Correction guide

Slow LCP: diagnose and improve main content rendering

By Antoine Le Louët · Estimated reading : 6 min · References

A slow LCP means the largest visible content arrives late. Find what holds that element back before changing your page.

Understand the measurement

Largest Contentful Paint marks the rendering of the largest eligible element in the viewport: an image, text block or video. Aim for 2.5 s or less at the 75th percentile of visits, separately for mobile and desktop; over 4 s is poor. Results between those thresholds need improvement.

A Lighthouse or Argos run is a laboratory measurement. Compare it with CrUX or RUM field data when available. Check whether the data describes this URL or the entire origin; missing data does not prove that the page is fast.

Find the part that is waiting

Record a page load in Chrome DevTools Performance. Select the LCP marker, record the element and, for an image, find its request in Network. The element can change with viewport size and content.

On WordPress, find the block, theme or cache to check

  1. Late HTML (high TTFB): inspect redirects, server processing and caching. Only test a shared cache when the response contains no private data.
  2. HTML received but late LCP request: look for a JavaScript-inserted image, a background discovered through CSS or deferred loading.
  3. Request started but slow: compare transferred size, dimensions, format and resource caching.
  4. Resource ready but element invisible: inspect blocking CSS, fonts, initial hiding and long JavaScript tasks. Start here for a text LCP.

Fix one cause at a time

For a main image identified as LCP, make its URL discoverable in the initial HTML. Remove loading="lazy" and test fetchpriority="high". Reserve this priority for the critical resource; preloading every image adds contention.

Serve appropriate dimensions using srcset/sizes and compare WebP/AVIF at acceptable quality. Reduce unnecessary scripts and blocking critical CSS. A smaller file cannot fix an element that stays hidden until a script finishes.

<img src="/hero.webp" width="960" height="540"
     fetchpriority="high" alt="Product presentation">

Measure before and after

Keep the same test URL, viewport, browser version, CPU/network throttling and cache state. Run at least five passes before and after; retain individual values and the median, then check the LCP element and delays in the trace.

The local example below uses only a demonstration page. It illustrates removing a discovery delay; its results do not predict your site’s improvement. Check afterwards for regressions in interactions, image quality and field data.

Read the Argos finding

The report exposes available browser measurements and associated Lighthouse findings, including runtime.lcp and lighthouse.audit.largest-contentful-paint. Values come from different runs and may vary. An unavailable module does not mean a zero LCP.

Argos cannot determine every delay’s cause on its own and does not collect your field percentile. After a fix, rerun the same page and compare it with your traces.

Measured before/after on a local demonstration

16 October 2026, integrated Chromium browser, 390 × 844 viewport, no CPU/network throttling, no-store responses over loopback. Same image and priority: before, JavaScript inserts it after an artificial 1,000 ms delay; after, it is in the initial HTML. Verified LCP element: #hero.

LCP in milliseconds, five independent page loads
VariantSamplesMedian
Before1060, 1052, 1048, 1060, 10481052 ms
After60, 52, 52, 56, 5656 ms

Reproduce from the repository: scripts/lcp-guide-demo.mjs, /before and /after routes. This test isolates an artificial delay; it measures neither a client domain nor Argos production performance. Server or rendering improvements would require another experiment.

Official references

References checked on 16 October 2026. Adapt and test the examples within your own scope.

Verify your fix

After your tests, rerun an audit of the same public URL and compare available findings.

Run a new audit