Technical SEO

Core Web Vitals 2026: How to Measure and Improve LCP, INP and CLS on a Business Website

A practical method to measure LCP, INP and CLS, diagnose causes and verify business website improvements.

Core Web Vitals 2026: How to Measure and Improve LCP, INP and CLS on a Business Website

Core Web Vitals should be assessed using two types of data: measurements from real visitors, which show user experience over a longer period, and laboratory tests, which help reproduce a problem and identify its technical cause. For a business website, the aim is to achieve a ‘good’ LCP, INP and CLS level at the 75th percentile. However, optimisation should not begin by implementing random PageSpeed recommendations. First identify the problematic page type, device group and component, then address the shared cause and verify the result both in the laboratory and after new field data have accumulated.

What Core Web Vitals Measure in 2026

In 2026, Core Web Vitals comprise three user experience metrics covering loading, responsiveness and visual stability. They are not a general ‘website speed score’. Each metric has its own mechanism, so the methods used to improve them also differ.

Metric

What it describes

Good level

Poor level

LCP — Largest Contentful Paint

When the largest significant image or text element appears in the viewport

up to 2.5 s

over 4.0 s

INP — Interaction to Next Paint

How quickly the page displays the next visual result after a click, tap or keyboard action

up to 200 ms

over 500 ms

CLS — Cumulative Layout Shift

How much visible content moves unexpectedly

up to 0.1

over 0.25

Between the good and poor levels is the ‘needs improvement’ range. Assessment uses the 75th percentile: a fast experience for the average user is not enough if a significant proportion of visitors still experience a slow or unstable page. Mobile and desktop devices should be analysed separately because processor performance, network conditions, viewport size and page layout differ.

Search engines use Core Web Vitals as one page experience signal, not as a guarantee of rankings. A good result does not replace relevant content, clear information architecture or alignment with search intent.

Why One PageSpeed Insights Test Is Not Enough

Measurement tools answer different questions. They should not be treated as interchangeable.

Field data describe the experience of real Chrome users. PageSpeed Insights may show Chrome UX Report, or CrUX, data for a specific URL or an entire origin when sufficient data are available. These data cover a rolling 28-day period, so a newly published improvement will not appear immediately.

The Google Search Console Core Web Vitals report helps identify problematic groups of similar URLs. It is suitable for determining scale and priority, but it usually does not identify the precise line of code or component causing the problem.

Laboratory data are collected in a controlled test. Lighthouse and Chrome DevTools make it possible to repeat a page load and analyse network requests, main-thread activity and layout shifts. A single test result is affected by the selected device, network simulation, server state and cache, so diagnosis requires several comparable measurements.

A laboratory test cannot fully measure INP because it does not include real user interactions across the page’s entire lifecycle. Lighthouse can use Total Blocking Time, or TBT, as a diagnostic signal for main-thread load, but TBT is not a substitute for INP.

Your own real user monitoring, or RUM, is useful when you need visibility into specific page types, devices, versions or interactions. Google’s web-vitals JavaScript library can collect these measurements and diagnostic attributes. Data should be sent without personally identifiable information and in accordance with the website’s privacy and consent procedures.

Establish a Measurement Baseline

Before making changes, prepare a short baseline report. It does not need to include every URL; page templates and business-critical actions are more important.

It is advisable to separate:

  •          the homepage, service pages, articles, categories and product pages;
  •          mobile and desktop traffic;
  •          new visits and situations where the cache has already been used;
  •          key actions such as opening the menu, using a filter, starting a form and adding an item to the basket;
  •          first-party code and third-party scripts.

For each problematic template, record the field metric, laboratory test configuration, suspected cause and change version. Without this record, the team may compare tests conducted under different conditions or attribute the result to the wrong change.

How to Diagnose and Improve LCP

An LCP investigation should examine the entire loading chain. Assuming that the only problem is an ‘image that is too large’ may be too narrow. First, use DevTools or PageSpeed Insights to check which element became the LCP element in that particular test. Depending on the viewport, it may be the main image, heading block or other content.

Next, determine where time is being lost:

  •          the server is slow to prepare the initial HTML;
  •          the browser discovers the LCP resource too late;
  •          the image or font takes too long to download;
  •          the resource has arrived, but rendering is delayed by CSS, JavaScript or client-side execution.

If the problem lies in the server response, check the cache, database operations, CDN and page generation. If the main image is added only after JavaScript executes, the browser begins requesting it too late. An LCP image should generally not be loaded with loading="lazy". Improvements may come from using the correct image format, a suitably sized file, srcset, early discovery in HTML and justified use of fetchpriority="high".

preload is not a universal solution. Excessive preloading competes with other critical resources for bandwidth. Priority should be given only to the resource genuinely needed in the initial view of that specific template.

How to Diagnose and Improve INP

An INP problem occurs when a user action is followed by a long input delay, slow event processing or delayed presentation of the next frame. The first step is therefore to identify the specific interaction rather than stopping at the page level: opening a menu, changing a filter, using a calculator or processing a form field.

In a Chrome DevTools Performance recording, look for long main-thread tasks and check which script caused them. Typical causes include a large JavaScript bundle, synchronous third-party code, a complex event handler, an excessively large DOM or repeated style and layout calculations.

A practical sequence is:

  •          do not load functions initially if they are only needed later;
  •          break up long JavaScript tasks and give the browser an opportunity to render the next frame;
  •          reduce the work performed in the event handler;
  •          avoid unnecessarily re-rendering the entire component tree;
  •          audit chat, analytics, advertising and other third-party scripts;
  •          move intensive calculations off the main thread only when they are not directly related to DOM processing.

Reducing the size of a JavaScript file does not by itself guarantee good INP. A small script may perform expensive work on every click, while a larger file may not interfere with interactions after loading. Optimisation should target the measured cause of the delay.

How to Diagnose and Reduce CLS

CLS measures unexpected layout shifts throughout the page lifecycle. They can also occur after the initial load. A laboratory test may therefore miss a shift caused by the appearance of a banner, form error message or content loaded later.

The most common fixes are practical:

  •          specify dimensions or aspect-ratio for images and videos;
  •          reserve space in advance for advertisements, maps and embedded content;
  •          do not insert new content above content that is already visible without an action the user can understand;
  •          choose an appropriate font-loading strategy and a similar fallback font;
  •          use transform for movement where possible rather than properties that trigger layout recalculation;
  •          provide stable space for validation messages and dynamic blocks.

DevTools can show layout-shift records and the elements involved. Fix the element causing the movement, which is not necessarily the element the user sees moving.

How to Prioritise Improvements

Start by fixing a shared component that affects many important URLs. For example, one change to the header, hero image, product grid or consent solution may be more valuable than polishing a single low-traffic page.

Priority can be determined using four questions:

  1. What proportion of traffic or URLs is affected?
  2. How far is the metric from the good threshold?
  3. Does the problem affect a business-critical action?
  4. How confidently has the cause been demonstrated, and what is the risk of the fix?

Do not optimise solely for the overall Lighthouse score. A change that improves the laboratory score but harms accessibility, form functionality, analytics quality or the conversion journey is not a good business outcome. A performance budget should be tied to specific metrics and templates.

How to Verify the Result After Publication

Before and after the fix, use the same test device, network mode, page state and measurement scenario. Save the Performance recording or Lighthouse report so that you can verify the final figure and understand how the underlying mechanism changed.

After publication, check:

  •          whether the diagnosed request, long task or layout shift has disappeared;
  •          whether another Core Web Vitals metric has deteriorated;
  •          whether key functions work on mobile devices;
  •          whether the new version behaves as expected in RUM data;
  •          whether the 75th percentile and the status of affected URL groups improve after the field data are refreshed.

CrUX’s 28-day window means that the field result changes gradually. This is not a reason to wait without monitoring: laboratory tests and your own RUM data can show earlier whether the specific cause has been resolved.

Limitations and Risks

CrUX does not represent every single visit. Lower-traffic pages may have no URL-level data, while an origin-level result may conceal differences between a fast homepage and a slow product or service template. The distribution of users’ devices, networks and geographical locations also affects the observed result.

For single-page applications, establish how measurements will be attributed to virtual routes in the RUM system. The behaviour of third-party scripts may vary according to the user’s consent choice, advertising content or the response time of an external service. These conditions cannot be controlled completely, but they can be segmented, limited and monitored.

Aggressive caching can serve outdated content, incorrect deferred loading can hide an important function, and excessive script delays can disrupt analytics or interactions. Every fix should be assessed alongside functional testing, accessibility, privacy and content freshness.

Core Web Vitals definitions, tools and recommendations may evolve. Before including thresholds in a contract or long-term KPI, check the current Google and Chrome documentation. Even meeting all three good thresholds does not by itself guarantee either search rankings or sales results.

Short Implementation Checklist

  •          Select business-critical page templates and interactions.
  •          Record the CrUX, Search Console and laboratory baselines.
  •          For each weak metric, identify a specific element, action or component.
  •          Fix shared causes with a large impact first.
  •          Verify the fix under identical laboratory conditions.
  •          After publication, monitor RUM and the 28-day field trend.
  •          Include performance testing in the acceptance process for future releases.

Managing Core Web Vitals is not a one-off speed initiative. A reliable system connects real-user data, repeatable technical diagnosis and release control. This keeps the team focused on causes that genuinely affect users of the business website and prevents the score from a single test becoming an end in itself.