What a Typing Accuracy Percentage Means

The strictest reading. Every key you press is either right or wrong, and accuracy is the proportion that were right:

keystroke accuracy = correct keypresses ÷ total keypresses

Under this rule a mistake is permanent the moment you make it. Correcting it does not remove the error from the count — and, on some implementations, the backspaces themselves count as additional keystrokes, which makes correcting actively worse for the figure.

This definition measures what your fingers did. It is the most honest description of motor performance and the least forgiving description of a finished piece of text, and it is why a typist who habitually corrects can be surprised by how low it comes out. This site uses this definition, which is why its accuracy figure and its error-handling differ: corrections cost time rather than accuracy in the speed calculation, while the accuracy figure still records that the mistake happened.

The most forgiving reading. Only the text you ended up with is examined:

final-text accuracy = correct characters in the output ÷ total characters in the output

Under this rule, corrections are free. A typist who mistypes forty characters and fixes every one of them scores 100%, because the finished text is perfect.

This is not a lax definition so much as a different question. It asks whether the work you produced was correct, which is the question that matters for a document, an email or a database field. Keystroke accuracy asks whether your hands are reliable, which is the question that matters for practice.

The two diverge most for exactly the people who care most: careful typists who correct constantly. Their finished work is near-perfect and their motor accuracy is ordinary, and which number they see determines whether they think they need practice.

The coarsest reading, and the one most common in employment testing:

word accuracy = fully correct words ÷ total words

A word with one wrong letter is wrong. A word with four wrong letters is equally wrong. This is a much blunter instrument than the other two and it behaves oddly in a specific way: it punishes errors distributed across many words far more than the same number of errors concentrated in one.

Eight mistakes in eight different words is eight bad words. Eight mistakes in two words is two. The typist with the second pattern scores much higher on identical error counts, which has very little to do with how good their typing is.

It survives because it maps onto a document-level question — how many words would need fixing — and because it is easy to explain. It is a poor measure of typing and a reasonable measure of proofreading load, and those are not the same thing. Employment tests using it are discussed further in typing tests in hiring.

Take a typist who types 500 characters, makes 15 incorrect keypresses, corrects 12 of them, and leaves 3 wrong in the final text. The 12 corrections also required 12 backspaces.

Keystroke accuracy, counting backspaces as keystrokes: 500 correct out of 527 total presses, about 94.9%.

Keystroke accuracy, not counting backspaces: 500 out of 515, about 97.1%.

Final-text accuracy: 497 correct characters out of 500, 99.4%.

Word accuracy, if those 3 remaining errors fall in 3 separate words out of 100: 97%.

Four numbers between 94.9% and 99.4%, all describing one session, none of them miscalculated. When someone reports their accuracy without saying which they mean, that spread is the uncertainty attached to the figure.

For practice, keystroke accuracy. It is the only one of the three that responds to your hands rather than to your editing, so it is the only one that will tell you whether a drill worked. A figure that lets corrections erase mistakes cannot show improvement in the thing you were training.

For work, final-text accuracy, because it answers the question your output is judged on. If you correct as you go and the finished document is clean, your keystroke accuracy is a fact about your process rather than about your results.

For comparing yourself against a threshold — a job posting, a certification — find out which definition the test uses before assuming your number transfers. A 98% under final-text accuracy is a considerably weaker claim than 98% under keystroke accuracy, and the two are quoted interchangeably.

If you want to see the gap on your own typing directly, take an ordinary test and then a no-backspace test. The first lets corrections hide your errors; the second does not.

Percentages flatten scale, and typing produces a great many characters, so the intuition attached to a figure like 97% is badly calibrated.

At 97% keystroke accuracy and a typical working pace, you are making an error every thirty-odd characters — several times a sentence. That sounds much worse than 97% does, and it is the same fact stated twice.

This is why accuracy work has a much larger effect on effective output than the percentage suggests, and why the arithmetic in accuracy versus speed comes out the way it does. Moving from 97% to 98.5% halves your error count, which sounds like a rounding adjustment and is in fact the difference between correcting twice a sentence and correcting once.