Different performance measurements describe different parts of a visit. A fast server response does not guarantee that an image appears quickly, and a page that loads quickly can still freeze when someone opens its menu.
The three Core Web Vitals
Google’s Web Vitals reference defines these user-experience measures and recommends evaluating the 75th percentile of visits, separately for mobile and desktop:
| Metric | What it describes | Good threshold |
|---|---|---|
| Largest Contentful Paint (LCP) | When the largest visible content element appears | 2.5 seconds or less |
| Interaction to Next Paint (INP) | How promptly the page responds visually to interactions | 200 milliseconds or less |
| Cumulative Layout Shift (CLS) | Unexpected movement of visible content | 0.1 or less |
Meeting a threshold is a useful experience target, not a guarantee of a particular search position.
Use TTFB to investigate the initial response
Time to First Byte measures the interval before the response begins arriving. Depending on the measurement, that interval can include connection setup and network travel as well as server processing. It should not be read as database execution time.
For example, if an article has a prompt initial response but a late main image, inspect the image request and rendering path. If many uncached pages wait before receiving their first byte, investigate the application, hosting resources and network route.
Separate field data from lab tests
Field data represents visits from actual users over a collection period. A lab test runs under selected conditions and is useful for diagnosing a repeatable problem. They can disagree because visitors use different devices, networks, pages and interactions. A site without enough field samples is not automatically fast or slow.
Keep a small measurement record
- The page and visitor task being tested.
- The tool, device and network conditions.
- The relevant metric before and after a change.
- Whether the result is a lab run or a field measurement.
- Whether the task still works correctly.
Use the speed optimization checklist to turn that evidence into a focused change. Treat engagement or conversion measures separately: they also depend on content, audience and the offer, not just loading speed.