Web performance
Core Web Vitals: measure the experience, not just the score
LCP, INP, and CLS describe different parts of a page experience. Used carefully, they help teams find slow loading, delayed interaction, and unstable layouts. Used as a single trophy score, they can send effort in the wrong direction.
The three user-centered signals
Largest Contentful Paint measures when the main visible content appears. Interaction to Next Paint measures the responsiveness of interactions over a visit. Cumulative Layout Shift measures unexpected visual movement. Field assessment uses the 75th percentile and needs enough real-user data.
1. Know what each metric can tell you
LCP often points toward server response, render-blocking resources, image delivery, or client rendering. INP exposes long main-thread work and handlers that delay visual feedback. CLS reveals content inserted without reserved space, late font changes, or unstable components. None of them alone proves the root cause.
- Aim for an LCP of 2.5 seconds or less, an INP of 200 milliseconds or less, and a CLS of 0.1 or less at the 75th percentile.
- Review metrics by important template, device class, and market; an origin-wide average can hide a broken checkout or landing page.
- Treat the thresholds as experience guardrails, not substitutes for conversion, completion, accessibility, or qualitative feedback.
2. Use field and lab data for different jobs
Field data comes from real Chrome visits and includes varied devices, networks, caching, and user behavior. Lab data runs in a controlled environment and is easier to reproduce while debugging. Lighthouse Total Blocking Time is a useful lab signal for main-thread blocking, but it is not the same measurement as field INP.
- Start with field data to identify an affected population and confirm that the problem is real and sustained.
- Use lab traces, request waterfalls, coverage, and performance profiles to isolate the mechanism behind the field symptom.
- Repeat lab tests and compare like with like; a single run is sensitive to environment, cache state, extensions, and background load.
3. Fix the delivery path, not the number
For LCP, shorten time to first byte, remove avoidable render blocking, preload only the truly critical resource, and serve correctly sized modern images. For INP, break up long tasks, reduce unnecessary JavaScript, and provide prompt visual feedback. For CLS, reserve dimensions and avoid inserting content above what the user is reading.
- Find the actual LCP element and trace every dependency required before it can render.
- Profile a real slow interaction and reduce the work before the browser can paint the next frame.
- Give images, embeds, banners, and dynamic panels predictable space, including their mobile states.
4. Protect performance after the fix
Performance regresses when new scripts, fonts, media, experiments, or components enter the critical path. Set page-type budgets, test representative routes in delivery checks, and watch real-user trends over time. Search data changes slowly, so pair it with faster technical and product signals.
- Record a baseline and annotation for each meaningful release so trend changes have context.
- Monitor the 75th percentile and the distribution; a healthy median can coexist with a poor experience for many visitors.
- Recheck consented third-party scripts and tag-manager changes because their cost can vary outside your deployment cycle.
A reliable Core Web Vitals workflow
- 01 Identify the page type, device segment, geography, and time window represented by the field data.
- 02 Confirm which of LCP, INP, or CLS is failing and inspect the distribution, not only the summary label.
- 03 Reproduce the page in a controlled lab and capture the LCP element, layout shifts, main-thread tasks, and request chain.
- 04 Change the smallest shared cause likely to improve the real experience without damaging accessibility or functionality.
- 05 Repeat several comparable lab runs and test the complete user journey on a representative physical device.
- 06 After release, monitor field data, errors, conversions, and the performance budget long enough to confirm the result.
Performance is a product quality
The best optimization makes an important journey visibly faster or more stable for real people. A higher lab score is useful evidence, but it is not the outcome. Keep field and lab data connected, document the cause, and guard the improvement against regression.
Primary references
Continue with a related guide
These checks influence one another. Follow the next guide to verify the surrounding signals, not only one tag or score.