Passing Core Web Vitals is worth doing — for rankings, and because slow sites lose money. But the metrics measure what’s measurable, not everything users feel. Optimising to the number while ignoring perceived experience leaves conversion on the table.
Build a journey map beside the performance report
Choose the pages that start a meaningful journey: a service page, a product category, a booking page and the form or checkout that finishes it. For each, record the main task, the first element a visitor needs, the point where the page becomes usable and the most common exit. A performance score alone does not tell you whether the task works. A fast hero followed by a confusing form can still lose the enquiry.
A practical review combines field performance data with a manual mobile test. Open the page on a typical connection, try to read the first answer, tap the primary action and complete the form. Watch for layout shifts that move a button while someone is touching it, overlays that block content, and delayed scripts that make an apparently loaded page unresponsive. Record each issue as a reproducible step rather than a vague complaint that the page feels slow.
Prioritise by the number of affected journeys. If the same image component slows every service page, a template fix can help the whole commercial section. If one decorative animation affects a rarely visited guide, it can wait. Keep the business goal in view, but do not claim that a faster page caused more sales without a suitable comparison.
Test one change at a time where possible
For Largest Contentful Paint, identify the actual largest element in field and lab reports, then check its image dimensions, loading priority, server response and render path. For Interaction to Next Paint, inspect slow interactions and the JavaScript work around them. For Cumulative Layout Shift, reserve space for images, fonts, ads and embedded elements. These are diagnostic steps, not a promise that every site will hit the same threshold after one edit.
After a change, compare the same page type and device group. Field data takes time to move and can vary with traffic and device mix, so keep a dated release log. Use a conversion measure such as completed enquiries or checkout starts alongside the technical metric. If performance improves but the action rate does not, test the copy, offer and form friction instead of assuming the speed work was wasted.
Our own website uses static pages and a small set of shared templates. That makes a template-level image or script problem more important than a single-page illustration issue. It also means the route should be checked live after deployment: a green build says the code compiled, while a live request confirms that the asset and page reached the CDN.
The question is not "when did the largest element paint?" but "when could the user do the thing they came to do?"
What the three metrics actually capture
LCP (Largest Contentful Paint) measures when the biggest element renders — a proxy for "is it loading?". INP (Interaction to Next Paint) measures response time to taps and clicks — "is it listening?". CLS (Cumulative Layout Shift) measures visual stability — "can I trust where things are?". Google evaluates all three from real Chrome users, not lab tests.
These are good proxies, and the field thresholds (2.5s / 200ms / 0.1) are sensible. The mistake is treating "passing" as "fast enough to convert" — the metrics are floor, not ceiling.
Perceived speed diverges from measured speed
A page can pass LCP while feeling broken: the hero image painted quickly, but the price, the reviews, and the add-to-cart button straggled in afterward. Users judge readiness by the elements they came for, and no metric knows which elements those are.
Skeleton screens, prioritised rendering of decision-critical content, and progressive image loading shape perception beyond what the numbers capture. The question to optimise is not "when did the largest element paint?" but "when could the user do the thing they came to do?"
Interaction friction the metrics miss
INP catches slow event handlers; it doesn’t catch touch targets too small for thumbs, forms that fight mobile keyboards (wrong input types, no autocomplete), date pickers that require surgical precision, or the cookie-consent-newsletter-appstore banner stack that buries the actual page.
These frictions routinely cost more conversions than a 300ms INP ever will. Session recordings and rage-click analysis find them; Lighthouse never will.
Layout stability after the measurement window
CLS is scored on shifts during loading, but trust-destroying movement often happens later: personalisation banners that push content down on scroll, late-loading recommendation widgets, chat bubbles that cover the checkout button on exactly the screen sizes your analytics says matter.
The rule is reserved space, always: every dynamically injected element gets explicit dimensions before it loads. Users’ motor memory is part of your UX — betraying it at the moment of click is uniquely expensive.
Connect performance work to revenue, not scores
Segment conversion rate by real-user LCP/INP buckets in your analytics: the revenue curve across speed cohorts makes the business case better than any audit score, and shows where returns diminish. Many sites find conversion flattens past a threshold — spend the remaining engineering budget on the frictions above instead.
Report both numbers: the Vitals pass rate for the ranking story, and the speed-cohort conversion delta for the money story. When the two disagree about priorities, the money story wins.
The useful takeaway
Pass the Vitals for the floor, then optimise what they can’t see: decision-critical rendering order, mobile interaction friction, post-load stability, and the speed-to-conversion curve from your own field data. The metrics are instruments — the experience is the target.
