Skip to the guide

Performance · Correction guide

High CLS: find what makes your page shift

Identify layout shifts, find their cause and verify the correction. A practical method with a before/after demonstration.

By Antoine Le Louët · Estimated reading : 7 min · References

Moving content can make an action miss its target

You start reading, then an image pushes the paragraph down. You aim for a button, but a banner takes its place. The Cumulative Layout Shift (CLS) measures this visual instability. It measures neither loading time nor interaction speed.

The value has no unit. The target is at most 0.1 at the 75th percentile of visits, separately for mobile and desktop. Above 0.25 the result is poor; between these thresholds, it needs improvement. The calculation uses the largest window of shifts, rather than adding up every movement during the visit. Official CLS definition.

An animation does not necessarily increase CLS. A new element must displace already visible content to cause a layout shift. Some movements near an interaction are excluded; a late insertion may still count.

Record when the content shifts

Choose a URL and a precise scenario: arriving on a product page, scrolling to reviews, or opening a variant. Record the viewport width, consent state and network conditions. A screen without a banner may not reproduce a first visit.

  1. Open the page in a private window without extensions, then open Chrome’s developer tools.
  2. In Performance, enable screenshots and record a reload. For a late-occurring issue, also record the journey after loading.
  3. Select an event in the Layout shifts. Comparez les captures immédiatement avant et après.
  4. Identify the first block added or resized upstream. The paragraph reported as displaced may be affected by the problem without causing it.
  5. Keep the trace, URL, scenario and an annotated screenshot. They let someone else reproduce the issue.

If the issue only appears on a slow connection, replay the scenario with consistent network throttling. Also test a narrow screen: a heading or banner may take an extra line. Record a trace in DevTools.

Fix the responsible block rather than the displaced content

On a small screen, scroll the table horizontally.

Start with the observation to choose a correction
ObservationWhat to examineCorrection to test
Text moves down when an image arrivesMissing dimensions or an inconsistent aspect ratio.Declare intrinsic width and height; preserve the aspect ratio in responsive CSS.
A widget pushes down the product pageAn empty area before reviews, advertising or an integration.Reserve a suitable space from the first render.
The heading changes sizeA fallback font is replaced by the final font.Compare font metrics and adjust the fallback font.
The page jumps after a few secondsA banner or data inserted above the content.Reserve space before the content arrives; test real variants.

For a 1,200 × 800 pixel image, the browser can reserve a 3:2 aspect ratio before receiving the file:

<img className={styles.photo}
     src="/atelier.webp" width="1200" height="800"
     alt="Établi de réparation de vélos" />

/* Dans le module SCSS du composant */
.photo {
  display: block;
  width: 100%;
  height: auto;
}

In this JSX example, styles refers to the SCSS module imported by the component. The dimensions must match the image’s actual aspect ratio. loading="lazy" does not reserve space on its own. For a widget, choose a suitable height for each viewport width and also check what happens when it returns nothing. Detailed causes and corrections.

For fonts, font-display: swap makes text available sooner but does not guarantee stability. A similar fallback font, possibly adjusted with size-adjust and the metrics for @font-face, reduces differences in geometry. Also measure long headings and the available languages. Font best practices.

Demonstration: reserve space for a late-loading block

The two pages below display the same block after an artificial delay of 1.2 seconds. On the first page, the block pushes the text down. On the second, a 128-pixel space exists from the initial render.

Before: no space reserved

The text initially occupies the area where the block will appear.

Open the before demonstration (new tab)

After: space reserved

The text stays in place while the content arrives.

Open the after demonstration (new tab)

Record each page in DevTools with the same viewport. These examples isolate an insertion; they do not present customer scores, traffic gains or production results. The fixed height suits this demonstration block, not every kind of content.

A fix must hold throughout the journey

Replay the same scenario several times before and after, under the same conditions. Check that the targeted event disappears and that the change does not hide content, clip enlarged text or impair interactions.

Then compare field data where available. A Lighthouse loading audit may miss a shift caused later by a widget or scrolling. In PageSpeed Insights, check whether the data describes this URL or the entire origin. L’absence de données terrain ne signifie pas que le problème est résolu.

In your fix ticket, keep the URL, viewport width, scenario, before trace, changed component, after trace and mobile test results. This evidence is more useful than an isolated screenshot of an overall score.

What Argos contributes to the diagnosis

Argos gathers performance signals and the Lighthouse score when the service is available. They provide context for choosing checks. A good overall score does not prove that your user journey is stable.

To isolate a CLS issue, use the browser trace and supplement it with field data. Argos does not measure your visitors’ 75th percentile or replay all their journeys. After fixing the issue, run the same URL again and compare the available findings with your manual evidence.

When publishing, include this check in the acceptance checks before and after a redesign.

Official references

Sources consulted on 2 November 2026. Service interfaces and rules may change.

Check your page with Argos

Run a free audit of a public URL and examine the available evidence. Automated checks complement the checks in this guide.

Analyse a page for free