Read the Metrics as User Outcomes
Core Web Vitals cover loading, responsiveness and visual stability. They should be reviewed at the 75th percentile and segmented by mobile and desktop. A site can pass one metric while a specific template or journey still feels slow.
| Signal | Investigation direction |
|---|---|
| Slow LCP | Identify the LCP element, then inspect server response, discovery delay, resource load and render delay. |
| Poor INP | Find slow real interactions and long main-thread tasks, event handlers and third-party script work. |
| High CLS | Locate unexpected shifts from unsized media, embeds, injected UI or font changes. |
| Slow TTFB | Review redirects, origin processing, database work, hosting capacity, caching and network path. |
| Lab good, field poor | Compare devices, networks, geography, logged-in states, consent scripts and real page variants. |
Performance Audit Workflow
- Choose representative pages. Include the homepage, primary landing pages, content, conversion and the slowest important template.
- Record the environment. Note device, network, location, cache state, authentication and consent state so results can be compared.
- Separate field from lab data. Use real-user evidence for impact and controlled runs for diagnosis.
- Trace the loading path. Inspect redirects, TTFB, critical requests, the LCP resource, render-blocking assets and cache behaviour.
- Profile interaction work. Reproduce slow clicks or inputs and find long tasks, layout work and scripts involved.
- Audit visual stability. Test initial load and post-load interactions for movement.
- Prioritise by impact and confidence. Fix work affecting important templates and many users before chasing tiny score gains.
- Release and verify. Compare equivalent lab tests, monitor field data over its collection window and protect against regression.
Typical High-Impact Fix Categories
Common work includes faster document response, correct full-page and object caching, early discovery and right-sizing of the main image, removal or deferral of unused scripts, smaller critical CSS and JavaScript, image dimensions, font strategy, and reducing third-party work. The right order depends on the measured bottleneck.
Adding optimisation products without mapping ownership can duplicate minification, caching or lazy loading and make regressions harder to isolate.
Audit Mistakes
- Optimising only the homepage when the commercial landing page uses another template.
- Comparing mobile and desktop scores as though they are the same test.
- Using one lab run as proof of field performance.
- Removing essential functionality for a higher score without testing the business journey.
- Ignoring server response because an image is the visible LCP element.
- Deploying several optimisation layers at once without a rollback.
Sources and Further Reading
Platform requirements and recommended practices change. These primary sources were checked when this guide was reviewed.
- web.dev: Web Vitals — primary documentation consulted for this guide.
- web.dev: Optimise LCP — primary documentation consulted for this guide.