Speed reading 2.0: why EyeStill doesn't show every word for the same time
A metronome doesn't read. Classic speed-reading apps flash each word for the same fraction of a second; EyeStill gives every word the time it needs, and still keeps the speed you chose.
The idea, and the catch
Rapid serial visual presentation — RSVP — shows a text one word at a time in the same spot. Your eyes stop travelling along lines and jumping back, and a lot of the time normal reading spends on moving the eyes is gone. Most speed-reading apps are built on it, and most of them work like a metronome: at 300 words per minute every word stays on screen for exactly 200 milliseconds, perhaps with a little extra after a full stop.
The catch is that people don't read like that. Eye-tracking research has measured for decades how long the eyes rest on each word, and the answer is: it depends. A short, common, expected word gets a glance or is skipped altogether. A rare word, a name, a surprising turn gets a longer look. At the end of a sentence the reader pauses to put it together. A metronome gives the as much time as photosynthesis — too much for one, too little for the other.
There's a second catch, and we'd rather say it ourselves. A large review of speed-reading research (Rayner and colleagues, 2016) found the same trade-off every time: push the speed up and understanding goes down, because what limits reading is the language, not the eyes. No app turns anyone into a 1,000-words-a-minute reader with full understanding, and EyeStill doesn't promise that. What it does is spend the time you give it where the text needs it.
Every word gets its own time
Before you read a passage, the reader analyses the next few hundred words and gives each one a weight. Here is what goes into it.
How rare the word is
The best single predictor of how long the eyes stay on a word is how often it appears in the language — the word frequency effect. EyeStill carries frequency data for the 30,000 most common word forms in English and in Russian (from wordfreq, by Robyn Speer and colleagues) and holds rare words longer than common ones. Long words and numbers get a little more too.
How surprising it is right here
A common word can still be unexpected in its place. Psycholinguists call this surprisal: how unlikely a word is given what came before it, measured in bits (Hale, 2001; Levy, 2008). A word you could have guessed is shown about 20% shorter; an ordinary one at the base time; a surprising one up to 45–60% longer. For Russian, EyeStill also knows the usual patterns of the language — a preposition followed by a noun, an adjective by its noun — and the most common word pairs, so these run through quickly. Rarity and surprise are combined and capped, so no single word freezes the line.
Where the sentence ends
Readers pause at the end of a sentence to integrate it — the wrap-up effect (Just and Carpenter, 1980). The pause in EyeStill isn't a fixed "extra half-second after a full stop": it grows with the length of the sentence and the number of clauses in it, and is a little longer at the end of a paragraph. Inside a sentence, a comma after a short introductory word gets a breath; a comma closing a long clause gets more.
What you've already seen
In normal reading the eyes pick up the next short word before they land on it. The drum shows the following words to the right of the focus point, so a short, predictable word you've already glimpsed is shown about 13% shorter.
What the story is doing
A small language model (RuBERT-tiny, running in your browser — the text never leaves it) compares neighbouring sentences. When the scene, the place or the subject changes sharply, the reader holds that moment longer so you can re-orient. In dialogue it slows a little when the speaker changes and glides through "he said" and "she replied". Words that carry the passage — the names and motifs that keep coming back — get a moment more.
…and the speed is still yours
If every adjustment only added time, 300 words per minute would quietly become 250. So EyeStill keeps a running average of how much time it has added. After a dense stretch it lets the next predictable words go a little faster, until the average is back at the speed you set. The overall pace stays yours; only its distribution changes. In psycholinguistics this is close to the idea of uniform information density (Levy and Jaeger, 2007): an even flow of information, not an even flow of words.
| What happens | Effect on time |
|---|---|
| A predictable word in context | about −20% |
| A short word already seen on the drum | −13% |
| A rare or surprising word | up to +45–60% |
| A key word of the passage | +28% |
| A new speaker in dialogue | +45% |
| A sharp change of scene | +70% |
| End of a sentence | grows with its length |
| Any single word, at most | 0.55× to 3.6× the base time |
Starting, stopping and going back
A metronome starts at full speed. EyeStill eases in: after a pause the first word stays 60% longer, the next three a little less each, until you're back in the flow. If you step back to re-read, the next few words slow down by 45% so you can catch the thread again.
When you pause, you aren't left with a single word: a card shows the whole sentence around it, and you can click any word to jump there. Going back is a known weak point of one-word reading — the eyes can't simply glance back — so EyeStill makes it one click.
Where you meet it
- The reader at eyestill.com/read: books, articles and files — everything described above.
- AI answers in ChatGPT, Claude, Gemini and DeepSeek, through the Chrome extension: read while the answer is still being written.
- Video subtitles: there the speaker sets the pace. EyeStill follows the timing of each spoken word and puts it at the speaker's lips instead.
- Inside Claude: read books and answers in the chat and ask about them.
What EyeStill is not. It doesn't promise to triple your reading speed. It's a way to read with your eyes still, at a speed you choose, with the time spent where the text is hard and saved where it's easy. Start at 300 words a minute and go up only as long as the meaning keeps up.
Further reading
- 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).
- Rayner, K. (1998). Eye movements in reading and information processing: 20 years of research. Psychological Bulletin, 124(3).
- Just, M. A., & Carpenter, P. A. (1980). A theory of reading: From eye fixations to comprehension. Psychological Review, 87(4).
- Hale, J. (2001). A probabilistic Earley parser as a psycholinguistic model. NAACL.
- Levy, R. (2008). Expectation-based syntactic comprehension. Cognition, 106(3).
- Levy, R., & Jaeger, T. F. (2007). Speakers optimize information density through syntactic reduction. NIPS.
- Zwaan, R. A., & Radvansky, G. A. (1998). Situation models in language comprehension and memory. Psychological Bulletin, 123(2).
- Speer, R. et al. wordfreq. github.com/rspeer/wordfreq