What does p75 mean in Core Web Vitals?

The p75 is the value three in four page views match or beat. It decides whether a metric passes, and doesn't say how slow the slowest quarter is.

Arjen KarelArjen KarelUpdated November 22, 2026

What is the p75 in a Core Web Vitals report?

In short

  • p75 is short for the 75th percentile. Put all page views in order from fast to slow, and it's the value three quarters of the way along.
  • A metric passes when its p75 is at or under the good line. That's the same as at least three in four page views being good.
  • It doesn't show how much slower the other quarter of visits was. A quarter just past the line and a long tail of very slow ones can give the same p75.

PageSpeed Insights, Search Console and RUM tools such as CoreDash give each metric one number, and that number is the p75. Take 100 page views, sort them from the fastest LCP to the slowest, and the LCP of the 75th is your p75.

Google judges every metric at that point, separately for mobile and desktop.

What does a p75 say for LCP, INP and CLS?

The rule is the same for all three: three in four page views did at least as well as the p75. What changes is the value each page view gets.

An LCP p75 of 2.4 seconds means three in four page views had their largest element on screen within 2.4 seconds. Good is 2.5 seconds or less and poor is over 4 seconds, so 2.4 seconds passes with very little room.

A page view's INP is its slowest interaction, with one interaction in every 50 ignored as an outlier. A page view without a click, tap or keypress has no INP at all. So an INP p75 of 180 ms says that of the page views with an interaction, three in four had an INP of 180 ms or less. Good is 200 milliseconds or less, poor is over 500 milliseconds. On CoreDash's INP page, the line beside the title counts "page views with an interaction".

For CLS, a page view's value is its largest burst of layout shifts: shifts less than a second apart, five seconds at most. A CLS p75 of 0.08 means three in four page views had no burst bigger than 0.08. Good is 0.1 or less, poor is over 0.25.

How does the p75 relate to the share of good page views?

The p75 and the share of good page views are the same test. A p75 at or under the good line means at least 75% of page views are good, and the other way round.

Take 100 page views, sorted from fast to slow. If 79 of them are good, the 75th falls among the good ones and the metric passes. If only 68 are good, the 75th is already past the line and the metric doesn't pass. So wherever a report shows the share of good page views, that share tells you pass or fail on its own: 75% good or more is a pass.

CoreDash shows both on every metric card. The big number is the p75. The bar under it is the share of page views that were good, needs improvement and poor, and the dark marker on it is the p75.

The p75 sits at 75% of the bar. If at least three quarters of page views are good, it lands in the green.

Hover a part of the bar and you see its range, its share and how many page views it holds. On a card that doesn't pass, the marker falls in needs improvement or poor: hover that part to see the range your 75th page view is in.

Start with the p75 of each metric on mobile. That's also the view CoreDash opens on: Mobile, last 28 days, p75. From there, two cards on the LCP page go further than the bar: Distribution, with the page views behind it, and Sub-phases, with the part of the load that was slow.

Why does Google use the 75th percentile?

Google uses the 75th percentile because most page views should meet the target, and a few outliers shouldn't decide the result. web.dev gives the arithmetic in how the thresholds were defined: out of 100 page views, five bad ones are enough to move the p95, while the p75 only moves after 25.

In practice the page views that decide your p75 are the slower phones on mobile networks. The longer answer is on corewebvitals.io, under "Why the 75th percentile?" in What are the Core Web Vitals.

What does the distribution show that the p75 doesn't?

Read the distribution and you see how slow the slowest quarter of page views is. Two sites can have the same p75 while one has many page views just past the good line and the other a few very slow ones far out in a long tail. Only the distribution tells them apart.

For LCP, CoreDash's Distribution card counts page views per 200 ms. The two dashed lines are the good line and the poor line, and the dark line marked "p75" is your p75: the bars to the right of it are the quarter of page views the number doesn't describe. The share on the right is every page view above 8 seconds, where the scale ends.

When those bars bunch up just past the good line, many page views are a little slow. A small gain on all of them moves the p75, because the p75 only moves when page views on its slow side get faster than it.

When the bars thin out into a long tail, a few page views are very slow. Fixing them matters to the visitors who get them, but it barely moves the p75: those page views were already past it, and most will still be past it after the fix.

Either way, the next step is to split the view by device, page or country and see which group fills those bars. If one country fills the tail, that is a correlation: visitors in that country differ from the rest in more than where they are, such as their phones and their connections.

On my own site, corewebvitals.io, LCP on mobile over the last 28 days is 0.70 seconds at p75, with about 96% of page views good. Even the slowest quarter is mostly good: about four in five of those page views still have their LCP within 2.5 seconds, and most of the rest sit just past the line, between 2.6 and 4 seconds.

With a few dozen page views, the p75 itself can still move a lot. Under 50 page views CoreDash says so next to the count of page views, and still shows the number.

Why don't the sub-phase p75s add up to the LCP?

For one page view, the LCP sub-parts add up to its LCP exactly, so you can see how much of that load each phase took. Their p75s don't add up, because each one is taken over all page views on its own. The page views that are slow in one phase are not all the same ones that are slow in another.

Say a quarter of your page views get a slow server response and a different quarter download the LCP image slowly. The Time to First Byte p75 comes from the first group and the Load Time p75 from the second. Stack them and you get a page view that had both, which may never have happened. Chrome says the same about the subparts in CrUX: "the sum of the subparts aren't expected to equal the overall LCP 75th percentile" (CrUX methodology).

On corewebvitals.io, the four sub-phases of a mobile image LCP add up to 0.79 seconds at p75, against an LCP of 0.83 seconds. That's close by chance. Switch the Percentile to p90 and they add up to 1.36 seconds, more than the 1.15-second LCP.

The Sub-phases card in CoreDash shows each sub-phase as its own p75 in milliseconds, against its own good line. "Over good" says how far past that line it is. The switch above the rows splits image LCPs from text LCPs, because a text LCP has no Load Delay or Load Time.

Each row is the p75 of that sub-phase on its own, so the rows don't add up to the LCP above them.

Read each sub-phase against its own good line and start with the one furthest over it. The verdict on top names that phase. Time to First Byte is good at 800 ms or less and Load Delay at 400 ms or less, so two rows with about the same milliseconds can be one good and one poor. Each row's colour and its "Over good" show which.

The same goes for INP's three sub-phases, Input Delay, Processing Time and Presentation Delay, and for the five parts of TTFB. INP's good lines add up to its 200 ms on purpose. The p75s on the card still don't. What each LCP sub-phase is and how to shorten it is on corewebvitals.io, in Largest Contentful Paint.

Is a passing p75 enough, or should you look at p90 or p95 instead?

A passing p75 is all Google asks for: the p75 decides pass or fail. Keep it as your target and read p90 and p95 next to it. p90 is where you look for the slow tail.

Passing means three in four page views are good. The fourth is still a real visitor, and still worth fixing.

In CoreDash the Percentile select changes every card at once: p50, p75, p80, p90 or p95. The page views stay the same, and you read further along the same bar. On corewebvitals.io, mobile LCP goes from 0.70 seconds at p75 to 1.15 seconds at p90. When the number crosses the good or poor line, the status word changes with it, and the shares on the bar stay as they are.

One bar, three percentiles: the slow end only shows when you read further along it.

If you'd rather have CoreDash flag that tail, a project can judge its issues at p90 or p95 instead of p75.

What each metric measures and how to fix it is on corewebvitals.io: the Core Web Vitals, LCP, INP and CLS.

Sources: How the Core Web Vitals metrics thresholds were defined (web.dev), CrUX methodology: metrics (Chrome), Interaction to Next Paint and Cumulative Layout Shift (web.dev).

Track your first site free.No card needed. Setup takes about 5 minutes.

Start free trialBook a demo