Blog · 16 September 2026
Your site is fast on your laptop. Real users wait two seconds for the first byte.
Lab tools measure from a developer's machine. Google ranks on what real users experience. On one site the two disagreed by a factor of sixty — and only one of them counts.
There are two ways to measure how fast a page is, and they answer different questions.
A lab measurement loads the page from one machine, once, under conditions you choose. It is repeatable, which makes it useful for comparing two templates or two deploys. It says nothing about your visitors.
A field measurement is the 75th percentile of what real Chrome users experienced over the last 28 days, published by Google as the Chrome UX Report. It is noisy and lags by weeks. It is also the only performance number Google’s ranking systems read.
The gap, on a real site
This week Vercatus measured a small commercial site both ways:
| Lab (our network) | Real users, p75 | |
|---|---|---|
| Time to first byte | 38 ms | 2,290 ms |
| Largest Contentful Paint | 204 ms | 2,568 ms |
The lab said the site was fast — sub-second on every metric. Real users wait over two seconds before the first byte of HTML arrives. That is a poor time to first byte by Google’s published thresholds, and the site’s Largest Contentful Paint is in the needs improvement band as a direct consequence.
Nothing on the page explains the gap. The origin does: uncached HTML, a slow application response, or users far from the server. A lab run from near the origin cannot see any of that.
Why first byte matters beyond visitors
Most performance metrics are about people: layout shift, blocking time, responsiveness. AI crawlers do not render pages, so those metrics are irrelevant to what they read. First byte is the exception. It reaches everyone — visitors, Googlebot’s crawl rate, and the AI crawlers that fetch HTML only. A slow origin is slow for all of them.
How Vercatus reports it
Both numbers, side by side, per template, coloured by Google’s bands — and a finding when they disagree that says which one to believe: “Real users wait 2,290 ms for the first byte, while the lab measured 38 ms from our network. The gap points at the origin, not the page.” The recommendation is to cache full HTML at the edge for anonymous visitors and confirm the cache-hit ratio by region; the number to watch is the field p75 on the next run.
The thresholds and the reasoning are in the knowledge base as perf.cwv-thresholds, perf.field-over-lab and perf.lcp-causes, all grade A, all sourced to Google’s own documentation. The method is described in Performance by template.