Core Web Vitals: measure real experience before fixing speed

A page can contain correct metadata and still load slowly or respond badly. Performance needs its own measurements, with a clear distinction between a test run and real visitors.

Know when a performance investigation is needed

Use this workflow when visitors report slow loading, delayed interactions or content that moves while they try to click. It is especially useful after adding large images, a new theme, analytics scripts or an embedded booking tool. Start with an important page and the device on which the problem occurs.

SingCompare’s static HTML diagnostic can inspect technical signals such as a viewport declaration. That is not a speed test, mobile usability certification or Core Web Vitals measurement. A passing technical finding cannot tell you how quickly a person can use the page. Capture separate performance evidence.

Understand the three metrics

Google’s web.dev documentation defines the following good thresholds. Evaluate the 75th percentile, separately for mobile and desktop, rather than substituting an average or the fastest run.

Documented good thresholds, reviewed 17 November 2026
MetricWhat it measuresGood threshold
LCPWhen the largest relevant visible content element renders2.5 seconds or less
INPResponsiveness to user interactions200 milliseconds or less
CLSUnexpected visual movement during the page lifecycle0.1 or less

A good assessment requires the relevant thresholds for all three metrics. These numbers describe user experience; they are not a promise of search traffic or a complete measure of business quality.

Collect a baseline with the right scope

  1. Run the URL in PageSpeed Insights. Save the URL, time, device category and result. The browser tool is available without activating a paid SingCompare AI service.
  2. Read the real-user section separately. PageSpeed’s methodology describes Chrome User Experience Report data from the preceding 28 days. Check whether the result describes this URL or the whole origin.
  3. Use the lab section to investigate. Lighthouse runs under controlled conditions and provides diagnostics. It is useful for reproducing issues; its score is not the same measurement as field Core Web Vitals.
  4. Check an actual journey. In browser developer tools, inspect loading and try the menu, filters and booking flow. A load-only test cannot reproduce every interaction that contributes to INP.

Some pages or sites lack sufficient real-user samples. Record “field data unavailable,” not zero milliseconds or a passing result. Origin-level data may describe many pages unlike the one you are changing.

Choose a fix that matches the evidence

For LCP, first identify the actual LCP element and what delays it. An oversized hero image, a slow server response and a resource discovered late need different changes. Resize and compress the appropriate image; avoid lazily loading an image that is immediately needed in the first screen.

For INP, investigate the interaction and long main-thread work. Reduce unnecessary JavaScript, split expensive work and review third-party code. For CLS, reserve space for images and embeds, and investigate late content insertion or font changes. Do not remove useful content merely to make one score look better.

In WordPress or another CMS, test a theme or plugin change in a staging copy first. In a custom application, use a versioned branch. Keep a rollback and verify that consent, navigation, accessibility and checkout still work. Changing several optimisation plugins at once makes cause and effect harder to understand.

Worked example: a moving booking button

Illustrative example, not SingCompare measurement: a service page loads a customer-review embed above its booking button. The embed arrives late and pushes the button down. A screen recording and performance trace show that insertion, while the current field CLS is 0.18.

The developer reserves the embed’s expected space, tests short and long content, and confirms in repeated lab runs that the button stays in place. This establishes that the reproduced layout problem improved. It does not immediately change the preceding 28 days of field data, and it does not prove every visitor’s CLS now passes.

Verify the experience, then revisit field evidence

Repeat the same lab setup after the change and retain both results. Compare several runs because network and device conditions vary. Test the real interactions that matter, including narrow screens and keyboard use. Recheck field data as the reporting window incorporates the updated experience.

Avoid reporting a Lighthouse score as an INP result, comparing mobile with desktop, or claiming an origin-level improvement for one specific page. Keep search performance separate: the measurement guide explains why a technical fix and a traffic change need different evidence. If the main problem is access or indexing, address that through the crawl and indexing workflow as well.

Sources and editorial approach

Primary documentation reviewed on . Examples are explicitly illustrative; they are not customer results or measured search data. Workflows are SingCompare’s practical recommendations. Provider documentation remains the authority for its own rules.

  1. Google web.dev: Web Vitals definitions and thresholds
  2. Google: About PageSpeed Insights, field data and lab data
  3. Google: PageSpeed Insights browser tool

Continue reading

Diagnose crawling and indexing problemsCheck the technical foundationsKeep performance and traffic evidence separate

Start with your website.

Get a free technical SEO and GEO readiness check, with findings you can inspect. No sign-in required.

Run free diagnostic