Prioritise Core Web Vitals improvements by the metric failing for real users and the affected page group. LCP measures loading, INP responsiveness and CLS visual stability. Select a problem from field evidence, then use lab recordings to investigate its cause. One Lighthouse score does not represent all users or guarantee rankings.
Understand the experience behind each metric
web.dev’s good-experience thresholds are LCP at 2.5 seconds or less, INP at 200 milliseconds or less and CLS at 0.1 or less. Assess the 75th percentile of visits, with mobile and desktop considered separately. These are measurement targets, not claims that your site already achieves them.
Identify the actual LCP element instead of guessing. Record the slow menu, filter or form interaction for INP. For CLS, locate what shifted, when and why. Where a shared template causes the issue, plan the correction for that component rather than one URL.
Give field and lab data different roles
Web.dev distinguishes field data from actual experiences and lab measurements under controlled conditions. A lab test can help diagnose a cause, but one test condition is not the entire audience. Record the period and page scope represented by field evidence too.
Insufficient field data on a new page does not prove good or bad performance. State the limitation and record the device, network and scenario used in controlled checks. This helps prevent changed test conditions from being mistaken for the effect of development work.
Rank findings by impact and practical scope
Record the failing metric, sample URL, template, user task and observed cause. A late-discovered main image needs different LCP work from a slow server response. Long JavaScript tasks versus delayed rendering for INP, or unsized images versus inserted banners for CLS, also need different fixes.
Start with verified causes affecting important page groups. Removing unnecessary lazy loading from the main image, breaking up long tasks or reserving media dimensions are possible fixes selected against evidence. Rerun the same scenario and check that forms, navigation and measurement still function.
| Metric / symptom | Diagnosis | Verify the change |
|---|---|---|
| LCP: main image arrives late | Resource discovery and download timing | Unnecessary lazy loading or resource delay addressed |
| LCP: content waits for server | Initial response and resource timings | Comparable tests show server improvement |
| INP: menu responds slowly | Long JS and rendering work during interaction | Visible response after interaction improves |
| CLS: content is pushed down | Image or banner causing the shift | Space reserved; later insertion no longer shifts content |
| Shared template issue | Same cause in representative URLs | Fix works on other important pages too |
Check that a performance correction preserves the task
Imagine an image compressed enough to load faster but no longer showing product details clearly. Transfer size alone would be an incomplete success criterion. This hypothetical example requires checking both speed and the visual information supporting a purchase decision.
Measure the change again under comparable conditions and complete the core task. Removing a script might improve a score while breaking enquiry measurement or functionality. Verify both sides. Track later field-data changes separately instead of reporting a lab gain as an already observed audience result.
Assign responsibility for performance checks when new images, campaign tags or components are added. A one-off cleanup can be undone by later releases. Durable improvement needs an ongoing publication habit.