One PageSpeed Run Is Not Your Core Web Vitals
PageSpeed Insights shows two different things on one page: 28 days of real Chrome visits and one simulated load. Read the real-user numbers first; when a site has none, take the median of several lab runs.
On November 6, 2026, PageSpeed Insights scored airbnb.com's homepage 41 on mobile, with an LCP of 17.7 seconds. In the same report, real Chrome users of that page had an LCP of 3.3 seconds. Before anyone rebuilds a homepage over a red score, check which half of the report the number came from. PageSpeed Insights shows two kinds of data on one page, and they answer different questions.
Field data is what visitors lived through. It comes from the Chrome UX Report: real Chrome visits over the previous 28 days, summarized at the 75th percentile. When a page has too few visits, PageSpeed falls back to the whole site (the origin), and when the site has too few as well, there is no field data at all. The Core Web Vitals assessment is computed from these numbers: it passes when LCP, INP and CLS are all good at the 75th percentile. Google Search describes Core Web Vitals as metrics that measure real-world user experience.
Lab data is one simulated load. Lighthouse loads the page once, emulating a mid-tier phone on a mobile network, from a Google datacenter in North America, Europe or Asia. One device, one network, one location. Google describes lab data as useful for debugging, and notes that scores move with conditions outside the test, such as which ad or A/B variant was served.
What the gap looks like. Three sites I measured on November 6, 2026, all mobile:
- airbnb.com, three runs started at the same moment: performance 41, 41 and 46, with lab LCP at 17.7, 17.7 and 18.2 seconds. Real users of that page, in the same reports: LCP 3.3 s, INP 365 ms, CLS 0.14. Both say the page needs work, but the lab LCP is more than five times the real one.
- A client's Webflow homepage, two single runs a few minutes apart: lab LCP 3.3 s, then 7.1 s. Real users: LCP 1.8 s, INP 97 ms, CLS 0.01, good on all three. Either lab run on its own would have reported a problem that visitors do not have.
- My own site, for which PageSpeed returns no field data at all: medians of three runs taken the same day ranged from 90 to 100, with LCP between 0.9 and 1.3 seconds. Without field data, the lab is all there is, and even a median moves. The score on my home page is now the median of five runs, with stalled or broken test loads run again.
How I read the report.
- Field data first. If the page, or failing that the site, passes the Core Web Vitals assessment, a red lab score is a debugging lead, not a verdict.
- Check which level the field data describes. Page-level numbers are for that URL; origin-level numbers blend every page of the site.
- No field data: run the lab test at least three times and report the median, mobile against mobile.
- Use the lab report to find causes: the LCP element, render-blocking files, the elements that shift. That is what it is good at.
- Expect field data to move slowly. It covers a trailing 28 days, so a fix shipped today shows up gradually, not the next morning.
If you report speed to a client, give the real-user 75th percentile with its date, or the median of several lab runs labeled as such, not a single screenshot.
My free SEO audit works the same way: its instant check shows real-user LCP, INP and CLS when Google has them, and one lab run, labeled as one, when it does not.
Want This Level of Care on Your Site?
Start with a free audit of your site, checked by me. Or book thirty minutes with your project on screen: you leave with a scoped plan and a number, whether or not we work together.
