Three metrics. A clearer picture.

Largest Contentful Paint marks when the largest visible text or image is rendered. It is a useful proxy for when the page’s primary content seems available. “Good” is at or under 2.5 seconds at the 75th percentile. A large hero image, slow server response, render-blocking code, or late-discovered resource may contribute. Find the actual LCP element before optimizing: a page’s biggest file may not be its LCP element.

Interaction to Next Paint measures the latency of interactions throughout a visit, such as tapping a control, and reports a representative responsiveness value. “Good” is at or under 200 milliseconds at the 75th percentile. Heavy JavaScript and long main-thread tasks are common things to investigate, but use a trace or field diagnostics to identify the actual culprit. Total Blocking Time (TBT), often shown in a lab report, can help flag lab-thread blocking; it is not INP and does not represent a visitor’s full set of interactions.

Cumulative Layout Shift measures unexpected movement of visible content. “Good” is at or under 0.1 at the 75th percentile. Images without reserved dimensions, late-loaded ads, injected banners, and changing fonts can move content after it appears. Set image dimensions or aspect ratios, and reserve space for content that arrives later. This helps visitors keep their place and reduces accidental taps.

The thresholds come from Google’s Web Vitals guidance and classify field experiences; intermediate ranges need improvement and poor ranges are above LCP 4 seconds, INP 500 milliseconds, and CLS 0.25. The overall Core Web Vitals assessment uses all three at the 75th percentile when sufficient data exists. The page or origin may lack enough samples to report a value.

Field metrics aggregate real Chrome visits over a rolling collection window. In PageSpeed Insights this is commonly a 28-day period, and if URL data is unavailable, PSI may fall back to origin data. Lab tools simulate a test with a controlled device and network profile; reruns may vary. Field data tells you what eligible visitors experienced. A lab run is a repeatable place to investigate possible causes. Neither tells the whole story alone.

Bare Audit pairs a single mobile Lighthouse snapshot with separately labeled CrUX data when available. It may show TBT as a lab responsiveness clue, but does not label TBT as real-user INP. Use Search Console for site-level Core Web Vitals groups and field trends, then debug representative pages with browser tooling. Google uses Core Web Vitals among page-experience signals; passing them does not guarantee a ranking position or compensate for irrelevant content.

A one-page mobile scan cannot represent all templates, devices, journeys, or returning-user conditions. Thresholds guide diagnosis; they are not a guarantee of success.