Most typing-speed pages repeat a number they read somewhere else. This site is built differently: it’s a data project with named sources, a documented model, and — uniquely — its own growing dataset. Here’s exactly how the numbers work.
Where our benchmarks come from
Three layers. Published studies: the general-population baseline (~40 WPM) comes from Teresia Ostrach’s 3,475-person study; the online-test average (52 WPM) comes from Aalto University’s 2018 analysis of 136 million keystrokes from 168,000 volunteers — the largest typing dataset ever published. Job postings: role requirements (data entry, transcription, admin, support) are compiled from live listings, not guessed; the growing dataset lives on the jobs page. Our own tests: every test on this site records an anonymous result — WPM, accuracy, test length, nothing else, no accounts, no personal data — and those aggregates progressively recalibrate our model.
The percentile model
Percentiles use a normal-distribution model, currently mean 43 WPM and standard deviation 15, fitted to the published studies above. When you score, say, 58 WPM, your percentile is the share of that distribution below 58 — about 84%. Two honest caveats: real typing speeds are slightly right-skewed (a long tail of very fast typists), and our initial age-group figures are modeled from study demographics rather than measured directly. Both improve as our own dataset grows; when we switch a figure from modeled to measured, the page’s verified date changes and this page says so.
How the test itself works
The test uses common English words in randomized order — no punctuation or capitals, which makes results slightly flattering versus real-world typing but keeps them comparable across attempts and with major online tests. Net WPM counts only correctly typed words; gross counts all; accuracy is per keystroke. One-minute tests run a few WPM hotter than five-minute tests (sprint versus sustained pace) — which is why our result screen and the test offer 1, 3 and 5-minute modes.
What we don’t do
No fabricated precision (a figure that’s modeled is labeled modeled), no personal data collection (results are anonymous aggregates), and no recycling of unsourced numbers — if we can’t trace a figure, we don’t publish it.
Corrections
Spot a figure that’s outdated or a study we’ve misread? Tell us — corrections are credited, and every data page carries the date its figures were last verified.
Who runs this
This site is researched and maintained by Umar Attique, a data researcher who builds measurement tools — including the typing test on our homepage, which runs on the same model documented here. More at umarattique.com.