A field guide to speed readers

The phrase “speed reading app” covers at least six different machines. Some train a habit. Some redraw the page. Some remove the page altogether. Some replace the text with a shorter text, and one family is not interested in speed at all. Telling them apart is the most useful thing anyone can do for a reader before they spend money, so here is the taxonomy.

We sell one of these things, which makes this a difficult piece to write straight. The obvious temptation is to arrange the field so that our own corner comes out looking inevitable. We have tried for the opposite. Every family below gets its best case made for it, and the family we belong to is judged hardest, because that is the one whose failure modes we know from the inside rather than from a review.

One fact constrains all six. Competent adults read continuous prose at roughly 200–300 words per minute, and that figure has been stable across a century of measurement. It is not a bad habit waiting to be corrected. It is what the machinery costs. Every tool in this field is therefore doing one of two things: reclaiming a small mechanical overhead sitting on top of that number, or quietly changing the task from reading into something else. Both are legitimate. Confusing them is not.

The six families

The families are defined by mechanism, not by branding. Products routinely belong to more than one — plenty of readers bolt a word stream onto a summariser, or ship a bold-prefix toggle inside a text-to-speech app. So the question to ask is never what the thing is called. It is: what has this removed from the reading task, and what has it put back in exchange?

• Technique trainers. Courses and drills that try to change how you use your eyes and your inner voice. The oldest family by decades, and the only one that sells you a habit rather than a piece of software.
• Word streams. One word at a time at a fixed point, at a rate you choose. The eye stops moving entirely. This is our family, and the one this article is least gentle with.
• Fixation aids. The page stays where it is, but the type is altered — most famously by bolding the opening letters of every word — to give the eye somewhere deliberate to land.
• Listening tools. Synthetic voices and audiobooks, which do not make reading faster so much as change the channel the words arrive through, and free your hands and eyes for something else.
• Summarisers. Human-written condensations and, more recently, machine-written ones. They do not make you read faster. They make there be less to read.
• Retention tools. Spaced repetition, highlight managers, annotation systems. The only family here that is not chasing speed, and the one with the cleanest evidence behind it.

Trainers, and the habit they attack

The technique trainers descend directly from the reading dynamics courses of the 1950s and 1960s, and the drills have barely changed. You run a finger or a pen down the page as a pacer. You practise taking in “chunks” rather than single words, widening what you believe your peripheral vision can deliver. You are told to stop saying the words in your head. The delivery has moved from a hotel conference room to a subscription app, but the instructions are the same instructions. Their history is the subject of a separate article.

What they genuinely buy is real, and it is not what the sales page says. A pacer imposes a steady rate on a reader who was drifting, and most adults do read well below their own comfortable ceiling out of habit alone. The training gives people permission to skim strategically and a vocabulary for triage. Where the evidence stops is at the mechanism: the perceptual span is a property of the visual system rather than a muscle, and suppressing subvocalisation costs comprehension on difficult material, because the phonological route was doing work. Best suited to readers with large piles and no method.

Word streams, including ours

Rapid serial visual presentation puts one word at a fixed point and replaces it on a timer. Spritz brought the idea to a wide audience, and Reedy, Outread, the flashing mode inside Spreeder and a long tail of browser extensions all work on the same principle. The mechanical saving is genuine and easy to state. Reading is saccadic — fixations of roughly 200–250 milliseconds separated by jumps during which you see nothing — and something like ten to fifteen per cent of reading time is spent in motion. Remove the movement and you get that back.

Then the evidence stops, quite abruptly. The bottleneck in reading is language processing, not vision. When eye movement is taken away entirely, reading gets somewhat faster, not multiples faster, and this is the single finding that ends most of the claims made for this family. Past roughly 400–500 words per minute on unfamiliar prose, comprehension slides. What you are doing above that line is skimming — a useful skill, honestly acquired, but not reading.

The costs are structural and we cannot design them away. A word stream cannot be re-read: skilled readers make a backwards saccade on roughly one saccade in seven, and many of those are comprehension failures being repaired in flight. It cannot be skimmed ahead, so you cannot glance at the end of the paragraph to find out where the sentence is going. It destroys spatial memory of the page — the vague sense that the thing you want was two-thirds down on the left, which is how a great many people navigate documents they have already read.

Removing eye movement buys real time. It is a smaller purchase than the category advertises.

The word-stream family, summarised

Three families with their own articles

Fixation aids leave the paragraph intact and change the letterforms. Bionic Reading is the well-known instance, and there are now many relatives: the opening letters of each word are bolded to create artificial anchors, on the reasoning that word recognition leans heavily on initial letters. It is enormously popular and very widely described as feeling easier. The controlled studies that exist have largely failed to find a reliable speed or comprehension benefit, with some finding a small cost — a gap between subjective ease and measured gain that deserves its own treatment rather than a dismissal.

Listening tools — Speechify, the voices built into every operating system, and the whole audiobook business — are the family least often counted as speed reading and arguably the most useful. They remove decoding entirely and hand the material to a different sense, which is why they can be used while walking, driving or cooking. The ceiling sits near the rate at which speech stays intelligible, and going back to a sentence you missed is clumsy in a way that turning a page is not. Compared properly in the listening article.

Summarisers are the honest outlier, because they never pretended to speed up your reading. Blinkist-style human condensations and the newer language-model ones both hand you a shorter text and someone else's judgement about which parts mattered. Sometimes that judgement is exactly what you needed, and triage of this kind is a genuine service. What you cannot do is treat the output as having read the original — the argument, the evidence and the qualifications are precisely what compresses away. Taken apart in its own piece.

The family that attacks forgetting

The sixth family barely belongs in this article, which is why it is worth putting here. Spaced repetition systems, highlight managers and annotation tools such as Readwise and its relatives are built on a different premise entirely: the problem is not that you read too slowly, it is that three weeks later you cannot recall what the chapter argued. Their central mechanism is the testing effect, which is about as well established as anything in this field — being tested on material retains it better than reading it again.

Their honest ceiling is that they make reading slower on purpose, and they demand a habit rather than a session. They also make an uncomfortable point about the rest of the shelf. If the thing you actually want is to still know this in a month, a tool that halves your reading time and leaves retention untouched has not helped you much. We took that seriously enough to build the comprehension check into the app and ship it switched on.

Ask what a tool removed from reading, and what it put back. The name tells you nothing.

How to read a product page

Where we sit

Reader Inc. is family two. We do not get an exemption from the family's weaknesses because we added something to it. Our stream cannot be re-read, cannot be skimmed ahead, and leaves you with no spatial memory of the document. On syntactically dense prose — a contract with three nested conditions, a proof, a paragraph of philosophy — that is a serious handicap, and the correct response is often to close the app and read the page normally, slowly, with a pen. We would rather you did that than distrust the tool everywhere.

What we will defend is narrow. The motion saving is mechanical and real. The pairing of every word with a sense-chosen image is our attempt to put something back where the other members of this family put nothing, and it rests on findings we did not invent but a combination nobody has yet tested. And the recall test at the end of a passage exists so that you can find the point where your own comprehension started falling, on your own material, and hold that number against ours. If the number disappoints you, the honest conclusion is the one we would draw too.