Why Your Lighthouse Score and Your Real Users Disagree

It's one of the more confusing moments in web performance work: a Lighthouse audit reports a 95, and the same page shows "needs improvement" in real-user Core Web Vitals data. Neither number is wrong. They're answering different questions.

Lab data answers "how fast can this page be?"

A Lighthouse run happens under controlled, idealized conditions — a specific simulated device, a specific simulated network speed, no other tabs competing for resources, usually run from a server with a fast, stable connection to your site. It's a consistent, repeatable benchmark, which makes it genuinely useful for tracking whether a specific change made things faster or slower. What it doesn't tell you is how the page performs for an actual visitor on an actual phone, on actual WiFi that's shared with three other devices in the house.

Field data answers "how fast is this page for real people?"

Real-user performance data — sometimes called RUM, for real-user monitoring — comes from measurements taken during actual visits: real devices, real network conditions, real interference from browser extensions and background processes. This is what search engines actually use for ranking-related signals, because it reflects genuine experience rather than a simulated one.

Why the gap happens

The most common explanation is audience mismatch. If a meaningful share of your real visitors are on older Android devices, or on 3G/4G connections instead of stable broadband, or in a region with higher latency to your server, their real experience will be measurably worse than a lab test run from well-connected infrastructure — even though the page itself hasn't changed at all.

A second common cause: lab tests typically load a page cold, once, with no other tabs or extensions. Real users often have several tabs open, ad blockers or other extensions intercepting requests, and background processes competing for the same CPU your page needs to render smoothly.

What to actually act on

Lab data is the right tool for iterating — make a change, re-run the test, confirm it helped, before it ever reaches real visitors. Field data is the right tool for understanding whether your actual audience is well served, and it's the number that should carry more weight when the two disagree, since it reflects what's actually happening rather than what's theoretically possible.