Core Web Vitals, fixed at the cause.

• Choose Core Web Vitals optimization when field data fails.

  If Search Console or PageSpeed Insights field data shows URLs as poor or needing improvement, the problem is measurable, and this is the work that addresses it.

• Choose technical SEO when pages are missing from the index.

  A slow page can still be indexed and ranked. When pages are not being indexed at all, crawling and rendering come first, and speed work follows.

• Choose a rebuild when the platform itself is the bottleneck.

  When a theme or framework ships more JavaScript than any tuning can remove, website redesign and modernization is the honest answer, informed by these measurements.

PageSpeed Insights shows two different things: field data from the Chrome User Experience Report and a Lighthouse lab test run once on a simulated device. Search Console uses the same field data but groups similar URLs together. A high lab score beside failing field data means real visits differ from the test in device, network, cache state or interaction.

PageSpeed Insights passes the Core Web Vitals assessment when the 75th percentile of all three metrics is good over the previous 28-day collection period. Lighthouse produces a separate score weighted across five lab metrics, so a perfect Lighthouse score does not mean the field assessment passes.

Search Console groups URLs that look similar and reports each group by its poorest metric. Pages with too little traffic may have no field data of their own, because the Chrome User Experience Report only includes pages that are publicly discoverable and sufficiently popular, and Search Console then reports them within a broader origin group.

Largest Contentful Paint is slowed in one or more of four parts: the server responds late, the browser discovers the LCP resource late, the resource takes long to download, or rendering is delayed after it arrives. As a rough guide, most of the time should go to server response and download, with discovery and render delay kept small.

• Time to First Byte: the server, redirects and caching.
• Resource load delay: an LCP image found late, for example only in CSS or JavaScript.
• Resource load duration: file size, image format and delivery distance.
• Element render delay: render-blocking CSS, JavaScript or fonts after the file arrives.

Interaction to Next Paint is slow when the main thread is busy. Each interaction has three phases: input delay while earlier tasks finish, processing time for the event handlers, and presentation delay before the next frame is painted. Any task longer than 50 milliseconds counts as a long task, and long tasks hold interactions back.

INP observes the latency of clicks, taps and key presses throughout a visit and, for most pages, reports the interaction with the worst latency, so one heavy handler on one control can fail the metric for the whole page. On pages with many interactions, the calculation ignores one of the highest for every 50. Hovering, zooming and scrolling are not counted. Good is 200 milliseconds or less, and anything above 500 milliseconds is poor.

Common fixes include splitting the work in event handlers into separate tasks, doing as little work in each handler as possible, deferring work that is not needed before the next frame, and keeping the DOM small so rendering after an update stays quick. Third-party tags are part of the review, because their tasks run on the same main thread as the site.

Layout shift happens when visible content moves after it has painted without the visitor causing it. The most common causes are images, ads, embeds and iframes without dimensions, dynamically injected content, and web fonts that render at a different size from their fallback. Reserving space and adjusting fallback font metrics are among the standard fixes.

• Explicit width and height, or an aspect ratio, on every image, video and iframe.
• Reserved space for embeds, ads and banners before they load.
• No content inserted above what the visitor is already reading.
• Fallback fonts adjusted with size-adjust and ascent and descent overrides.
• Animations that use transform and opacity instead of layout properties.

Fix the metric that fails on the page types with the most traffic first, starting with causes that affect every page built from the same template. A fix in a shared template, font setup or tag manager reaches every URL built on it, while a fix on a single page reaches only that page.

Within a failing page type, the work is ordered by how directly each cause moves the 75th percentile: for LCP, the part of the timing that takes longest; for INP, the slowest interactions visitors actually use; for CLS, the largest shifts.

Work that improves a lab score without touching the failing field metric is postponed, however easy it looks.

• Core Web Vitals optimization
• Largest Contentful Paint diagnosis
• Interaction to Next Paint diagnosis
• Cumulative Layout Shift elimination
• Field and lab measurement

Every fix follows the same four steps, from field data back to field data. Skipping the first step is how a team spends weeks improving a lab score while the metric that fails stays exactly where it was.

• Work aimed at the metric that fails

  Field data shows which metric fails on which page type, so effort goes to the real problem instead of a lab score.

• Causes fixed, not symptoms

  Attribution names the element, request or script responsible, so the fix holds instead of trading one metric against another.

• Pages that load, respond and hold still

  Content appears sooner, controls answer sooner and nothing jumps under a finger or a cursor, which is exactly what the three metrics describe.

• Results that survive the next release

  A budget and a release check catch regressions before they reach visitors, rather than after the field data drops again.

• A marketing site with a slow hero

  The largest element is a hero image discovered late. Loading it early, sizing it correctly and removing render-blocking work brings LCP forward.

• A store with heavy third-party tags

  Tag managers, review widgets and chat scripts create long tasks. Deferring, removing or rescheduling them is what moves INP.

• A content site with shifting layouts

  Embeds and web fonts push text down after paint. Reserved space and adjusted fallback font metrics remove the shift.

• Field data from the Chrome User Experience Report, read through Search Console and PageSpeed Insights, is the measure of success; lab tools find and test causes.
• A Lighthouse page load involves no user input, so it cannot measure Interaction to Next Paint; responsiveness is investigated through Total Blocking Time in the lab and real interactions in DevTools.
• The LCP element is identified before anything is optimized, because it can be a heading or a block of text rather than the image a team expects.
• Images carry explicit width and height and modern formats such as WebP or AVIF, and only the LCP image is loaded with priority, never lazily.
• Web fonts are self-hosted, preloaded where they are critical, and paired with fallback faces whose metrics are adjusted to the real font.
• Server response is its own subject: as a rough guide, most sites should aim for a Time to First Byte of 0.8 seconds or less, and TTFB is not a Core Web Vital.