statisticspercentileswebsite auditscore web vitalstechnical seo

Percentile Meaning: Definition, Examples, Formula, and Website Audit Uses

Percentile meaning is the relative position of a value in a dataset, helping website teams see typical and worst-case performance instead of relying on a single average.

· 13 min read

A website can have an average Largest Contentful Paint of 1.8 seconds yet still fail a 75th-percentile assessment if enough real visitors have slower experiences. Understanding percentile meaning helps SEO teams, agencies, and site owners interpret performance data accurately, prioritize the pages affecting more users, and avoid drawing the wrong conclusion from an average.

A percentile is a position within an ordered dataset. It answers a practical question: what value marks the point at which a given share of observations falls at or below it? For instance, a 90th-percentile page-load time of 4.2 seconds means roughly 90% of observed page loads took 4.2 seconds or less, while the slowest 10% took longer. NIST defines a percentile in terms of the percentage of measurements below and above a value, reflecting why percentiles are useful even when data is not evenly distributed. (itl.nist.gov)

What does percentile mean?

The simplest percentile meaning is relative standing. A percentile compares one value with all other values in the same relevant dataset.

If a page’s interaction delay is at the 75th percentile, that result identifies the boundary below which about three quarters of recorded visits fall. It does not mean that the page earned 75 out of 100, that it is 75% complete, or that 75% of users had exactly that experience.

Here is the plain-language interpretation of common percentiles:

  • 10th percentile: 10% of values are at or below this point; 90% are above it.
  • 25th percentile: one quarter of values are at or below this point; also called the first quartile, or Q1.
  • 50th percentile: half of values are at or below this point; this is the median.
  • 75th percentile: three quarters of values are at or below this point; also called the third quartile, or Q3.
  • 90th or 95th percentile: highlights the slower, larger, or otherwise more extreme end of the distribution.

The U.S. Bureau of Labor Statistics describes the median as the 50th percentile: the midpoint with half the values greater and half lower. (bls.gov) Percentiles extend that same idea beyond the middle, allowing an analyst to inspect the experience near the best, typical, or worst end of a group of results.

Percentile vs. percentage: the key difference

Percentile and percentage sound similar but describe different things.

A percentage is a share of a whole. If 82 out of 100 audit URLs have a meta description, the percentage is 82%.

A percentile is a location in a comparison group. If a page is at the 82nd percentile for speed, its result is faster than roughly 82% of the observations in the comparison set, assuming lower time is better and the rank is framed that way. The exact direction must always be stated: higher is not automatically better.

Consider this example:

  • An ecommerce category page returns 200 responses for 98% of crawled URLs.
  • Its 95th-percentile server response time is 1.9 seconds.

The first number tells the team what share of URLs met a condition. The second tells the team how bad the slower tail of response times is. Both numbers matter, but they answer different questions.

This distinction becomes especially useful in reports. Writing “95% percentile” is usually a warning sign: the intended measure may be a 95th percentile, a 95% pass rate, or a percentile rank. Reports should name the metric, the population, the time period, and whether a lower or higher result is desirable.

Percentile meaning with simple examples

Test-score example

Suppose a student is in the 80th percentile on an exam. That means the student scored as high as or higher than about 80% of people in the comparison group. It does not mean the student answered 80% of questions correctly.

A score of 72% could be in the 90th percentile if an exam was difficult, while a score of 92% could be in the 50th percentile if most test takers scored near 92%. The comparison group controls the interpretation.

Child-growth example

The Centers for Disease Control and Prevention uses percentile curves to compare a child’s measurement with a reference population of children of the same age and sex. A 5-year-old girl at the 25th percentile for weight weighs the same as or more than 25% of the reference group and less than 75% of it. The CDC also cautions that growth charts contribute to an overall picture and are not a diagnostic tool by themselves. (cdc.gov)

Website-performance example

Imagine 100 mobile page loads with LCP results sorted from fastest to slowest. If the 75th observation is 2.5 seconds, then the 75th-percentile LCP is approximately 2.5 seconds. About 75 visits loaded the largest visible content in 2.5 seconds or less; about 25 took longer.

That gives a web team a much clearer operational question than “what is the average?”: what is causing the slowest quarter of visitor experiences?

How to calculate a percentile from a small dataset

To calculate a percentile manually, start by sorting the values from smallest to largest. Then select a percentile convention and document it. This last step matters because software products can use different interpolation and ranking methods, particularly for small datasets or datasets with repeated values.

For a simple nearest-rank approach, use:

rank = ceiling((p / 100) × n)

Where:

  • p is the requested percentile, such as 75.
  • n is the number of observations.
  • ceiling means round up to the next whole number.

Worked 75th-percentile example

A crawl records these eight HTML response times in milliseconds:

120, 180, 210, 240, 300, 410, 560, 1,100

They are already sorted. To find the 75th percentile:

  1. Calculate the rank: ceiling(0.75 × 8) = ceiling(6) = 6.
  2. Take the sixth value.
  3. The 75th-percentile response time is 410 ms.

This result means six of eight observations, or 75%, are at or below 410 ms. The remaining two URLs take 560 ms and 1,100 ms.

For 10 values, the same method would place the 75th percentile at rank ceiling(7.5), or the eighth value. Other methods may interpolate between the seventh and eighth values instead. Neither approach is inherently deceptive, but comparing results from different tools without checking their calculation method can create avoidable confusion. NIST notes that percentile estimates depend on the ordered observations and discusses several estimation approaches. (itl.nist.gov)

Why the 50th, 75th, 90th, and 95th percentiles matter

Each percentile reveals a different slice of a distribution. A single average can hide spikes, outliers, and inconsistent user experiences.

For performance and audit work, the following pattern is often practical:

PercentileWhat it commonly showsUseful audit question
50th (median)The middle or typical observationWhat does a typical visitor or URL experience?
75thThe threshold affecting the slower quarterAre enough users receiving a reliably good experience?
90thA more difficult tail of resultsWhat affects one in ten visits or URLs?
95thThe most problematic 5% without focusing on only the single worst valueAre there severe issues needing targeted investigation?

The 75th percentile is prominent in Google’s Core Web Vitals guidance. Google states that a page or site is classified using the 75th-percentile value of page views, and a metric is considered good when at least 75% of page views meet the good threshold. (web.dev)

That policy is deliberate: it avoids letting excellent results for a majority fully mask a poor experience for a substantial minority. For internal monitoring, Google also suggests observing the 90th or 95th percentile when teams need to understand more severely affected visitors. (web.dev)

Percentiles in website performance audits

Percentiles are not just a statistics lesson; they shape how performance tools report real-user data. Google PageSpeed Insights, for example, presents real-user experience data over a rolling 28-day period when sufficient Chrome UX Report data is available. It can report at page or origin level, and the Core Web Vitals assessment uses 75th percentiles for LCP, INP, and CLS. (developers.google.com)

Field data and lab data answer different questions

A Lighthouse run is a controlled lab test: it provides a reproducible diagnostic snapshot under predefined device and network conditions. Field data comes from real visits, which vary by device, connection, geography, cache state, and user behavior.

Google explains that field data is a distribution rather than one number, while lab and field measurements can differ for valid reasons. (web.dev) A strong workflow uses both:

  1. Use field percentiles to identify whether real visitors have a broad problem.
  2. Segment results by mobile and desktop, page type, template, country, or connection where data permits.
  3. Use lab tests to reproduce and diagnose likely causes.
  4. Re-check field percentiles after release, allowing for the reporting window and sufficient samples.

For an audit across many URLs, local crawl results can also be grouped by template or directory. A tool such as Audra’s whole-site Lighthouse workflow helps turn isolated page-level checks into a repeatable site-wide review, where medians and upper percentiles make prioritization easier.

Percentile vs. mean and median in audit reporting

The mean, median, and percentile are complementary, not competing, statistics.

The mean adds all values and divides by the number of observations. It is useful for estimating the overall arithmetic average but can shift dramatically when a few values are unusually high. In the earlier response-time set, the 1,100 ms URL lifts the average even though most URLs are faster.

The median is the 50th percentile. It is more resistant to extreme values, but it can still understate a problem affecting 25% or 30% of visitors. A median LCP of 1.9 seconds does not guarantee that the slowest quarter has a good experience.

The 75th percentile focuses directly on that slower quarter. The 95th percentile shows whether a small but material segment experiences a particularly serious issue.

A useful client-ready summary might read:

  • Median mobile LCP: 1.9 seconds
  • 75th-percentile mobile LCP: 3.1 seconds
  • 95th-percentile mobile LCP: 6.4 seconds
  • Main affected template: filtered category pages

That tells a far more actionable story than an average alone. For a deeper comparison of these measures in audit work, see Percentile vs Mean: What the Difference Means for Website Audits.

Common mistakes when interpreting percentiles

Percentiles are easy to calculate incorrectly and even easier to explain imprecisely. These errors commonly weaken SEO, performance, and accessibility reporting.

Treating percentile as a score out of 100

A 90th-percentile LCP is not a “90% score.” It is a time value, such as 3.8 seconds, located at the point where about 90% of observations are at or below it.

Forgetting the comparison population

“80th percentile” is incomplete without the group. Is the comparison based on all mobile page views in the last 28 days, URLs in a crawl, competing sites, or pages within one template? A percentile is only meaningful relative to a defined population.

Assuming higher is always better

For rankings, conversion rate, or test scores, a higher percentile is generally favorable. For LCP, INP, response time, and crawl depth, a higher percentile often means a worse result because higher values mean more delay or distance.

Mixing calculation methods

Excel, SQL engines, analytics platforms, and custom scripts may select or interpolate values differently. The discrepancy is usually most visible with small samples. A report should name the tool or method when the exact number affects a decision.

Ignoring sample size and segmentation

A 95th percentile built from 20 URLs is much less stable than one built from 20,000 real page views. Likewise, combining desktop and mobile results can hide a mobile-only issue. Google recommends measuring Core Web Vitals separately across mobile and desktop page loads. (developers.google.com)

A practical percentile workflow for SEO and agencies

Percentiles are most useful when connected to an action, not dropped into a dashboard without context. A repeatable website-audit workflow can look like this:

  1. Define the population. Choose pages, visits, templates, or crawl results that belong together. Do not merge product pages, checkout flows, and blog posts without a reason.
  2. Choose a direction. State whether lower or higher values are better for the metric.
  3. Start with the 50th and 75th percentiles. This shows the typical result and the point affecting the slower or weaker quarter.
  4. Inspect the 90th or 95th percentile. Use it to find severe tail problems, such as a small set of pages with oversized images or slow third-party scripts.
  5. Open the underlying URLs or visits. A percentile identifies the boundary; it does not explain the cause.
  6. Track change with a consistent method. Use the same data source, segmentation, and percentile calculation over time.

For example, an agency auditing 500 pages might find a median Lighthouse performance score of 78 but a 25th-percentile score of 42. That does not prove that 25% of users have poor real-world performance—Lighthouse scores are lab measurements—but it does flag a meaningful lower-performing group of URLs for investigation.

A local-first audit process can preserve the URL-level evidence needed for that investigation. Audra combines technical SEO, performance, accessibility, best-practice, link, and AI-answer-engine visibility checks in a desktop workflow, which can help teams group issues before reporting them to a client. Teams comparing crawl-oriented software can also review Audra vs. Sitebulb for technical SEO and AI search workflows.

When percentiles are not enough

A percentile summarizes position; it does not diagnose causation, establish business impact, or guarantee statistical stability.

For example, a 75th-percentile LCP of 3.0 seconds identifies a threshold, but it does not reveal whether the cause is unoptimized hero images, render-blocking CSS, slow server response, client-side rendering, consent banners, or a particular geography. The analyst needs URL-level diagnostics, waterfall data, template comparisons, and ideally real-user segmentation to answer that next question.

Percentiles also need appropriate context when the dataset is tiny. With just four URLs, the 75th percentile may effectively point to one URL depending on the chosen method. In that case, show all four observations rather than creating a false sense of statistical precision.

Finally, a percentile does not replace pass rates. An accessibility audit may need both the percentage of pages with a missing form label and the 90th percentile of issue counts per page. The first shows prevalence; the second shows how severe the heavier-issue pages are.

FAQ

What does being in the 75th percentile mean?

Being in the 75th percentile means a value is at or above roughly 75% of values in its comparison group, or that about 75% fall at or below it depending on the convention being described. For page-load time, a 75th-percentile value marks the experience of the slower quarter boundary, where lower time is better.

Is the 50th percentile the same as the average?

No. The 50th percentile is the median, meaning half of observations are lower and half are higher. The average, or mean, is calculated by adding values and dividing by the count. A few extreme values can change the mean substantially while leaving the median largely unchanged. (bls.gov)

How is a percentile different from a percentage?

A percentage describes a share of a whole, such as 85% of URLs returning a 200 status code. A percentile describes relative position in a dataset, such as a 85th-percentile response time of 900 ms. The first is a rate; the second is a ranked threshold value.

Why does Google use the 75th percentile for Core Web Vitals?

Google uses the 75th percentile so that a site must provide a good experience to at least 75% of relevant page views to meet the recommended threshold. This avoids judging a site as healthy solely because its fastest or most typical visits perform well while a substantial minority has slower experiences. (web.dev)

Is a higher percentile always better?

No. It depends on what is measured. A higher percentile rank is usually positive for a test score or revenue result. But a high percentile value for LCP, INP, response time, error count, or number of accessibility issues often indicates a worse outcome because lower values are preferable.

Sources