How to Increase Typing Speed, Starting With the Errors

The order that works, in one page

If you want to type faster, stop trying to type faster and start removing what interrupts you. Every mistake costs time twice: the time inside it, and the time spent noticing, finding, undoing and getting back to where you were. A clean run is already fast, because nothing had to be repaired.

The second lever is specificity. People stuck on a number are rarely slow at typing in general; they are slow at eight or ten particular key sequences, and they pay for them on every page. Fixing that short list moves the average. Re-running text you can already type does not.

The third lever is structure. Short, frequent sessions aimed at one named failure build the pattern; long ones decay into performance with a climbing error rate.

• Accuracy before speed: take the errors out and the words-per-minute figure follows.
• Practise where you stay clean; raise the tempo only when clean gets easy.
• Keep an even rhythm rather than racing easy words.
• Stop looking at your hands; wean off with anchors, not a keyboard cover.
• Train capitals, digits and punctuation on real text; drills under-train them.

Accuracy before speed, because errors cost more than they save

Take the cheapest possible mistake. You hit the wrong key in a three-character sequence, see it immediately, press backspace three times and retype the three characters: six keystrokes to produce three. That does not count the attention spent spotting it, which breaks your reading ahead. Miss the noticing and the cost grows — every character after the mistake sits one position away from where the passage expects it, so you either leave a tail of garbage in place or delete good text along with bad.

That is why accuracy comes first, and not as a preference for neatness. Speed is the rate at which your hands execute sequences they can execute reliably. Whatever that rate is, the number is already available and errors are the subtraction from it. Remove the subtraction and the figure rises without anyone pushing it.

There is a worse reason to hurry: practising through your errors rehearses them. Key sequences are learned as timing patterns, and a pattern repeated at speed gets stronger. Mistype the same pair of letters every time you run flat out and you are not failing to learn the right sequence — you are learning the wrong one, with the recovery gesture welded to it. That is why people who force tempo feel they have regressed.

So define clean before you define fast. Clean means you never reached for backspace and nothing wrong was left in the text. Hold that and every speed number after it rests on something repeatable under pressure: in an email to a client, in notes taken during a meeting, in code where a stray character is a bug.

How net WPM, raw WPM and accuracy are actually calculated

The accuracy question deserves a real answer, and it starts with a convention. Typing tests do not count words. They count characters and divide by five, because an average English word plus its space lands close to five — which is why two different tests can agree on your number even though neither knows what a word is. Each passage in the linked test runs to about 230 characters, so finishing one is roughly 230 / 5 = 46 words.

Here are its three numbers, in the test's own terms. Net WPM counts only the characters still correct in the box at the moment of measurement, converted to words and divided by the elapsed minutes. Raw WPM uses every character in the box, right or wrong. Accuracy is correct characters divided by typed characters, checked against the passage you were given.

Work one through. Suppose the box holds 200 characters when the run ends and 180 of them match the passage. Net WPM comes from the 180: 180 / 5 = 36. Raw WPM comes from all 200: 200 / 5 = 40. Accuracy is 180 / 200 = 0.9, shown as 90%. Notice what the gap between raw and net describes: characters you typed and left wrong. It is not a record of how much you corrected.

Metrics are recomputed from the current contents of the box, which changes how you read your scores: backspace over a mistake and retype it, and it vanishes from net WPM, from raw WPM and from accuracy as though it never happened. Only uncorrected characters count against you. A timed run lasts a fixed 60 seconds; a free run ends when you finish the passage. The passage comes from twelve hardcoded English texts, drawn at random and never repeating the previous one, with no custom text and no non-English passage.

Slow practice on purpose, and an even rhythm

Every typist has a tempo at which their error rate starts to climb. Below it you are executing; above it you are guessing and repairing. Set practice by accuracy rather than speed, and let the tempo be whatever keeps accuracy above the line you refuse to cross.

Working on the clean side of the threshold

Practising at the edge of control feels productive because it feels difficult, and it is the least productive thing available to you. You spend your repetitions inside recovery, your error rate stays flat because you never give your hands a clean example to learn from, and the tension shows up as missed keys. It is rehearsing mistakes at tempo.

Deliberate slow practice feels like wasting your ability, and it is not. Drop the tempo until you can finish a run with no backspaces and nothing left wrong, stay there until it is boring, then step up a little and step back down the moment errors return. A pattern that is easy at a slow tempo is one that exists; a pattern that only works where you make errors does not exist yet.

• Pick a tempo you can hold without errors, not the fastest one you can survive.
• Step up in small increments; step back down as soon as errors return.
• Count reaching for backspace as a failure of the run, not as a repair.

Rhythm, which is the part everyone skips

Burst speed and steady speed are not the same skill. Racing easy words and stalling on hard ones feels fast, but acceleration and deceleration are where missed keys cluster, and a stall is your hands waiting on a pattern they do not have. An even cadence removes the timing variability that makes you overshoot a key.

Train against a steady pulse from any device that offers one, because this test does not analyse your consistency or your bursts — that part you have to hear. Set it low enough that one character per beat needs no rushing, and aim at the beat rather than at a number. Refuse to speed up for easy words: hold the same interval through “the” that you hold through “bureaucratic”. Treat the space bar as part of the rhythm, not the period after the word, because many boundary errors are rhythm errors.

Stop looking down, without covering the keyboard

Looking at your hands is the most expensive habit here, and not mainly for the travel time: it moves your attention off the text. Typing is a lookahead task — while your hands are on the current word your eyes should be on the next one or two, and a glance down spends the attention that needs. Finding your place again when you look up costs a second time, usually larger than the first.

The usual prescription is to cover the keyboard. Skip it. Blindfolding yourself does not build a map of the layout; it converts looking into guessing, which teaches your fingers that key positions are uncertain. Wean off by making the map concrete.

Anchors instead of a cover

Open every session by finding the two bumps on F and J and settling the rest of the hand on the home row without looking; those bumps exist for exactly this. When a key feels uncertain mid-word, return to the bumps and navigate by touch. Slow for a week, permanent after that.

Allow glances, but schedule them — between runs rather than inside words, and for one named key rather than for the whole keyboard. Checking where the semicolon lives is a useful glance. A gaze drifting down mid-passage is not.

The tap-and-report drill

To find out what your fingers actually know: say a letter, press its key, and only then look to see whether you were right. Log the misses. Twenty of these replace one more high-speed run, and they give you an honest list of the regions of the layout your fingers were never told about — which is what slow practice should spend its repetitions on. Expect the weak spots to cluster: the number row and right-hand punctuation.

Drills are not typing: capitals, digits, punctuation and code

Training passages and real text differ in distribution, not difficulty. Drill passages are overwhelmingly lowercase letters and spaces, because that makes prose readable at a training tempo. Real typing is loaded with what a drill gives you twice an hour: capitals, punctuation on shifted keys, digits, and symbols.

Capitals are the classic under-trained case. Shift is a pinky or thumb reach that drills rarely combine with the letter it modifies, so most people learn the two-finger version — one finger holds shift while another hunts — slow, unreliable, and the first thing to collapse when you are thinking about what you mean rather than where the keys are. Digits are worse: the number row has no home position and no bumps, so unless it was drilled on purpose it stays a lookup, and lookups stop you reading ahead.

Punctuation costs time at word boundaries, where errors are dearest. A missed apostrophe or an unopened quote sends you back into a word you already finished, and dialogue is the worst case: quotes, apostrophes and a capital on the next word, stacked in a sequence drills never rehearse.

Code adds its own tax. Identifiers put a capital mid-word rather than at the start, where nothing is braced for it; snake_case wants a shifted underscore; brackets and operators sit on keys drill text never uses; and indentation turns the space bar into something you have to count. Syntax is less predictable than prose, so the lookahead you built on passages does not transfer: in code you type in tokens. So retype a source file you work in daily, character for character, indentation included; transcribe a page of dialogue; type the email you owe someone. Then measure on the test, remembering it shows you prose — your best case, not your working case.

From 40 to 60 wpm: where the time actually goes

The honest framing first: nobody can hand you a schedule for this, including this page. How long it takes depends on how much real typing you do, which errors are load-bearing for you, and whether you practise clean or through your mistakes. The mechanism is predictable and encouraging: at this stage the gap is rarely finger speed. Your hands can do more than your average shows, and the difference goes to recovery.

The uncaught error, which is the expensive one

At 40 wpm you read roughly a word ahead, and where people stall that lookahead is unstable, so errors go unnoticed for several characters. Everything after the mistake is misaligned, you find out at the end of the line, and you delete a long run and retype it while reconstructing what you meant. The highest-value change is not tempo but the delay between making a mistake and knowing about it.

Shifted keys: capitals, digits and symbols

Ask which errors people at this stage make and they say “random typos”. Their text says otherwise: the capital that came out lowercase, the digit that came out as the symbol above it, the apostrophe missed entirely. These are reach errors on shifted keys, and drills hide them because drill passages are mostly lowercase.

• Digits: pick one route — top row or keypad — and train only that.
• Capitals: drill shift-and-key as one gesture, then in real combinations.
• Punctuation: transcribe dialogue, which stacks quotes, apostrophes and capitals.

Word boundaries and the correction loop

Missing and doubled spaces cost far more than one character each, because they merge or split words and a merged word has to be reopened. The other boundary failure is over-correction: you go back for one letter, hold the backspace a beat too long, lose the whole word, and retype it from memory instead of from the passage. Practise the minimum repair — one backspace, one character — the fastest option, because it does not destroy good work.

None of this shows in your score. The test keeps no error count and no record of corrections, so a run full of repairs and a flawless one can produce identical numbers, which is why the tally below exists.

What a realistic plateau looks like, and how to break one

A plateau in typing is not a shortage of headroom. It is your average being dominated by a handful of sequences you cannot do reliably: the outliers cost enough that everything you type smoothly stops mattering. That is why plateaus feel invisible from inside: the easy passages feel great, the number does not move, and practice has become performance, demonstrating your current speed instead of changing it. The commonest shape is test numbers holding while real typing does not, which says the missing material is what you no longer practise on.

Change the input before you change the tempo. Move practice onto real text for a while. Take the failure list from your tally and drill only those patterns, slowly, out of context. Drop your tempo for a couple of sessions and rebuild it, which resets where your threshold sits. Add one constraint at a time — harder characters, a longer lookahead, or a steady beat — because with two you cannot tell which cost you the accuracy.

Then leave it alone. Motor patterns consolidate between sessions rather than inside them, which is why a night's sleep changes how your hands behave more than an extra hour of practice. If none of that moves the number, the remaining time is elsewhere: in reading ahead, in trusting the layout without checking it, or in symbols you dodge by rewriting sentences so you never type them.

How to structure practice: short, frequent, aimed at one failure

Marathon practice backfires in a specific way. Attention degrades before your fingers do, so the back half of a long session carries a higher error rate than the front half, and in that half you are rehearsing errors at speed. Fifteen focused minutes most days beats an hour once a week, because motor patterns come from repetition with feedback, not endurance.

A fifteen-minute block

Two minutes of warm-up at a deliberately slow tempo, no target and no measurement, just finding the home row. Eight minutes on the one failure you named before sitting down. Three minutes on your own material: a real file, a draft you owe someone. Two minutes on one measured run, written down somewhere durable. That is 2 + 8 + 3 + 2 = 15. The closing run is not a report on the session; it is the marker you compare with the runs either side of it.

Name the failure before you start

“Practise typing” is not a session. “Capitals in the middle of words” is. “The apostrophe in contractions” is. “Digits on the number row under the right hand” is. A named failure gives your attention something to monitor, which is what lets you tell whether the drill worked; with no target, a useful ten minutes and a wasted ten minutes feel identical. Pick from your tally.

Keep a manual error tally

This follows from how the test computes accuracy. Corrections leave no trace: type a word wrong, backspace it, retype it correctly, and the box ends up as though you never made the mistake — same net WPM, same raw WPM, same accuracy. Two runs can score identically while one was clean and one was a dozen repairs deep.

So keep a pencil and something to tick beside the keyboard, in two columns: characters you left wrong, and mistakes you corrected. Compare the corrected column with the accuracy figure at the end of the run. If the figure reads 98% and your tally says nine corrections, the number describes the text in the box, not the quality of the typing. It is also the only place your plateau information lives, since the test keeps nothing durable.

What a score from this test can tell you, and what it cannot

What it can tell you: whether you were accurate on one sample of English prose, how much of the passage you got through in sixty seconds, and whether the text you left in the box matched. That is a useful check on clean timed prose typing.

What it cannot tell you is most of what you would want from a training tool, and a misread number will send your practice in the wrong direction for weeks. There is no error count, so it cannot say how many mistakes you made, only how many you left in place. There is no per-key timing, no consistency or burst analysis, and no keyboard-layout analysis, so it cannot tell you which keys are slow, whether your cadence is even, or whether one hand carries more work. It keeps no history beyond your last five runs and a best score, all in memory and gone when the page reloads, so it cannot show a trend. And with twelve passages of about 230 characters, roughly 46 words each, it says nothing about how you type across a real document.

• Corrected mistakes are invisible. A 98% figure means the text you left in the box was right, not that you typed it cleanly: leave 196 of 200 characters correct and you get 196 / 200 = 0.98 however much backspacing got you there. Practising with the error count hidden can fool you, which is why the tally is part of the method.
• There is no paste guard, so pasting the passage and producing a huge number is possible. That measures your clipboard and teaches your fingers nothing.
• A fast typist can finish the passage with time still on the clock, and the history entry will still read 60s — so the label is no proof you typed for sixty seconds.
• Timed mode is a fixed 60 seconds with no other length; free mode ends when the passage is finished.
• No custom text and no non-English passages: twelve hardcoded English ones, drawn at random so the previous one never repeats.
• It cannot place you among other typists: bands of “average speed” quoted elsewhere are not measurements this tool makes.

Frequently asked questions