A PageSpeed score of 98 tells you how a simulated device performed once. It does not tell you what your customers experienced. Those are different questions and only one of them matters.
Lab versus field, briefly
Lab data comes from a single synthetic test under controlled conditions. Field data comes from real visits on real devices over the trailing 28 days. Search engines evaluate the field data. Agencies screenshot the lab score, because it is prettier and it moves faster.
Where the two diverge most
- Device mix. If forty per cent of your traffic is on mid-range Android hardware over cellular, a test run on a simulated fast connection is fiction.
- Third-party scripts. Chat widgets, review embeds and tag managers behave differently under real conditions and often load conditionally.
- Cache state. Lab tests are usually cold. Real traffic is a mix, and the mix varies by page type.
What to fix, in order
- Largest Contentful Paint. Usually the hero image. Serve it in a modern format, size it correctly, set fetchpriority, and stop lazy-loading the one image that is above the fold.
- Cumulative Layout Shift. Almost always missing width and height attributes on images, or an injected banner. Reserve the space.
- Interaction to Next Paint. Long JavaScript tasks blocking the main thread. Audit what your tag manager is actually loading; on client sites we routinely find scripts for tools cancelled two years ago.
A realistic target
Chasing a perfect score is a poor use of budget. Passing the thresholds for seventy-five per cent of real visits on the templates that generate revenue is the goal. Get the product page and the service page right; the archive page from 2019 can wait.
Related reading
Not sure where your search traffic is leaking?
Send us your domain. We will return a 12-point technical and commercial audit within three business days — no fee, no obligation, and no automated PDF dump.