counts as your largest content. On most WordPress sites it's the hero image at the top of the page.
If it's an image, don't lazy-load it. Remove loading="lazy" from that one image and add fetchpriority="high". WordPress 6.3 and later already does this for the image it guesses is the main one, but sliders, page builders and lazy-load plugins often add lazy loading back. Look for an option such as "exclude the first image" or "skip above the fold" in the plugin's settings.
Test again. The lab figure changes straight away. The real-visitor figures are a 28-day average, so give them a few weeks.

Why Amazon scores 66 and still passes

When this was checked on 12 Nov 2026, amazon.co.uk scored 66 in the lab test, with a lab LCP of 5.5 seconds. Its real visitors, over the previous 28 days, got their main content in 1.6 seconds and passed all three Core Web Vitals.

The circle is one page load on an emulated mid-range phone over a throttled connection. The top section is what millions of real visits actually got. When they disagree, the top section is the one that reflects your visitors, and it's the one Google's page experience signals use.

A high score is still a decent sign, and a low one is worth looking into. Don't chase 100 at the expense of things your visitors use.

Tested

12 Nov 2026: both reports were run on pagespeed.web.dev in desktop Chrome on 12 Nov 2026, Mobile tab. amazon.co.uk: real visitors passed Core Web Vitals (LCP 1.6 s, INP 189 ms, CLS 0.01) while the lab test scored 66 with an LCP of 5.5 s. elkdigital.co.uk: No Data for real visitors on Mobile and Desktop, lab score 98 and LCP 1.8 s. Your figures will differ from day to day.

Where this comes from

Google's page experience documentation points to Core Web Vitals, which use field data from real Chrome users through the Chrome UX Report. The good thresholds and the 75th percentile come from web.dev. INP replaced First Input Delay as a Core Web Vital in March 2024. Sources: Understanding Core Web Vitals and Google search results, Web Vitals, Optimize Largest Contentful Paint, About PageSpeed Insights.

Transcript

Your PageSpeed score isn't the number Google uses. Let me show you the one that is.

Put your site into PageSpeed Insights. Skip the big circle and look at the top section. That's what real visitors got.

Those are your Core Web Vitals. Your main content should load within 2.5 seconds, and a tap should get a response within 200 milliseconds.

Mine says No data. Small sites often don't get enough Chrome visitors for Google to report it. If yours says that too, use the lab figure further down.

If your main content is slow, check your hero image. When it's lazy-loaded, the browser fetches it late. Turn lazy loading off for that one image.

Full checklist at blog dot elkdigital dot co dot uk slash page speed.

On screen: a rebuilt PageSpeed Insights report for amazon.co.uk: a lab Performance score of 66 crossed out, then the real-visitor section showing Core Web Vitals Passed with LCP 1.6 s and INP 189 ms. Then ELK Digital's own report showing No Data in the real-visitor section and a lab LCP of 1.8 s, and a code view of a hero image tag where loading="lazy" is swapped for fetchpriority="high". The figures are real, taken on 12 Nov 2026. The voice is generated from a sample of Jamie's own voice.

The presenter in the video is an AI avatar made in HeyGen, speaking with a clone of Jamie's voice. The script, demo and test were made by ELK Digital with Claude.