statisticspercentilesdata analysisseo reportingwebsite audits

Percentiles Explained: Formula, Examples and Practical Uses

Percentiles show where a value sits within a dataset, making them useful for interpreting rankings, quartiles, performance tails, and website-audit results.

· 15 min read

A page-load time of 3.9 seconds at the 75th percentile does not mean the page earned 75% of a score: it means 75% of measured pages loaded in 3.9 seconds or less under the stated dataset and method. This guide explains percentiles step by step so readers can calculate them, distinguish them from percentages, compare common formulas, and use them to report website-audit data more clearly.

What is a percentile?

A percentile is a value that indicates a position within an ordered dataset. The nth percentile is commonly described as the value below which n% of observations fall, with the remaining observations above it. Minitab uses that practical definition for percentiles, while also noting that calculation details depend on the method used. (support.minitab.com)

For example, suppose an agency audits the mobile load time of 100 pages:

  • A 25th percentile of 1.8 seconds means roughly one-quarter of pages loaded in 1.8 seconds or less.
  • A 50th percentile of 2.5 seconds means roughly half of pages loaded in 2.5 seconds or less.
  • A 75th percentile of 4.1 seconds means roughly three-quarters of pages loaded in 4.1 seconds or less.
  • A 90th percentile of 7.8 seconds highlights the slower tail: only about 10% of pages take longer than 7.8 seconds.

Percentiles are therefore about relative standing, not a score out of 100. They answer a question such as, “Where does this value fall compared with the rest of the sample?”

This makes percentiles particularly useful when an average hides variation. If 90 pages load in 2 seconds and 10 pages load in 12 seconds, the mean is 3 seconds. That average sounds manageable, but the 90th percentile exposes the group of pages that users are more likely to experience as slow.

Percentile versus percentage: the distinction that causes most confusion

A percentage measures a portion of a known total. A percentile measures a value’s rank or location relative to other values.

Consider a student who answers 79 out of 100 questions correctly. Their percentage score is 79%. If that student scored better than 79% of the comparison group, their percentile rank is approximately the 79th percentile. Those figures can be the same number, but they answer different questions.

MeasureWhat it tells the readerExample
PercentagePortion of a total achieved79 of 100 questions correct = 79%
PercentileRelative position in a datasetBetter than about 79% of a comparison group = 79th percentile
Percentile valueThe data value at a specified positionThe 75th-percentile load time is 4.1 seconds

A useful wording check is this:

  • Say “the page received 82% of possible points” for a percentage.
  • Say “the page’s 3.2-second load time is at the 68th percentile” for a relative rank.

The distinction matters in client reporting. A technical SEO audit might show that 18% of URLs have missing meta descriptions. That is a percentage of affected URLs. It might also show that the 90th-percentile crawl response time is 2.7 seconds. That is a percentile value summarizing the slower end of the distribution.

Percentiles and quartiles: Q1, median, Q2 and Q3

Quartiles are named percentiles that split ordered data into four broad sections. They are easier to remember because there are only three main cut points:

  1. First quartile (Q1) = 25th percentile.
  2. Second quartile (Q2) = 50th percentile, also called the median.
  3. Third quartile (Q3) = 75th percentile.

The median is the middle of an ordered dataset. Half of observations are at or below it and half are at or above it, subject to how ties and even-sized samples are handled. The 25th percentile marks the lower quarter, while the 75th percentile marks the point below which roughly three-quarters of observations sit.

Quartiles are commonly used alongside the interquartile range (IQR):

IQR = Q3 − Q1

If Q1 is 1.7 seconds and Q3 is 3.9 seconds, the IQR is 2.2 seconds. That describes the spread of the middle 50% of values without letting a handful of unusually slow pages dominate the summary.

For audit reporting, Q1, median and Q3 can create a useful compact story. A report might say: “The middle 50% of HTML pages responded between 280 ms and 720 ms, with a median of 440 ms.” That is more informative than a single average when response-time data are skewed.

How to calculate percentiles with one ordered dataset

The following dataset contains 12 page response times, in seconds, already sorted from lowest to highest:

1.1, 1.4, 1.6, 1.9, 2.0, 2.4, 2.7, 3.1, 3.5, 4.0, 4.8, 6.2

To calculate a percentile manually, first choose and state a convention. One commonly taught interpolation approach, also reflected in Minitab’s percentile-function documentation, uses this location formula:

L = (n + 1) × p

Where:

  • L is the location in the ordered list.
  • n is the number of observations.
  • p is the requested percentile written as a decimal: 0.25, 0.50 or 0.75.

For this dataset, n = 12.

Step 1: Calculate the 25th percentile, or Q1

L = (12 + 1) × 0.25 = 3.25

The location is 3.25, which falls one-quarter of the way between the third and fourth values:

  • Third value = 1.6
  • Fourth value = 1.9

Interpolation gives:

Q1 = 1.6 + 0.25 × (1.9 − 1.6) = 1.675

Using this convention, the 25th percentile is 1.675 seconds.

Step 2: Calculate the 50th percentile, or median

L = (12 + 1) × 0.50 = 6.5

The 50th percentile lies halfway between the sixth and seventh values:

  • Sixth value = 2.4
  • Seventh value = 2.7

Median = 2.4 + 0.5 × (2.7 − 2.4) = 2.55

The median is 2.55 seconds.

Step 3: Calculate the 75th percentile, or Q3

L = (12 + 1) × 0.75 = 9.75

The location is three-quarters of the way between the ninth and tenth values:

  • Ninth value = 3.5
  • Tenth value = 4.0

Q3 = 3.5 + 0.75 × (4.0 − 3.5) = 3.875

The 75th percentile is 3.875 seconds under this method. Minitab documents the same general structure: calculate a position using p(N+1), identify the lower ordered value, then use the fractional part of the position for interpolation. (support.minitab.com)

The percentile formula is not always the same

There is no single percentile formula used everywhere. This is why two spreadsheets, statistical packages or dashboards can produce slightly different answers from the same small dataset.

NIST notes that multiple percentile-calculation methods are in common use and references the nine methods catalogued by Hyndman and Fan. (itl.nist.gov) The disagreement is most visible with small samples, tied values, or percentiles near 0 and 100.

Three common approaches illustrate the issue:

MethodPosition rule for the 75th percentile when n = 12Result in the worked dataset
Nearest rankceil(np) = ceil(9)3.5
(n + 1)p interpolation13 × 0.75 = 9.753.875
Inclusive linear interpolation1 + (n − 1)p = 9.253.625

None of these results is automatically “wrong.” They reflect different definitions of the estimated quantile in a finite sample.

The practical reporting rule is simple: name the tool and method when comparability matters. If last month’s report used Excel’s inclusive percentile function and this month’s report uses another application, a small change in Q3 may be methodological rather than operational.

For a large crawl of 20,000 URLs, differences among reasonable methods often become less material. For a 12-page sample, the difference between 3.5 and 3.875 seconds can be noticeable. That is why audit templates should preserve the calculation method, the crawl date, the sample size and any exclusions.

Percentile rank: finding the rank of a known value

A percentile value answers, “What response time marks the 75th percentile?” A percentile rank reverses the question: “What percentage of observations are at or below this response time?”

In the 12-value dataset, a response time of 3.1 seconds is the eighth listed value. Using an inclusive cumulative approach:

Percentile rank = (number of values at or below x ÷ n) × 100

= (8 ÷ 12) × 100 = 66.7%

So 3.1 seconds is at approximately the 66.7th percentile in this sample. About two-thirds of recorded times are at or below 3.1 seconds.

However, percentile-rank conventions vary too. The cume_dist() function in dplyr counts values less than or equal to a value and divides by the number of observations. Its percent_rank() function instead counts strictly lower values and divides by one fewer observation. (dplyr.tidyverse.org) With ties, those choices can materially change the reported rank.

For client-facing reporting, avoid overprecision. “Around the 67th percentile” is clearer and more honest than “66.6667th percentile,” particularly when the sample is small or contains tied values.

How to read a cumulative relative frequency graph

A cumulative relative frequency graph, often called an ogive, shows the running percentage of observations at or below each value or class boundary. The vertical axis runs from 0% to 100%; the horizontal axis shows the measured variable, such as page weight, crawl duration or response time.

For grouped response-time data, a table might look like this:

Response-time boundaryCumulative countCumulative relative frequency
1 second315%
2 seconds945%
3 seconds1470%
4 seconds1785%
5 seconds1995%
6 seconds20100%

The cumulative relative frequency at each row is calculated as:

cumulative count ÷ total observations

For example, 14 ÷ 20 = 0.70, or 70%. Educational statistics references describe cumulative relative frequency as cumulative frequency divided by sample size, and note that an ogive is useful for reading percentile positions. (stats.libretexts.org)

To estimate the 75th percentile from the graph, start at 75% on the vertical axis, move horizontally to the curve, and then drop to the horizontal axis. In this example, 75% falls between 3 and 4 seconds, so the estimated 75th-percentile response time is somewhere in that interval.

Grouped graphs estimate rather than recover exact percentiles because the original values have been placed into bins. Smaller, sensible intervals can improve precision, but they may also make a chart noisy when the sample is small.

What percentile is the top 10 percent?

The cutoff for the top 10% is the 90th percentile. If a metric is better when it is larger, values at or above the 90th percentile are generally in the top 10% of the dataset.

For example, if the 90th-percentile organic-traffic value across 500 landing pages is 4,200 monthly visits, pages with at least 4,200 visits are around the top tenth of that group. But direction matters:

  • For traffic, conversions or visibility, higher values are usually better, so top 10% means at or above P90.
  • For load time, crawl time or error count, lower values are usually better, so the slowest or highest-error 10% is at or above P90—but that is the worst tail, not the best one.

Ties complicate a strict top-10% label. If many URLs share the same value at the P90 boundary, more than 10% of URLs may be included. A report should say “at or above the 90th-percentile threshold” rather than imply an exact number of pages when ties are common.

Is a 79th percentile good?

A 79th percentile means a value is higher than, or at least as high as, roughly 79% of the chosen comparison dataset under the selected rank convention. Whether that is good depends on what is being measured and whether higher or lower is desirable.

A 79th-percentile conversion rate among comparable landing pages is usually encouraging. A 79th-percentile time to first byte is less encouraging if higher time means slower delivery. The percentile supplies context, not a verdict.

Before calling a percentile good or bad, check four details:

  1. Direction: Is a larger number desirable, such as revenue, or undesirable, such as JavaScript payload?
  2. Comparison set: Is the page compared with all URLs, only indexable HTML pages, or a specific template type?
  3. Time period: Was the dataset collected in one crawl, one week or a rolling 90-day window?
  4. Measurement method: Were the same device, crawler settings, throttling rules and percentile convention used?

This is especially relevant when measuring AI-search visibility. A brand might appear in the 79th percentile for mentions across a defined prompt set, but that result only has meaning when the prompts, competitors and answer engines are held consistent. Teams building that workflow can compare it with an Ahrefs Brand Radar ChatGPT mention audit workflow before deciding what their benchmark should be.

Using percentiles in SEO, performance and accessibility audits

Percentiles are often more useful than averages when audit data have a long tail. Website datasets frequently do: one slow template, a few redirect chains or a limited set of pages with huge image payloads can distort the mean.

A practical audit can report the following for each metric:

  • P50 (median): the typical page or user experience.
  • P75: a realistic upper-middle threshold that reveals whether a substantial minority of URLs is struggling.
  • P90 or P95: the problematic tail that deserves prioritization.
  • Count above the threshold: the number of URLs requiring action.

For example, an audit of 2,000 pages might find a median HTML response time of 410 ms, a P75 of 690 ms and a P95 of 2.8 seconds. The median alone suggests the site is healthy. The P95 reveals that around 100 pages sit in a much slower segment worth investigating by template, server region, cache status or redirect behavior.

The same idea applies to issue volumes. Suppose most pages have zero accessibility issues, but a legacy template has 18 missing form labels per page. The average may understate the problem; a high percentile makes the affected tail obvious. A local audit workflow can preserve the underlying URL list so the report connects each percentile to concrete pages rather than treating it as an abstract statistic.

Audra’s local-first approach is designed for this type of analysis: teams can audit technical SEO, performance, accessibility, links and answer-engine visibility without treating a single average as the whole story. When reporting AI visibility alongside conventional SEO signals, percentile summaries can also help separate typical page performance from outlying templates or content groups. For related measurement context, see the overview of Ahrefs Adaptive Mode and Brand Radar updates.

Calculating percentiles in StatCrunch and Minitab

For routine coursework or exploratory analysis, software is usually preferable to manual arithmetic once the calculation convention is understood.

In StatCrunch, percentile calculation is available through summary-statistics workflows; an instructional StatCrunch example directs users to enter a requested value such as 90 in the Percentiles field. (statcrunch.com) StatCrunch is Pearson’s web-based statistical analysis platform, so it is commonly encountered in introductory statistics courses. (statcrunch.pearson.com)

In Minitab, the Calculator’s percentile function uses a specified probability and sorted observations. Its documentation explicitly gives the p(N+1) position framework and describes interpolation using the fractional portion of the calculated position. (support.minitab.com)

Regardless of the tool, a reliable process is:

  1. Remove or explicitly handle missing values.
  2. Confirm whether values are raw observations or grouped data.
  3. Record the software, function and version if the result will be compared over time.
  4. Export the ordered values or URL-level data behind any client-facing percentile.
  5. Pair P50 with at least one tail percentile, commonly P75, P90 or P95.

A tool can calculate the number, but it cannot decide whether the relevant comparison group should be every crawled URL, only canonical pages or only pages in a specific template family. That decision belongs in the audit methodology.

FAQ

How do you calculate a percentile?

Sort the dataset from lowest to highest, choose a percentile convention, calculate its position and read or interpolate between the surrounding ordered values. One common formula is L = (n + 1)p, where n is the number of observations and p is the requested percentile as a decimal. Different software may use a different position formula. (support.minitab.com)

What does a percentile mean?

A percentile indicates where a value sits relative to other values in an ordered dataset. The 75th percentile is commonly interpreted as a value at or below which about 75% of observations fall. It is a measure of relative standing, not a percentage score or a guarantee about every individual observation. (support.minitab.com)

What is the difference between a percentile and a percentage?

A percentage reports a share of a total, such as 45 broken links out of 900 URLs being 5%. A percentile reports position within a comparison set, such as a 2.9-second response time sitting at the 80th percentile. The first describes quantity achieved or affected; the second describes relative rank.

What percentile is top 10 percent?

The 90th percentile is the usual threshold for the top 10%. For a metric where higher is better, values at or above P90 are in the top tenth. For a metric where lower is better, such as load time or crawl duration, values at or above P90 are generally the slowest or worst-performing tenth.

Quartiles are specific percentiles. Q1 is the 25th percentile, Q2 is the 50th percentile and is also the median, and Q3 is the 75th percentile. Together, Q1 and Q3 describe the spread of the middle half of the dataset, while the median identifies its central position.

Sources