Website performance work starts with measurement, not with a particular plugin or technology. First establish whether real users experience slow loading, delayed interactions or an unstable layout, then fix the largest bottleneck.
Separate real-user data from a laboratory test
Where enough observations are available, PageSpeed Insights presents both field data collected from Chrome users and laboratory data produced in a controlled environment. Field data describes the experience of real visitors across devices and connections. A laboratory test is useful for reproducing a problem and investigating its technical causes.
A single test run is not a final verdict. Results vary with server load, test location, caching and third-party services. Assess the most important page types separately, check mobile and desktop, and compare changes over a suitable period. Search Console's Core Web Vitals report can reveal groups of similar pages affected by the same issue.
The current Core Web Vitals are LCP, INP and CLS
Core Web Vitals describe loading, interaction responsiveness and visual stability. The good threshold is evaluated at the 75th percentile separately for mobile and desktop visits, meaning that at least 75 percent of measured page loads should meet the target. INP replaced the former FID metric in 2024, and TTI is not part of the current Core Web Vitals set.
Use the metrics as a starting point for diagnosis, not as the only definition of website quality. Google says Core Web Vitals are used in its page-experience assessment, but good scores do not guarantee high rankings. Helpful content, clear journeys, accessibility and commercial outcomes need to be reviewed alongside performance.
| Metric | Good threshold | What it describes |
|---|---|---|
| LCP | 2.5 s or less | loading of the largest visible content element |
| INP | 200 ms or less | responsiveness to user interaction |
| CLS | 0.1 or less | unexpected movement of visible content |
Fix the LCP bottleneck at the right point in the loading chain
A slow LCP can begin before the browser receives the first HTML response. Review redirects, server response time, caching and geographic distance before endlessly compressing images. If the server responds quickly but the main content is discovered late, the problem may be render-blocking CSS, JavaScript, a font or a key image that is only revealed through a script.
A hero image should be delivered at an appropriate size and in an efficient format, but not every image needs identical handling. Below-the-fold images can usually load lazily, whereas lazy-loading the LCP image may delay it. Use preloading selectively for genuinely critical resources so that less important content does not compete for bandwidth.
- remove unnecessary redirects and reduce server response time
- serve responsive images at suitable dimensions and in efficient formats
- discover the key image and font early, then load other resources as needed
- apply long-lived caching to versioned static assets
INP improves when the main thread is available to the user
INP measures the responsiveness of clicks, taps and keyboard interactions throughout a visit. Poor results often arise from long JavaScript tasks, heavy client-side rendering or third-party scripts. The issue may not appear during initial loading and only becomes visible when someone uses a menu, form or filter.
Remove unused code, divide expensive work into smaller tasks and avoid unnecessary rerendering of whole views. Advertising, analytics, chat and personalisation scripts all consume resources on the visitor's device, so evaluate each against its value. Test important journeys on a slower mobile device and verify that every action receives prompt visible feedback.
CLS improves when space is reserved before content arrives
Images, videos, advertising slots and embeds need predefined dimensions or an aspect ratio so that surrounding content does not move when they load. A cookie banner, notification or validation error should not unexpectedly push content that is already visible. Reserve space for dynamic elements or reveal them in a way that does not move the user's target mid-action.
A web font can alter text dimensions when it replaces a fallback. Load only the necessary styles, choose a fallback with compatible metrics and define the font-loading behaviour. Check CLS in field data because a laboratory test may not capture shifts that occur later in a visit.
Prioritise improvements by business impact
Start with pages that receive substantial traffic or perform an important commercial task, and with issues that affect a shared template. Correcting one image, font or component pattern may improve hundreds of URLs. Record a baseline before the change and confirm afterwards that the improvement appears in both technical measurement and the user journey.
Performance can decline as new images, tracking tags, embeds and features are added. Agree performance budgets, automated checks and clear ownership for findings. Monthly monitoring is useful, but a critical regression should be caught during release rather than in the next report.
When you want performance built into the design, implementation and ongoing development of your website, explore our business website service.
Sources and further reading
The metrics, recommended thresholds and search-related qualifications are based on current guidance from Google and web.dev.
