Lab vs field: yardstick

Two instruments, two realities

PageSpeed Insights puts two kinds of data side by side, and most people only read the big number at the top. That number comes from a lab test: Lighthouse loads your page in a simulated browser, with an emulated mobile connection and throttling applied on purpose. Every run measures the same thing under the same conditions. That makes lab data excellent for comparison and for tracking down technical debt.

Field data is the other instrument. Google collects it anonymously from real Chrome users, aggregates it in 28-day windows, and reports it as CrUX: the p75 value of, say, your Largest Contentful Paint. Not a simulation but your visitors’ reality, with their networks, their devices, and their caches. For the same site those two sources can be far apart — not because one of them lies, but because they answer different questions.

What we measured

For the benchmark edition of September 10, 2026, we measured 19 website builders from the top 10 of “website laten maken” (having a website built) in Belgium and the Netherlands. Lab side: three mobile Lighthouse runs per homepage through the PageSpeed Insights API, median per value, as always. Field side, one day later: CrUX field data for the same domains. Both measurements are publicly verifiable.

One limitation up front: CrUX only collects data from sites with enough real Chrome visits. Of the 19 agencies, 7 had a usable field sample. The other 12 are missing from the table below — which is itself a finding we will return to.

The gap: up to 16.7x between lab and field

Put the lab LCP of those seven domains next to their field LCP, and there is no nuance — there is a chasm.

The median gap is 8.1x. Conversal “loses” 18.2 seconds of LCP in the lab; real visitors see the largest content element on screen in just over a second. Yourhosting, the number 1 in the Netherlands, measures 21.4 seconds in the lab — and 2.08 seconds with real users. That is not a small difference; that is a different conclusion.

Why the lab can diverge this much

A single lab test rarely tells the whole story. The causes we see in the data stack up:

• External scripts load slowly in the lab: campaign and chat tools, cookie walls, and marketing pixels get fetched serially in a fresh test browser. Real visitors often have those scripts cached or fetch them in parallel. Sites heavy on third-party tools show this pattern strongest.
• Artificial throttling: Lighthouse deliberately slows the connection to emulate an “average” user. Anyone on wifi or a fast network never experiences that brake. For an LCP that is mostly network time, the two drift apart automatically.
• One run versus a month of aggregation: the lab test is a snapshot on one day; the field value is the p75 over 28 days of visits. Daily fluctuation, a slow ad server that afternoon, a cache miss on first hit — a month of real visits smooths all of that out.

The strongest evidence: the number 1 was never slow

The most striking pillar in the data is Yourhosting. In the benchmark it received the worst lab score in the Netherlands: Performance 28, LCP 21.4 seconds. The obvious conclusion: speed plays no role in this ranking. But the CrUX history of the domain tells the opposite story.

Over the last 25 weeks, Yourhosting’s field LCP stayed between 1.78 and 2.89 seconds in every period, with the share of “good” visits around 74 percent in the most recent period. With real users, the site was never unacceptably slow. That changes the conclusion completely: Core Web Vitals simply could not hold this number 1 position back, because there was nothing to hold back.

What this means for your ranking

Three things are true at once. One: a low lab score is not a ranking alarm by itself. Two: speed does count, but through field data — the Core Web Vitals Google derives from CrUX, as part of page experience. Three: it is a tiebreaker, not a headline rule. Relevance and authority weigh more.

Our own data shows this. Across all 19 agencies, the sites with the best lab scores even had slightly lower positions on average (Spearman rank correlation +0.32 between position number and score — clearly not a ranking driver). Field LCP correlates zero with position (−0.07). What does track: the authority of the ranking page, the number of linking domains, and the brand behind the site. That is the subject of our follow-up research.

What we do with it

We measure lab, and we keep doing so. A lab test is reproducible, independent of visitor counts, and fair to everyone in the table — including the 12 agencies without a field sample. What we add from this edition onward is the field reality next to it, so you see not just the artificial score but also what visitors experience. Two yardsticks, two answers, and finally an honest story about both.

For your own site this means a fixed order: measure your field status first, then use lab tests to find the cause of whatever is red, and treat the lab primarily as a development tool. The concrete steps are in the guide to improving load time.

Do this yourself

Want to check this for your own site? Three tools are enough:

PageSpeed Insights (pagespeed.web.dev) shows both worlds on one screen: the field p75s from real visitors at the top, the lab score with diagnostics below. Start there.

The Core Web Vitals report in Search Console shows per page group whether real visitors clear the thresholds — the most direct source for your own site.

Google’s CrUX documentation includes the interactive dashboard where you can view the 25-week trend for any domain — the same history we used for Yourhosting.

And remember with every number: lab tells you where the technical debt sits, field tells you what your visitors feel. Use both, in that order.