Making CLS fairer for long-lived pages
The Chrome Speed Metrics Team has finalized a change to the Cumulative Layout Shift (CLS) metric aimed at pages that stay open for a long time. After reviewing community feedback and running a large-scale analysis over millions of web pages, the team settled on a new definition: maximum session window with 1 second gap, capped at 5 seconds.
How the options were evaluated
The team implemented the leading candidate strategies in Chrome and compared how they scored real-world pages. Two findings stood out across all options tested:
- Every option reduced the correlation between time spent on a page and its layout shift score.
- No option produced a worse score for any page compared to the current metric.
Why a session window approach
Developer feedback strongly favored grouping layout shifts into session windows, where a window starts with the first shift and continues until a gap with no shifts occurs. When the next shift happens after that gap, a new window begins. Each window's score is the sum of the layout shifts it contains.
In the example above, three gaps without layout shifts produce three session windows. Based on the large-scale analysis, a 1 second gap between windows worked well.
Maximum vs. average window scores
Two summarization strategies were compared: averaging the scores of very large windows, or taking the maximum score across smaller capped windows. During testing, a problem emerged with the averaging approach. Some pages showed a pattern with one substantial burst of shifts followed much later by a single tiny shift in a second window.
With averaging, the small second window dragged the overall score down. But if a developer fixed just that one tiny shift, the score would be based only on the larger window and would nearly double — making a genuine improvement look like a regression. Since removing a layout shift obviously shouldn't worsen a page's score, the average strategy was abandoned in favor of maximum session windows.
Why the window is capped at 5 seconds
Two competing concerns shaped the window size decision. If windows are too short, slower page loads or delayed response to user interaction could split layout shifts across more windows, improving the score — effectively rewarding sluggishness. Windows needed to be long enough to avoid that outcome. At the same time, some pages produce a continuous stream of small shifts — a sports score page updating with each score change is the canonical example. Those shifts aren't increasingly annoying over time, so the score shouldn't grow without bound.
Comparing a range of window sizes across real-world pages led to the 5 second cap as the best balance.
Expected impact on CLS scores
Because the update caps a page's CLS, no page will see its score worsen. The analysis indicates that 55% of origins will see no change at the 75th percentile, since those pages either have no layout shifts today or their shifts already fall within a single session window.
The remaining origins will see improved scores at the 75th percentile, with most improving only slightly. Roughly 3% of origins will move from a "needs improvement" or "poor" rating to a "good" rating — pages that tend to use infinite scrollers or have frequent slow UI updates.
Trying out the updated metric
Tooling updates are in progress. In the meantime, the new CLS definition can be tested using example JavaScript implementations or a fork of the Web Vitals extension.
