Core Web Vitals: What Each Metric Means and How to Fix It

Core Web Vitals measure three different things, with three different causes and three different sets of fixes. Treating them as one performance score is why optimisation effort frequently lands in the wrong place. This guide covers each, drawing on FISTA Solutions' web and mobile work.

What does each metric measure?

Three distinct aspects of experience, assessed at the seventy-fifth percentile of real visits.

Metric	What it captures
Largest contentful paint	When the main content becomes visible
Interaction to next paint	Responsiveness across the whole session
Cumulative layout shift	Unexpected movement while loading
Measured from	Real visits, not lab tests
Threshold	Seventy-fifth percentile
Segmented by	Device class and page type

How do you improve largest contentful paint?

Find the element first — it is usually a hero image, sometimes a heading, occasionally a video poster.

Then work through the causes in order: server response time, render-blocking resources, the element's own load time, and whether it is being lazy-loaded by accident.

The most common mistake is lazy-loading the largest contentful element because a blanket policy was applied. That single error can cost a second or more, and fixing it is a one-line change.

How do you improve interaction to next paint?

Reduce main-thread work. Long JavaScript tasks block the browser from responding, and the metric captures the worst interactions rather than the typical one.

Break long tasks, defer non-critical work, reduce the size of re-renders, and move computation off the main thread where possible. Event handlers doing heavy work synchronously are a common cause.

Measure on a mid-range device. A main thread that is never saturated on a development machine will be on the hardware most visitors actually use.

How do you eliminate layout shift?

Set explicit dimensions on images, videos, and embeds; reserve space for content that loads late; and use transforms rather than layout-changing properties for animation.

Font loading is a frequent cause: a fallback font at a different size shifts everything when the web font arrives. Matching fallback metrics or using a display strategy that avoids the swap addresses it.

Injected content — banners, consent notices, advertisements — that appears above existing content produces large shifts. Reserve the space rather than pushing content down.

Why does field data matter so much?

Because the assessment uses real visits. A lab score describes a controlled run on a fast machine; the field data describes what your visitors experienced on their devices and connections.

Those diverge substantially. A site scoring well in a lab and poorly in the field is the normal case, and optimising against the lab score in that situation produces no improvement to what is measured.

Use lab tools for diagnosing specific changes and field data for deciding what to work on.

Why the seventy-fifth percentile?

Because it captures the experience of a substantial minority rather than the typical visitor.

A site with a good median and a long tail fails. That tail is usually slower devices, poorer connections, or a specific page type, and finding which is what makes the improvement targeted rather than general.

Segment the field data by device class and page template. The problem is frequently concentrated rather than spread evenly.

How do these relate to search visibility?

They are part of how search evaluates page experience, alongside other signals, and their weight is modest relative to relevance and content quality.

That means they are worth fixing and they are not a substitute for having something worth finding. A fast page with poor content does not outrank a slower page that answers the question.

The stronger argument for fixing them is user behaviour: the same problems that produce poor scores produce abandonment, particularly on mobile.

What are the common mistakes?

Optimising from lab scores. Lazy-loading the LCP element. Measuring averages rather than the seventy-fifth percentile. Fixing layout shift by animating differently rather than reserving space. And treating the three metrics as one problem.

How do you test it?

Monitor field data continuously and use lab tools to verify specific fixes. Test on a mid-range device with a throttled connection rather than on a development machine.

Add regression checks to the pipeline for the metrics that matter most to your pages, because scores improved once and unmonitored return within a few releases.

What does it cost to operate?

Engineering time, mostly. The changes are typically small — image handling, dimensions, script loading strategy — and the diagnosis takes longer than the fix.

Ongoing cost is monitoring and budget enforcement, both of which are cheap and both of which lapse without an owner.

What should you measure?

Field values at the seventy-fifth percentile for each metric, segmented by device class and page template, plus the business metric the work is meant to affect. See web performance optimization guide.

What about AI-heavy interfaces?

They stress interaction to next paint and layout shift specifically. Streaming responses insert content continuously, which shifts layout unless space is reserved, and client-side processing competes for the main thread.

Reserve space for streamed content and keep model calls off the main thread. Both are straightforward when designed in and awkward afterwards. See streaming UI patterns for AI apps.

When is this the wrong approach?

These metrics are a poor priority for internal applications used on good connections by people who have no alternative. They matter where users can leave, which is most consumer and marketing contexts.

What should you do first?

Look at your field data segmented by device class and find which metric fails and on which pages. The answer is usually narrower than expected and it directs the whole effort.

How FISTA Solutions helps

FISTA Solutions builds and operates production systems through web and mobile, AI enablement, and staff augmentation: each metric diagnosed to its actual cause rather than treated as one score, field data segmented by device so effort lands where it matters, decisions documented with their reasoning, and handover that leaves your team able to maintain what was delivered. The record is 150+ projects for 50+ companies across 12+ countries.

To scope this work, message FISTA on WhatsApp, or read web performance optimization guide.