How to Fix Core Web Vitals: LCP, INP and CLS
Find the cause of poor Core Web Vitals and test fixes for loading, interaction responsiveness and page layout shifts.

Quick answer
Fix LCP by identifying when the main visible content becomes available, fix INP by tracing delayed interactions, and fix CLS by reserving space and eliminating unexpected shifts. Use field data to locate affected users and lab tools to reproduce and test the problem.
A fast local run disagrees with the Search Console report
Field metrics represent real users, while a local trace shows one environment. Use the trace to find which element rendered late, which interaction took time, or which content shifted. Use field data to assess the effect after enough data is available.
A before-and-after page-speed exercise
Consider an article page whose large cover is the LCP element on phones. A lab trace shows the browser discovers the image after a layout script finishes. The release moves a properly sized hero into the initial markup and leaves below-the-fold assets to load according to their role. Next, a filter on the related tools page has a slow event handler, and a delayed consent strip shifts the visible headings. These are three separate tests. Save an initial trace and verify image request discovery, then record the exact tap that is slow, then check a cold page load for unexpected motion. Use the same route, state and viewport in each comparison.
Check whether a redesign combines multiple changes that make the raw numbers difficult to interpret. A faster image request could coincide with a slower animation bundle. Inspect what starts when and which visible element or interaction the metric uses. After a release, field data may take time to reflect the new experience and might group several URLs. Give the site enough eligible traffic and observe the field report while continuing to test difficult conditions in the lab. Google describes the metrics as part of page experience. Treating a lab target as a promised rank increase is a different claim and cannot be established with the trace.
Read the three metrics
Understand a Search Console group with mobile warnings. Match LCP with loading of the largest visible content element, INP with interaction responsiveness and CLS with layout instability. Good targets are at most 2.5 seconds, 200 milliseconds and 0.1 respectively.
Common problem. A lab score is mistaken for the 75th-percentile field experience, or a sitewide average hides the affected template. Inspect field grouping and write down URL, device and reporting context when comparing data.
Keep lab and real-user evidence in separate columns so changes can be interpreted on the right timescale.
Find the LCP element
Improve a product page with a slow prominent image. Trace the actual LCP candidate, its network request, the point the browser discovers it and the time it can render.
Common problem. The developer optimizes a footer image while the initial view waits on script-driven discovery of its hero. Record a cold load with a performance trace and check whether server response, resource discovery, download or rendering dominates.
Test hero images and large headings separately if different screen sizes pick different elements.
Fix interaction lag
Make a filter control respond while a large results list is visible. Trace the interaction from input through event processing to the next rendered frame. Check whether unrelated main-thread work blocks it.
Common problem. A fast network response looks like proof of fast interaction even though the interface takes far longer to paint. Repeat the same filter on a slower device and find the longest task competing with it.
Measure user-visible feedback early in the interaction as well as completion of the full results update.
Reserve space for dynamic content
Stop a summary from jumping when photos or a banner arrive. Give images and embeds space that matches the expected aspect ratio and keep late dynamic elements clear of content read early.
Common problem. An asynchronous banner moves a paragraph just as the visitor starts reading or tapping. Record a slow load and observe the shifted elements and the source that appeared above them.
Test consent states, signed-in variants and localization where text length changes needed space.
Review third-party cost
Understand slow interactions after installing a widget. Compare traces with the widget on and off in staging. Check network load, execution timing and what user action actually requires it.
Common problem. A script apparently loads asynchronously but still runs expensive main-thread work during interactions. Test the same action under realistic CPU and network constraints and inspect the responsible script.
Delay optional execution until needed or replace the integration only after verifying its actual business purpose.
Set a performance budget
Keep a fast article template fast after redesign. Set limits for the critical image, early script work and meaningful responsiveness over representative traces. Tie each budget to a user outcome.
Common problem. A blanket size target misses how fonts, widgets and script timing affect the first view. Check several recently published articles, including the largest cover and longest list of related content.
Add a release check for useful representative pages and track exceptions with expected expiry.
Verify mobile results
Roll out a fix for a performance group affecting mobile users. Use a repeatable viewport, throttling and clean browser first, then compare with real devices and field reporting when enough data arrive.
Common problem. Launch-day field data still contains previous experiences, while a lab improvement is reported as complete field recovery. Save current traces and record the release date so measurements can be read against reporting windows.
Check important desktop and keyboard interactions as part of the normal release smoke test.
Prioritize changes by impact
Choose between server improvement, hero optimization and interaction cleanup. Rank candidates by affected pages, user time lost, confidence in evidence and implementation effort.
Common problem. A team chases a small metric gain on an obscure page while a repeated slow search interaction remains. Run one targeted fix on a stable sample before applying it more broadly.
Review errors, accessibility and conversion-critical behavior as well as the timing metric.
Match each metric to the right tool
Look at the Search Console Core Web Vitals group for a starting point, then examine a representative URL in PageSpeed Insights or a real-user monitoring setup where available. Capture a browser performance trace when a user can reproduce a delay or layout jump. A single navigation trace is useful for analyzing loading, while interaction analysis requires reproducing the taps, clicks or keyboard actions that cause the delay. Test cold and warm conditions when they reflect actual visitors. A first visit from search may wait for resources or show a consent interface that a returning visitor already has cached. Keep both results if they reveal different problems. Grouped field results often deserve focused fixes by template instead of changing every component on the site.
