Core Web Vitals for Business

What each metric is actually measuring

LCP: Fix the Main Element

Largest Contentful Paint reports when the largest image, text block or video visible in the viewport is rendered. Redirect time, connection setup and server response can all contribute before the browser reaches that element.

A useful LCP investigation follows the request path:

1. Server response: was the HTML delayed?
2. Resource discovery: did the browser find the hero image or font immediately?
3. Download: is the media correctly sized and compressed?
4. Render delay: did CSS, JavaScript or client-only rendering hold it back?

Do not install a generic speed plugin before identifying the LCP element. A page can have a small image and still fail because that image is discovered late.

INP: Test Valuable Interactions

Interaction to Next Paint measures responsiveness across clicks, taps and keyboard interactions during a visit. The final value is based on the slowest qualifying interaction, with outlier handling.

Start with journeys that matter commercially:

• opening navigation on mobile;
• changing a product option;
• adding to basket;
• submitting a form;
• opening account or checkout controls.

If a widget or old marketing tag blocks the main thread, removing it can be more valuable than optimising application code around it. Record the interaction, page, device class and release date so the fix can be verified in later field data.

CLS: Reserve Content Space

CLS is unitless. It reflects how much visible content moved and how far. Common failures include images without dimensions, ads or embeds without reserved space, late banners and web-font swaps.

Field and Lab Data Differ

Field data describes real visits across devices, connections and page states. Lab data gives a controlled reproduction that helps engineering diagnose a problem. Use field data to decide whether users are affected; use lab tooling and traces to find the cause.

1. 01
   Confirm the failing metric

   Use the page group, device split, percentile and sample context.

2. 02
   Identify the exact element or interaction

   A score without a target is not an engineering task.

3. 03
   Ship one attributable change

   Record the deployment date and avoid bundling unrelated redesign work.

4. 04
   Verify both lab and field movement

   Confirm the mechanism improved immediately, then watch real-user distribution.

Priority rule

Fix Failed Customer Journeys

Start with poor mobile pages that support enquiries, purchases or sign-in. A passing brochure page does not outrank a failing checkout because its traffic number is larger.

See performance-led website delivery

Primary sources

1. web.dev: Core Web Vitals and the 75th-percentile ruleweb.dev
2. web.dev: LCP thresholds and diagnostic breakdownweb.dev
3. web.dev: INP thresholds and optimisationweb.dev
4. web.dev: CLS thresholds and common causesweb.dev