跳至正文
网站性能与 Core Web Vitals · Google SEO

Failing Core Web Vitals in GSC: What to Fix First for LCP, INP and CLS

// / / 光算科技

When Search Console reports failing Core Web Vitals, first identify the device, URL group, failing metric and affected user task. There is no universal “CLS, then LCP, then FID” repair order. The current metrics are LCP, INP and CLS; prioritize by severity, reach, business importance and a cause you can reproduce.

Understand what the report represents

The Search Console Core Web Vitals report uses CrUX field data and groups similar URLs separately for mobile and desktop. A group's status reflects its worst metric category. An example URL is not necessarily the only page needing repair, and the values are not measured by Googlebot during a crawl.

MetricExperience measuredGoodNeeds improvementPoor
LCPRendering of the largest eligible visible content≤2.5 s>2.5 to 4 s>4 s
INPResponsiveness to clicks, taps and keyboard interactions during a visit≤200 ms>200 to 500 ms>500 ms
CLSVisual stability against unexpected layout shifts≤0.1>0.1 to 0.25>0.25

Assess the 75th percentile of real experiences, not the average. Mobile and desktop use the same numerical thresholds. INP replaced FID and includes input delay, processing and presentation of the next frame, rather than only the wait before the first interaction starts. See the Web Vitals documentation for definitions and measurement options.

Missing data does not mean a pass. GSC omits groups without sufficient LCP and CLS data, while PSI may fall back from a URL to origin-level data. The PSI documentation says all three metrics must be good when all are available. If only INP lacks data, LCP and CLS can determine the assessment; without sufficient LCP or CLS it cannot be assessed. A pass in that limited case is not evidence that INP was measured and found good.

Choose a valuable problem before choosing a technique

  1. Repair functional failures first: unusable controls, hidden content or lost form data should not wait for a performance report.
  2. Prioritize Poor groups: within the same category, compare reach and the importance of the affected pages.
  3. Identify a shared cause: a template-wide missing dimension may justify a template fix; one unusual embed does not establish a site-wide architecture problem.
  4. Include implementation risk: verify payment, consent, support and analytics behavior when a change touches them.

For example, unusably slow product filtering may deserve attention before a minor article-page shift. If one ad container moves content across many heavily read pages, it may instead be the best first fix. These are hypothetical prioritization examples, not a mandatory metric order.

For LCP, locate the element and break down the delay

The LCP optimization guide separates TTFB, resource-load delay, resource-load duration and element-render delay. Establish whether the LCP element is an image or text and inspect its waterfall before compressing a banner.

  • Late HTML: inspect redirects, connection time, cache misses and backend work.
  • A late resource request: check discovery, unnecessary lazy loading and script-dependent visibility.
  • A slow transfer: compare actual image dimensions, quality and formats.
  • A delayed render after transfer: inspect styles, fonts, main-thread work and reveal logic.

A smaller file may not improve LCP if the page still waits for the same script before displaying it. Preloading every resource is not a solution, and no single file-size ceiling establishes a pass.

For INP, reproduce the task rather than just reload the page

Try menus, search, filters, cart actions and form input both during and after loading. The INP guide recommends finding slow interactions in field evidence and reproducing them locally, then separating input, processing and presentation delays. Without field data, testing typical tasks is still useful, but it remains local diagnosis.

Long tasks, repeated handlers, broad DOM updates and expensive layout can contribute. Reduce unnecessary work, then consider splitting tasks or changing rendering. Adding defer does not guarantee responsive interactions. TBT in an automated Lighthouse loading test can provide clues, but it is not field INP or an end-to-end checkout test.

For CLS, inspect the visit beyond initial loading

The CLS guide includes shifts from later images, embeds, fonts and dynamic content. Check intrinsic ratios, mobile art direction, ad containers and font swaps. A fixed height that works on one screen may be wrong on another.

An overlay can obscure the main content without moving other elements, so a low CLS value does not certify a good dialog experience. Conversely, expected movement following an interaction needs to be assessed under the actual measurement conditions rather than assumed to count identically.

Verify the release, then follow field results

Record versions, device settings, network conditions, URLs and interaction paths. Use Lighthouse and a browser Performance trace to diagnose and retest the same cause, confirm public and origin versions agree, and check other affected template variants. Screaming Frog can help inventory URLs and resource defects; a crawl does not measure real-user performance.

GSC's Start Tracking begins a 28-day monitoring period. It does not force recrawling or provide a 48-hour fast-track review. Rolling field data can still include experiences from before deployment. If validation fails, inspect the remaining pages, conditions and served versions. Your own RUM can help identify a version-specific regression sooner, but its coverage and methodology may differ from CrUX.

Google's page-experience guidance confirms that Core Web Vitals are used by ranking systems while warning that good scores do not guarantee top rankings. Record release completion, field improvement, search clicks and qualified inquiries as separate outcomes.