Lighthouse Scores vs Real Users: What Performance Data to Trust

You can get a perfect 100 in Lighthouse and still be slow where it matters. You can also score a 62 and serve visitors brilliantly. If that sounds broken, it is not — the two tools measure different things, and confusing them wastes a lot of optimization effort.

Lighthouse runs in a lab. It loads your page in a simulated browser on a simulated device over a simulated network, under conditions you can repeat. That control is its gift and its limit: repeatability. Run Lighthouse twice on an untested site and you will get different numbers, because it measures one synthetic visit at one moment from one location — usually a data center nowhere near your visitors.

Field data is the opposite. Google collects real performance from real people: actual devices, actual networks, actual states — full caches, crowded subways, three-year-old phones. This is the Chrome UX Report, and it powers the field sections of PageSpeed Insights and the Core Web Vitals that influence ranking. Your field data is not an estimate of how your site performs; it is how your site performs.

Why the two disagree. Lighthouse typically tests a cold cache over a fast simulated connection — often flattering. Your real visitors arrive with polluted caches, shared IPs, extensions, and mid-range hardware — often sobering. A site can pass a lab run while its field data fails, and the reverse happens too: a page that scores modestly in the lab can feel instant in the field if it is small, cached and close to users.

So which do you trust? Field data for truth, lab data for debugging. Rankings follow the field. But field data cannot tell you what to fix — it only tells you something felt slow. When field numbers are bad, Lighthouse and its traces are the microscope: which request blocked rendering, which script monopolized the main thread, which image was the LCP element. You diagnose in the lab and verify in the field.

Practically: check PageSpeed Insights for both views at once. If field data exists for your site, treat it as the verdict and the lab score as a tool. If your site has little traffic and no field data yet, use lab scores as a proxy — but always pair them with a real device test, because the median visitor's phone is humbler than any laptop.

A small warning from experience: chasing a green 100 can become its own project, with diminishing returns after the real problems are fixed. The scores that move rankings and revenue are field numbers on real phones, and those improve by shipping less, caching more, and staying close to your users. A 90 with excellent field data beats a synthetic 100 every day of the week.

If your scores disagree and you want to know which one is lying, that is precisely what a performance audit answers — real-user data first, lab traces second, fixes in priority order.