The evidence

The reading science behind Fovea, including the parts that argue against it.

Speed reading is one of the few consumer categories where the marketing and the literature point in opposite directions. Reading one word at a time is a real technique with real limits, and the limits are measurable. This page sets out what the research says, what Fovea does about it, and what Fovea does not claim. Every number here has a source, and every source is linked.

Reviewed 1 August 2026

What this page claims, and on what basis

1. 01

   Moving your eyes takes roughly a tenth of reading time, not most of it.

   A reading saccade covers about two degrees and takes around 30 ms; fixations last 200 to 250 ms. Rayner (1998).

2. 02

   RSVP does not make people read several times faster.

   In a direct comparison, the RSVP group finished at 209 wpm and the ordinary readers at 200 wpm, a difference that was not significant. Benedetto et al. (2015).

3. 03

   Reading speed and comprehension trade against each other.

   A review of the whole literature concludes that doubling or tripling reading speed with comprehension intact is unlikely. Rayner et al. (2016).

4. 04

   Stopping readers from looking back measurably costs comprehension.

   Masking each word once the eye passed it lowered comprehension, on ordinary sentences and not only on garden-path ones. Schotter, Tran and Rayner (2014).

5. 05

   RSVP costs literal comprehension and suppresses blinking.

   Literal comprehension fell from 72% to 60%, and blink rate from 8.5 to 4.67 per minute. Benedetto et al. (2015).

6. 06

   The speed at which comprehension breaks is a property of the text, not a constant.

   INFERENTIAL comprehension held to 350 wpm and dropped above it in one study, the distinction the paper itself draws; in another, the same words in meaningful order supported 524 wpm against 269 wpm scrambled. Di Nocera et al. (2018), Primativo et al. (2016).

7. 07

   Words are recognised fastest when the eye lands slightly left of their middle.

   Naming and lexical decision latencies rise by roughly 20 to 30 ms for each letter away from that position. O’Regan and Jacobs (1992), Brysbaert and Nazir (2005).

8. 08

   There is no evidence that colouring that letter helps you read.

   No study we are aware of tests it. Fovea colours it so you can see where the anchor is. The layout does the work.

How reading actually works

Reading is not a smooth sweep along a line. Your eyes sit still for a moment, jump, and sit still again. The still moments are fixations, the jumps are saccades, and you take in almost no information during a jump. In English, fixations last about 200 to 250 ms and each saccade moves the eye 7 to 9 letter spaces forward. Roughly 10 to 15 percent of those movements go backwards over text already passed. Those are regressions (Rayner 1998).

Rayner tabulates ten skilled readers on easy text: a mean fixation of 216 ms, a mean saccade of 8.1 character spaces, 11 percent regressions, and 308 words per minute. That is what ordinary competent reading looks like when you measure it, and it is the baseline any speed claim has to beat (Rayner 1998, Table 3).

Eye movement is about a tenth of reading time

This is the number that decides the whole argument, and it is the one speed reading marketing leaves out. A saccade of about two degrees, the size typical of reading, takes around 30 ms (Rayner 1998). A fixation takes 200 to 250 ms. So one complete cycle costs roughly 30 + 216 = 246 ms, of which the movement is 30 ms. That is 12 percent. Take the widest published ranges and the share still lands between about 7 and 17 percent.

The consequence is arithmetic, not opinion. If you eliminated every eye movement perfectly and changed nothing else, you would recover about a tenth of your reading time. You would not double your speed, because the other nine tenths is your brain identifying words and building meaning, and no display technique touches that. Any product claiming a multiple has to be getting it from somewhere other than saccades, and the only other place to get it from is comprehension.

RSVP does not multiply reading speed

The strongest single result here comes from a study built to test exactly this claim. Sixty people read part of a book either normally or with Spritz, the RSVP app that popularised the fixed-letter idea. Its developers had said that eliminating saccades would reduce visual fatigue and improve comprehension. The RSVP group finished at 209 words per minute. The ordinary readers finished at 200. The difference was not significant (Benedetto et al. 2015).

Nine words a minute is what the technique bought, and it was not paid for in nothing. Literal comprehension, meaning questions about what the text explicitly said, fell from 72 percent to 60 percent. Inferential comprehension, which leans on what the reader already knows, held up. Blink rate more than halved, from 8.5 to 4.67 blinks per minute, which the authors connect to the rise in visual fatigue they measured (Benedetto et al. 2015).

The review of the field arrives at the same place from the other direction. After surveying what is known about reading, Rayner and colleagues conclude that “it is unlikely that readers will be able to double or triple their reading speeds … while still being able to understand the text as well as if they read at normal speed” (Rayner et al. 2016). If a thorough understanding is not your goal, skimming will get you through faster with moderate comprehension. That is a real and useful thing. It is not the same thing as reading faster.

Studies of actual trained speed readers say the same. Monitored at 600 to 700 wpm, they matched normal readers on questions about the gist of a passage, but could not answer questions about details in regions their eyes had never landed on. When ordinary readers were asked to skim, their eye movements and their comprehension both came to look like the speed readers’ (Just, Carpenter and Masson 1982). Speed reading, once you measure it, turns out to be skimming with a better name.

Blocking regressions costs comprehension

This is the strongest argument against reading one word at a time, and it is a good one. Regressions look like inefficiency, and a stream that only moves forward removes them by construction. That is usually sold as a benefit.

It has been tested directly. Schotter, Tran and Rayner had people read normally or with a trailing mask: each word was replaced by a mask once the eye moved past it, so the text could be read but never re-read. Comprehension dropped. Not only on garden-path sentences engineered to be misread, which was the expected result, but on ordinary sentences too. The authors conclude that regressions contribute to understanding what you have read, and say plainly that this calls into question the viability of speed reading apps that eliminate eye movements (Schotter, Tran and Rayner 2014).

Some of that one movement in seven is idle doubt. A lot of it is repair: you took a sentence the wrong way, and you go back to fix it before the error compounds. Take repair away and you do not stop making the errors. You only stop correcting them.

This is the objection Fovea is designed around rather than against. Going back is a primary control, not a buried setting: release to stop the stream, tap the left edge to step back a configurable one to ten words, swipe left to scrub back word by word or sentence by sentence. And when a passage needs more than a rewind, the paginated view opens on the exact word you were on, with the paragraph around it, to be read the ordinary way for as long as you like. The cost the research identifies is real. It is just not one you have to pay.

The comprehension ceiling belongs to the text

There is a speed above which comprehension falls away, but it is not a single number, and treating it as one is wrong in both directions at once.

Di Nocera, Ricciardi and Juola put 209 participants through ordinary reading and RSVP at 250, 300, 350, 400 and 450 wpm. Up to 350 wpm they found no comprehension difference against normal reading. Above it, comprehension dropped significantly (Di Nocera et al. 2018). Worth being precise about what degraded, because it is the harder half: what they measured was INFERENTIAL comprehension, the questions whose answers are not stated anywhere in the text and have to be put together from it. Recognising what a passage said survives higher speeds than working out what it meant. Primativo and colleagues found the ceiling moving with the material rather than with the reader: presented with twelve-word sequences, the same participants managed 524 wpm on ordered meaningful text, 369 wpm when those words were scrambled, and 269 wpm on unrelated random words (Primativo et al. 2016). Same eyes, same display, same vocabulary. What changed was how much the text helped.

So the honest range for where RSVP comprehension starts to break is roughly 350 to 500 wpm, depending on what you are reading. Fovea marks that on its speed slider and estimates it per book from the text’s own lexical rarity and sentence length. It is a mark, not a cap: you can set any speed from 150 to 900 wpm, and the app tells you which side of the line you are on instead of deciding for you. We should be equally clear that no published curve maps a text measurement onto an RSVP ceiling. The two endpoints come from different paradigms, and interpolating between them is our engineering judgement, shown as a hint because that is all it is.

The optimal viewing position, and what we do not know about it

Words are not recognised equally well from every angle. The time it takes to name a word, or to decide whether a string is a word at all, depends strongly on where in the word the eye happens to be when it appears. There is a best position, and for languages written in the Latin alphabet it sits between the word’s first letter and its middle, slightly left of centre. Move away from it and naming and lexical decision latencies rise by roughly 20 to 30 ms per letter of deviation (O’Regan and Jacobs 1992). The effect is largely visual: acuity falls off sharply with distance from fixation, so letters further from the eye’s landing point are simply less available (Brysbaert and Nazir 2005).

This is the one place where a fixed-point reader has a genuine structural advantage, and it is worth being precise about what that advantage is. In ordinary reading your eye has to land somewhere in each word, and it does not always land well. In Fovea the word is delivered to the eye instead, positioned so that the letter at its optimal viewing position falls on the anchor you are already looking at. Fovea places that letter about a third of the way into a word’s real letters and never right of centre, and it positions each word by measuring the glyph widths either side of that letter rather than by centring the word. Centring is what makes the point drift, and the point drifting is what makes you hunt.

Now the part nobody else in this category writes down. There is no evidence that colouring the pivot letter helps you read. We are not aware of a study that tests it. Fovea colours it so that you can see where the anchor is while you get used to it, and for no other reason. The work is done by the layout, which holds the letter still. The colour only points at it. Turn the colour off and the mechanism is unchanged.

The second caveat is about the evidence itself. The optimal viewing position was measured on isolated words, with the eye already in place and the word appearing around it. No study we are aware of shows that delivering a whole book at its optimal viewing position improves comprehension, retention or comfort against ordinary reading. It is a plausibility argument, well grounded in how word recognition works, and it should be read as exactly that rather than as a demonstrated benefit.

Why the word interval is not flat

A fixed interval per word would be the obvious way to build an RSVP reader, and it would be wrong, because reading time per word is not constant. Two variables dominate how long the eye holds a word: how long it is, and how common it is. Fovea’s pacing curve holds long words for longer, adds a pause at commas, sentence ends and paragraph breaks, eases in over the first few words after you start, and gives rarer words slightly more time while giving common words slightly less.

The frequency term is deliberately mean zero across the corpus: rare words are held longer only because common words are shown shorter by exactly as much, weighted by how often each band actually occurs. Without that, the speed you selected would quietly become a ceiling nobody reaches, and the number on screen would be a lie. The same disclosure applies here as everywhere else on this page: nothing in the literature shows that imposing a frequency effect on a machine-paced stream improves comprehension or recall. It is a design argument, not a validated benefit.

One practical consequence, which is also why the reading times below are measured rather than calculated: because the interval is not flat, dividing words by speed understates how long a text takes by about a fifth.

How long a summary actually takes

Fovea’s catalog holds 605 original summaries, each with an English and a French edition. A typical summary runs about 6,000 words. The figures below are medians across the English set, produced by running the shipped pacing curve over each summary’s real text and taking the middle value. They are not words divided by speed, which would come out roughly a fifth too low.

The 700 wpm row is listed because the app allows it, not because we recommend it. It sits well above the range where the evidence puts the comprehension ceiling, and at that speed you are skimming. Skimming has its uses. It is not reading, and we are not going to relabel it.

Dyslexia: what we offer and what we will not say

Dyslexic readers do make longer fixations, shorter saccades and more regressions than typical readers. It is tempting to conclude that a reader which removes eye movements addresses the problem. The literature does not support that inference: Rayner’s review argues that eye movements are generally not a cause of reading disability but a reflection of other underlying difficulties (Rayner 1998). Removing a symptom is not treating a cause, and Fovea makes no therapeutic claim of any kind.

Fovea bundles OpenDyslexic and Atkinson Hyperlegible, and both are free forever, along with every other typeface, because charging for an accessibility affordance is not something we are willing to do. We will not oversell them either. Controlled studies of dyslexia-specific typefaces have not found reading rate or accuracy benefits over ordinary fonts (Wery and Diliberto 2017), (Kuster et al. 2018), and where a benefit has turned up it has traced back to letter and word spacing rather than to letterform design. They are there because some readers prefer them, and preference is a real reason. That is the whole claim.

References

1. 1.Rayner, K. (1998). Eye movements in reading and information processing: 20 years of research. Psychological Bulletin, 124(3), 372-422. Link
2. 2.Rayner, K., Schotter, E. R., Masson, M. E. J., Potter, M. C., & Treiman, R. (2016). So Much to Read, So Little Time: How Do We Read, and Can Speed Reading Help? Psychological Science in the Public Interest, 17(1), 4-34. Link
3. 3.Schotter, E. R., Tran, R., & Rayner, K. (2014). Don’t Believe What You Read (Only Once): Comprehension Is Supported by Regressions During Reading. Psychological Science, 25(6), 1218-1226. Link
4. 4.Benedetto, S., Carbone, A., Pedrotti, M., Le Fevre, K., Bey, L. A. Y., & Baccino, T. (2015). Rapid serial visual presentation in reading: The case of Spritz. Computers in Human Behavior, 45, 352-358. Link
5. 5.O’Regan, J. K., & Jacobs, A. M. (1992). Optimal viewing position effect in word recognition: A challenge to current theory. Journal of Experimental Psychology: Human Perception and Performance, 18(1), 185-197. Link
6. 6.Brysbaert, M., & Nazir, T. (2005). Visual constraints in written word recognition: evidence from the optimal viewing-position effect. Journal of Research in Reading, 28(3), 216-228. Link
7. 7.Di Nocera, F., Ricciardi, O., & Juola, J. F. (2018). Rapid serial visual presentation: degradation of inferential reading comprehension as a function of speed. International Journal of Human Factors and Ergonomics, 5(4), 293-303. Link
8. 8.Primativo, S., Spinelli, D., Zoccolotti, P., De Luca, M., & Martelli, M. (2016). Perceptual and Cognitive Factors Imposing “Speed Limits” on Reading Rate: A Study with the Rapid Serial Visual Presentation. PLOS ONE, 11(4), e0153786. Link
9. 9.Just, M. A., Carpenter, P. A., & Masson, M. E. J. (1982). What eye fixations tell us about speed reading and skimming. Eye-Lab Technical Report, Carnegie Mellon University. Reported in Rayner (1998) and in Just & Carpenter (1987), The Psychology of Reading and Language Comprehension. Link
10. 10.Wery, J. J., & Diliberto, J. A. (2017). The effect of a specialized dyslexia font, OpenDyslexic, on reading rate and accuracy. Annals of Dyslexia, 67(2), 114-127. Link
11. 11.Kuster, S. M., van Weerdenburg, M., Gompel, M., & Bosman, A. M. T. (2018). Dyslexie font does not benefit reading in children with or without dyslexia. Annals of Dyslexia, 68(1), 25-42. Link