Core Web Vitals FAQ on percentiles

For each metric, visitors should have an experience below the following boundaries:

• Largest Contentful Paint should be below 2.5 seconds;
• First Input Delay should be below 100 milliseconds;
• Cumulative Layout Shift should be below 10% (0.1).

To create the three buckets, being green for good experience, orange for needing improvement and red for poor experience, Google looked at their very own CrUX data as well as user experience researches to come up with a threshold or bucket distribution. They also analyzed what the percentage of different amount of (in case of the LCP) seconds was. See also Defining the Core Web Vitals on web.dev website.

Luckily: no. There might always be visitors under really poor circumstances, when it comes to device capabilities and internet connectivity. For example when having an international website or webshop, users may come from different regions, working with different budgets and thus different smartphone devices.

As a result, at least 75% of visitors should have an experience within the three metric boundaries to be able to pass the Core Web Vitals. To put it mathematically: the 75th percentile of the data of each metrics, should be within the green thresholds.

Although PageSpeed Insights was using the 50th percentile (being the median) when using Lighthouse v4, Google changed this to the the 75th percentile over time to make sure that pages work well for the majority of users. By focusing on 75th percentile values for our metrics, it is ensured that pages provide a good user experience under the most difficult device and network conditions and developers can understand the most frustrating user experiences on their site

Google did use a higher percentile (90th percentile) for a short amount of time for some metrics. However, as faster websites and webshops are likely to increasingly attract more users under more difficult conditions, this could skew the data too fast. As a result, the 75th percentile is being used so data will not be overly impacted by outliers.