Core Web Vitals are the three field measurements Google publishes as the pass marks for page experience: Largest Contentful Paint, Interaction to Next Paint and Cumulative Layout Shift. A page is rated good when at least 75 percent of real visits clear all three marks, which is why the number on your report is a percentile and not an average.
Every platform fact below comes from Google's own threshold, Chrome User Experience Report and Search documentation, as of October 2026. The thresholds have not moved. What trips an owner up is which of the two numbers on the free report they are reading.
The three measurements, and the number each has to beat
Three measurements make up Core Web Vitals, and each has a published good mark: Largest Contentful Paint at or under 2.5 seconds, Interaction to Next Paint at or under 200 milliseconds, and Cumulative Layout Shift at or under 0.1. Google's threshold documentation sets the poor marks at 4 seconds, 500 milliseconds and 0.25, as of October 2026.
| Measurement | What the visitor feels | Good | Poor |
|---|---|---|---|
| Largest Contentful Paint | How long before the biggest thing on the screen, usually the hero image or the headline, actually appears | 2.5 seconds or less | Over 4 seconds |
| Interaction to Next Paint | How long the page takes to visibly answer a tap on a button, a menu or a form field | 200 milliseconds or less | Over 500 milliseconds |
| Cumulative Layout Shift | How far the page jumps under a thumb as it loads, moving the button that was being aimed at | 0.1 or less | Over 0.25 |
Why the mark is a 75th percentile and not an average
The marks are judged at the 75th percentile of page views rather than at the average, which changes what a failing number means. Google's threshold documentation rates a site good on a measurement when at least 75 percent of page views meet the good mark, and poor when at least 25 percent of them land in the poor range.
Google's stated reason for choosing that percentile is that it is less likely to be affected by outliers. A 75th percentile Largest Contentful Paint of 4 seconds does not say the site is slow on average; it says a quarter of real visits were worse than 4 seconds. On a page taking 2,000 visits a month, that is 500 people, and an average hides every one of them.
What changed, and what did not
One measurement has changed since Core Web Vitals began: Interaction to Next Paint replaced First Input Delay in March 2024, which Google's own launch post announces and its page experience documentation repeats. The thresholds themselves have not changed. Google's threshold page still sets good at 2.5 seconds, 200 milliseconds and 0.1, as of October 2026.
The launch post also dates the handover, when an old audit turns up: Interaction to Next Paint arrived as an experimental Chrome metric in May 2022 and was announced in May 2023 as the coming replacement. A report citing First Input Delay was run before March 2024, whatever date is on its cover.
A tightened set of numbers also circulates, usually a 2 second Largest Contentful Paint and a 150 millisecond Interaction to Next Paint. Those are not the published marks. The check takes a minute: open Google's threshold page and read the table. A tighter number quoted without that page behind it is quoting another article.
Reading the free report: two tests, one verdict
PageSpeed Insights shows two sections, and one of them is Core Web Vitals. The top section is field data: real Chrome visits to the page over the previous 28 days, per Chrome's documentation. The lower section is Lighthouse, a single controlled test run on Google's own servers. The pass or fail verdict comes from the field section.
The score out of 100 that everyone screenshots belongs to the lower section. Chrome's documentation notes that PageSpeed Insights runs Lighthouse on separate servers, so its results can differ even from a run on your own laptop. A green score above a failing field section is normal rather than a contradiction: the lab machine had a better morning than your customers did.
Each section is good at one thing: the field data says whether the page passes, and the Lighthouse list says why, because it names the oversized image, the blocking script and the font swap.
Three ways the free report misleads an owner
Three readings account for most of the confusion. PageSpeed Insights falls back to origin level data when the specific URL is not in the dataset, shows No Data when neither the page nor the origin has enough traffic, and reports a 28 day window, so a fix shipped yesterday is averaged against the four weeks in front of it.
- The origin fallback. Chrome's documentation says PageSpeed Insights attempts to fall back to origin level data when it cannot find the URL in the Chrome User Experience Report. Read the label before the number, or the whole site's average gets taken for the one page that was just rebuilt.
- No Data is not a failure. When the page and the origin are both absent from the dataset, Chrome's documentation says the field section cannot be displayed and the Lighthouse result stands as an approximation. A page has to be sufficiently popular to enter the dataset, so a quiet service page often has nothing to show.
- The 28 day window. Chrome's documentation describes the data as a 28 day rolling average, notes that the collection period always reads 28 days even when less data exists, and says PageSpeed Insights updates daily on the previous 28 days. A fix is fully reflected about four weeks after it ships, not the next morning.
What Google says the thresholds do for ranking
Google's page experience documentation is careful about this. There is no single page experience signal. Core Web Vitals are used in ranking, but good scores in Search Console or a third party tool do not guarantee that pages will rank at the top, and the other page experience aspects do not directly help a site rank higher.
Which puts the honest order the other way round from how speed is usually sold. The 75th percentile is the reason: a failing field number says one visit in four was worse than the mark, on the phones and connections real customers have, and some of those people left.
What this looks like when it runs
Treating the three marks as build targets rather than as an audit means the numbers are set before launch and watched after it. Mirastart builds the site, the booking path and the reporting for Charlotte service businesses as one system: pages served already rendered rather than assembled by script, the booking step on the same screen as the promise, and one report in booked jobs.
The field numbers get checked on the pages that take bookings rather than on the homepage alone, because that is where the failing quarter costs money. Where a mark fails, the fix is sized against what it earns: an image budget, a rendered page and a reserved space for anything that loads late are cheap, and a redesign chasing a lab score is not.
Every account stays in your name, hosting and analytics included, with Mirastart added as a user you can remove. Ongoing engagements start at $3,000 a month, flat rather than a share of ad spend. The live availability booking on this site is ours, as is the online booking and repair status software Quick Auto NC runs its bays on.
Sources
- web.dev, "How the Core Web Vitals metrics thresholds were defined" - Source of the good marks used here (Largest Contentful Paint 2.5 seconds, Interaction to Next Paint 200 milliseconds, Cumulative Layout Shift 0.1), the poor marks (4 seconds, 500 milliseconds, 0.25), the 75th percentile basis, the rule that a site rates good when at least 75 percent of page views meet the good mark and poor when at least 25 percent land in the poor range, and the stated reason for the percentile, that it is less likely to be affected by outliers. Read via search result summary on October 10, 2026; web.dev is blocked to direct fetches from this environment.
- web.dev, "Interaction to Next Paint is officially a Core Web Vital" - Source for Interaction to Next Paint replacing First Input Delay as a Core Web Vital in March 2024, and for the dates around it: experimental in Chrome from May 2022, announced in May 2023 as the coming replacement. Read via search result summary on October 10, 2026. The exact day in March is reported by secondary sources only and is not claimed here.
- Chrome for Developers, "CrUX Tools" - Source for the Chrome User Experience Report being a 28 day rolling average, for the collection period always reading 28 days even where less data exists, and for the CrUX API and PageSpeed Insights updating daily on the previous 28 days. Read via search result summary on October 10, 2026.
- Chrome for Developers, "How to view Chrome UX Report data on PageSpeed Insights" - Source for the field section being real user data over the previous 28 days, for PageSpeed Insights attempting to fall back to origin level data when the URL is not in the dataset, for No Data appearing when neither is available with the Lighthouse lab result left as an approximation, for a page needing to be sufficiently popular to be in the dataset, for the indexing and status code audits as the discoverability check, and for PageSpeed Insights running Lighthouse on separate servers so results can differ from a local run. Read via search result summary on October 10, 2026.
- Google Search Central, "Understanding Google Page Experience" - Source for there being no single page experience signal, for Core Web Vitals being used in ranking while good scores in Search Console or a third party tool do not guarantee top rankings, for the other page experience aspects not directly helping a site rank higher, and for Interaction to Next Paint having replaced First Input Delay as of March 2024. The page does not quantify the effect of any threshold on rankings, and no such figure is claimed here. Read via search result summary on October 10, 2026.