What words per minute actually measures

The biggest control in our app is a slider labelled in words per minute, which makes us part of an industry that has spent a century selling that unit rather harder than it deserves. So it is worth being precise about what the number is. Words per minute is a rate. It tells you how fast text moved past a reader. It does not tell you what arrived. On its own, it measures nothing worth having.

Start with the baseline, because it is one of the most stable numbers in the study of reading. Competent adults read continuous prose at roughly 200 to 300 words per minute, and the figure has held across a century of measurement — different laboratories, different decades, different texts. When someone asks what the average reading speed is, that range is the honest answer, and the width of the range is part of the honesty. Where a given reader sits inside it depends on the material, their familiarity with it, and what they are reading for.

The stability matters more than the value. If reading speed were mostly a matter of habit, a century of schooling, print culture and commercial ambition would have shifted the average by now. It has not shifted, because the constraint is not on the eyes. When eye movement is removed entirely — which is what RSVP does — reading gets somewhat faster, not multiples faster. That finding, replicated across many laboratories, is the one that ends most speed-reading claims. The bottleneck is language processing: parsing each clause, integrating it with the last, updating the model of what the text is saying. A display can feed words in faster than that machinery runs. It cannot make the machinery run faster by feeding it.

What a reading speed test actually measures

A reading speed test, as usually administered online, times how quickly your eyes can pass over a passage before you press a button. That is a genuine measurement of something — call it traversal. What it is routinely reported as is reading, and the gap between the two is exactly as wide as the comprehension check the test did or did not include. A traversal number with no comprehension attached is a statement about the display and the eyes. It says nothing about the head behind them.

Every rate has a numerator, a denominator and an unstated assumption that all the units counted. Words per minute inherits the assumption silently: it counts a word as read because it was present on screen while the clock ran. But a word can be present and skipped, present and half-parsed, present and gone by the time the sentence needed it. Whether it counted is not observable from the timer. A rate only becomes a score when an outcome is attached to it, and comprehension is the outcome.

None of this makes the unit useless. A rate is exactly the right thing to hold steady while you vary something else, which is how honest experiments work. It is the right thing to creep upward in small increments during practice, which is how honest training works. The failure mode is not measuring wpm. It is quoting wpm alone, as if the number certified that reading happened. It certifies that time passed.

The 900 words a minute claim

Which brings us to the person who reads at 900 words per minute. There is such a person; you may have met them; they may be you. What the eye-movement and comprehension literature says, quite consistently, is that beyond roughly 400 to 500 words per minute on unfamiliar prose, comprehension falls away, and what continues past that point is skimming — sampling the text, reconstructing the gist, trusting prior knowledge to fill the gaps. Skimming is a legitimate and genuinely useful skill. It is also a different activity, and the sentence “I read at 900 wpm” is a claim about it.

The tell is what happens when the material stops cooperating. On familiar ground — a news story shaped like a hundred previous news stories — skimming performs impressively, because most of the text was predictable and prediction is what skimming runs on. Hand the same reader a dense contract, an unfamiliar argument, a proof, and the 900 collapses back towards the ordinary range, or the comprehension does. The people and products that populate our field guide to speed readers differ mainly in how candidly they describe that collapse.

• Sampling replaces reading: the eyes land on some words and infer the rest, which works in proportion to how predictable the text is.
• Syntax goes shallow: long dependencies — the clause that modifies something four lines back — are the first thing lost at speed.
• Memory shifts to gist: the reader can tell you what the passage was about, and increasingly not what it said.
• Regressions vanish: on a page, skilled readers look back roughly once in seven saccades, and many of those are comprehension repairs. At skimming rates there is no time to notice the failure, let alone repair it.

Past a certain rate you are not reading faster. You are reading less, faster.

The trade-off, stated plainly

Our own number is not quite honest either

Now the part where the criticism lands on us. Set our slider to 300 words per minute and the app will not show you a word every 200 milliseconds like a metronome. The dwell on any given word stretches — up to 2.8 times — on long words, on commas, at sentence ends, at paragraph breaks, and on numerals. A short function word in the middle of a clause takes something like the nominal time. The word notwithstanding, or the figure 14,700, or the last word before a full stop, takes considerably more.

We do this because prose is not uniform and pretending otherwise is worse. A number needs checking against expectation. A clause boundary is where integration happens, and integration takes time; the eye-movement record of ordinary reading shows exactly the same stretching, fixations lengthening at the same places. A true fixed cadence — the pure metronome — delivers its worst service precisely where the text needs the most work, which is the design equivalent of a bridge rated for average load.

But the consequence is that the number on the slider is not the number in your head. Nominal 300 is an upper bound; the delivered average over real prose runs below it, and how far below depends on the text — more numerals, more punctuation, longer words, slower delivery. Two sessions at the same slider setting are not the same speed. We could hide this by racing between stretches to make the average come out exactly; that would keep the label honest and the experience worse. We chose the better experience and an overstating label, and you should know the label overstates.

Why recall sits next to speed

This is also why the number that follows a session in the app is not one number. After a passage you can take a recall check — real targets from the text mixed with plausible distractors that were never there — and the score it produces is shown next to the speed, not beneath it, not behind a tap. Sessions export with both. The pairing is the point. Speed is the thing you set; recall is the thing you find out; and the second column is the only reason the first one means anything.

The check itself is deliberately unflattering — distractors are chosen to be the kind of word that plausibly could have appeared, so that a warm feeling of familiarity is not enough to score well. How to run it fairly on yourself, and what a single passage can and cannot tell you, is its own subject, covered in testing your own comprehension.

Here is the whole argument in two sentences. A rate is a promise about the display; a score is a fact about you. Any product, page or person quoting words per minute reading speed without an outcome beside it is telling you how fast the fuel left the tank, and nothing at all about whether the engine was running.