Troubleshooting Core Web Vitals: From Symptoms to Root Causes

The most common CWV troubleshooting mistake is looking at a lab score screenshot and immediately starting to change code. The right approach is to first examine…

Insights:Troubleshooting Core Web Vitals: From Symptoms to Root Causes

Start by distinguishing two types of data

Data type Source Use case
Field data Search Console, CrUX Analyze real users, grouped by page
Lab data Lighthouse, PageSpeed Insights, WebPageTest Reproduce issues and identify root causes

Field data tells you whether to fix something, while lab data reveals why performance is poor. Do not substitute real user metrics with a single lab score, and do not let average field values hide severe issues affecting specific templates.

Group by template and device first

Do not start with the homepage. Homepage performance is often individually optimized; real problems typically appear on article pages, listing pages, product pages, or form pages.

In Search Console’s Core Web Vitals report, check:

  • How many URLs underperform on mobile versus desktop.
  • Which template type has the most “Needs Improvement” URLs.
  • Whether the bottleneck is LCP, INP, or CLS.
  • Whether changes show improvement within the 28‑day rolling window after launch.

LCP: Identify the LCP element first, then break it into three parts

A slow LCP can usually be decomposed into:

  1. TTFB: server response time, caching, and CDN delivery.
  2. Resource loading: whether the hero image, fonts, CSS, or JS are blocking.
  3. Render delay: whether the main thread is occupied by long tasks.

Troubleshooting steps:

  • Confirm the actual LCP element in DevTools Performance—do not guess which image loads first.
  • Check whether the LCP image uses preload and whether it is served at an appropriate size and format.
  • Verify that critical above‑the‑fold CSS is not blocked by large files.
  • Use curl to check TTFB and cache headers.
  • If TTFB is already high, fix the server and caching first — don't keep compressing images.

    INP: Identify Long Tasks and Event Handling

    A poor INP score usually means the main thread is blocked:

    • Third-party scripts: analytics, live chat, ads, A/B testing.
    • Large JS bundles: unnecessary components loaded on the initial screen.
    • Event handlers: synchronous heavy computation, requests, or rendering triggered by clicks.
    • Layout thrashing: a single interaction triggers multiple reflows.

    When troubleshooting, record a real user interaction rather than just page load. Observe what the main thread is doing between the click event and the next frame.

    Fix priorities:

    1. Defer non-critical third-party scripts.
    2. Split long tasks and offload computation to idle periods.
    3. Reduce duplicate requests within a single interaction.
    4. Add debouncing, caching, or asynchronous handling for inputs and clicks.

    CLS: Identify Elements Without Reserved Space

    Common sources of CLS issues:

    • Images without explicit width/height push content down after loading.
    • Font swap causing line-height shifts.
    • Ads, recommendation blocks, or cookie banners inserted into the layout, crowding the main content.
    • Asynchronous content suddenly appears above the fold.

    Fix actions:

    • Reserve fixed dimensions or aspect ratios for images, videos, and ad slots.
    • Use font-display and try to lock font families and weights.
    • Avoid injecting dynamic content that alters layout in the initial viewport.
    • Use DevTools' Layout panel to pinpoint which frame the shift occurs in.

    Root-Cause Diagnosis Order

    1. First group by template to confirm the issue isn't site-wide average.
    2. Then analyze by device, region, and network type.
    3. Next examine third-party integrations and ad placements.
    4. Finally review themes, plugins, and component code.
  • Finally, check servers, CDN, and databases.
  • This order avoids the mistake of swapping servers only to find the real issue was images missing width/height attributes.

    How to Verify After Fixing

    • Change only one variable at a time.
    • Re-test using the same device, network, and page.
    • Watch lab data for trends; wait for a 28-day window on field data.
    • Record the change date, template, metrics, and results.

    Don’t use “100 on Lighthouse today” as proof that real users have improved. Lab scores are diagnostic tools, not acceptance criteria.

    What Not to Do

    • Don’t remove body content, images, or features just to chase a higher score.
    • Don’t compare before-and-after scores across different networks or cache states.
    • Don’t run audits on the homepage only.
    • Don’t hand off every metric issue to “switch themes” or “add CDN.”

    CWV is a user experience metric, not an exam. Find where real users get stuck first, then decide what to change.

    Related reading: