Rohman
Almost everything written about how to read faster is written by someone selling a course. So here's the finding that usually doesn't make it into the sales page: when researchers put a one-word-at-a-time RSVP app in front of an eye tracker and measured it directly against ordinary reading, the speed gain rounded to noise. The comprehension loss didn't (Benedetto et al., 2015).
That is the honest starting point for this piece, and it still isn't an argument against the technique. It's an argument for knowing exactly what the technique is, and isn't, for.
What RSVP Actually Does
Rapid Serial Visual Presentation (RSVP) is a reading-display method that shows text one word, or a short chunk of words, at a single fixed point on the screen, instead of letting your eyes travel across a line. Your eyes stop moving; the text moves through them instead.
It began as a laboratory method for studying attention and language processing, decades before anyone tried to sell it as a faster way through a novel. The mechanism is genuine: by pinning every word to the same spot, RSVP removes saccades (the small jumps your eyes make between words) and regressions (the backward flicks you make when something didn't land). It also lets a reader line up each word on its optimal recognition point — the letter position where a word gets decoded fastest, roughly a quarter to a third of the way in from the left, a perceptual effect well documented in eye-movement research (see the review by Rayner, 1998) and the same rule OxygenPDF's reader uses to pick which letter to highlight in each word.
So far so good. The problem is the arithmetic of what removing eye movement actually saves you.
The 10% Ceiling
Your eyes are not the bottleneck. Fixations — the pauses where you actually process a word — run about 200 to 250 milliseconds each. Saccades, the jumps between them, take 20 to 40 milliseconds. Across ordinary reading, eye movement itself accounts for roughly a tenth of your total reading time; the other nine-tenths is your brain doing the work of recognising words and building meaning (Rayner, 1998).
Remove 100% of the eye movement and you recover about 10%. Yet RSVP tools routinely advertise two to four times faster. That extra speed has to come from somewhere, and it comes out of the part you cannot safely cut: processing time, and regressions.
Those regressions matter more than they sound. They happen something like 10 to 15% of the time in normal reading (Rayner, 1998), and when researchers built an experiment specifically to stop readers from ever looking back, comprehension fell (Schotter, Tran, & Rayner, 2014). Re-reading is not a bad habit you should train away. It's the repair mechanism that makes reading work, and RSVP removes it by construction — there is no "look back" once the previous word is already gone.
Pick a Gear, Not a Number
The most useful reframe I found is that reading isn't one speed you optimise. It has modes, and each has a natural rate — a distinction the reading researcher Ronald Carver spent much of his career mapping under what he called rauding theory (Carver, 1982).
Memorising runs around 138 wpm. Studying something you'll be tested on, about 200. Ordinary comprehension reading sits at roughly 238 wpm for English non-fiction — the measured average from a meta-analysis of 190 studies and more than 18,500 readers (Brysbaert, 2019), not the 300 wpm figure that circulated through textbooks for decades on much thinner evidence. Skimming reaches about 450, and scanning for a single fact, about 650.
Notice where the marketed range begins. Speed-reading products sell 500 to 1,000 wpm as a reading speed, when 450 and up is already the territory of skimming and scanning by Carver's own numbers. They aren't making you read faster. They're quietly changing what "reading" means, and charging for it.
Rayner and colleagues went through this literature carefully, in a review written specifically to evaluate speed-reading claims, and concluded that rates above 1,000 words per minute aren't achievable without severe loss of understanding (Rayner, Schotter, Masson, Potter, & Treiman, 2016). The durable gains come from something much less marketable: reading more, across varied material, until word recognition gets automatic.
The Settings That Matter More Than Speed
The wpm number gets all the attention, but a handful of other settings decide whether a passage is actually legible a few hundred milliseconds at a time:
- Smart pausing. A sentence-ending period, question mark, or exclamation point buys roughly two and a half times the base delay; a comma or semicolon buys about one and a half times; long words and numbers get a smaller bump of their own. Without this, every word gets the exact same sliver of screen time regardless of how much thinking it demands — which is where RSVP stops feeling like reading and starts feeling like a strobe light.
- Slow start. The first handful of words play back slower than your target pace and ramp up to it, instead of opening at full speed before your eyes have even found the display.
- Chunk size. Showing two or three words at once trades away some of RSVP's saccade-elimination for something closer to normal phrase-level reading, which suits familiar material better than strict one-word mode.
- Hyphen joining. A PDF exported from a justified layout — the kind academic journals and some scanned textbooks use — breaks words across the line: "exam-" on one line, "ple" on the next. Left alone, a word-stream reader flashes those as two separate, meaningless fragments. A dehyphenation pass rejoins them before either half ever reaches the screen.
Get these wrong and a technically-working RSVP reader still feels unusable. Get them right and the wpm number stops being the only thing you have to fight with.
Where It Genuinely Earns Its Place
None of that makes RSVP useless. It makes it specific.
It works on short, linear text, the same kind early studies used when they found it matched ordinary reading. It helps when the screen is small enough that a normal column is awkward. For some readers, pinning each word to one fixed position removes a specific kind of effort that ordinary line-tracking demands — the evidence there is genuinely mixed rather than settled, and it isn't a substitute for an actual reading assessment if reading itself is the persistent difficulty. And it is good at one particular thing: pulling you through a backlog at a forced pace so you stop re-reading the same paragraph while thinking about something else.
That last one is the real draw, and it is a discipline benefit rather than a perceptual one. A pace you cannot drift away from is worth something on a Tuesday afternoon.
Where it fails is predictable from the mechanism. Dense technical prose overloads it. Tables, code, citations and footnotes are structurally incompatible with a strictly linear word stream. There's no "jump back to the header row" in a format that shows you one word. And anything you need to retain in literal detail is exactly where the measured comprehension gap showed up.
Contracts, manuals, anything you will be held to: read those the slow way.
Speed Reading a PDF
- Open the Speed Reader
- Drop in a PDF, Markdown, or plain-text file, or paste text straight in
- Start at 300 wpm and let the slow-start ramp ease you in
- Press space to pause the moment something doesn't land
Speed runs from 50 to 1,000 wpm. The up and down arrow keys nudge it 25 at a time (50 with Shift held), which is the control that matters most — you want to be adjusting constantly, not picking a number once and living with it. Left and right skip ten words at a time (fifty with Shift), and comma and period step exactly one word, which is the precise way to back up after a pause instead of guessing how far to jump. Chunk size shows up to three words at once, which suits familiar material better than strict one-word mode does. A "Join hyphenated words" toggle exists specifically for justified PDFs that break words across line ends. Press Z for a distraction-free zen overlay that clears everything but the word itself and a way out — Escape or Z again brings the controls back.
A reasonable way to use it: run the first pass at a pace that feels slightly too fast, mark what you want to revisit, then go back and read those parts properly. Triage, then read. That plays to the format's strength and avoids its weakness entirely.
If your source is a Markdown file, the reader strips the syntax before streaming — headers, bold markers, and link brackets don't show up as literal punctuation flashing past. Paste raw text in directly and the same tokenising and pausing rules apply, no file required.
The file is parsed in your browser and never uploaded; it's the same local-first model the rest of the toolkit runs on. For the documents people work through — unpublished drafts, case files, medical records, anything under NDA — that is not a small detail. A backlog is a strange reason to hand a stranger your reading list.
Three Things Worth Doing First
Crop the margins if you'll also read it normally. A journal PDF on a phone is mostly whitespace, and Crop PDF gets the text filling the screen.
Check it has a text layer, too. A scanned PDF has no words to stream, only pictures of words. Run OCR first — the same fix covered in our guide to turning a scan into searchable text — and it becomes readable text a stream can actually use.
And if you only need one section out of a long report, extract those pages before you start streaming. RSVP has no table of contents to jump to mid-read; pulling the right pages first beats skipping through the whole document with the comma key.
Frequently Asked Questions
Does speed reading actually work?
It depends what "work" means. If it means reading faster with no comprehension cost, the eye-movement research says no — speed and comprehension trade off against each other past a fairly low ceiling, and gains marketed at two to four times normal speed come from cutting comprehension, not from an efficiency the technique unlocks for free (Rayner et al., 2016). If "work" means a forced, non-negotiable pace through a backlog you'd otherwise stall on, then yes — that's a real, if modest, use.
Is RSVP the same thing as speed reading?
Not exactly. Speed reading is the marketing category; RSVP is one specific mechanism inside it — flashing single words, or short chunks, at a fixed screen position instead of letting your eyes move across a line. Other techniques sold under "speed reading" (eliminating subvocalisation, widening peripheral vision, reading in chunks with a moving guide) are older, separate claims with their own research record. RSVP is the one this article, and OxygenPDF's tool, are actually about.
What reading speed should I start at?
Below where you think you should, not above. Something like 250 to 300 wpm sits close to ordinary silent-reading pace for English non-fiction (Brysbaert, 2019), which makes it a reasonable place to judge whether you're actually absorbing the material before pushing higher. Nudge it up 25 wpm at a time and watch your own comprehension, not the number on screen — the right speed is the fastest one where you could still explain the last paragraph back to yourself.
Can RSVP help with dyslexia or other reading differences?
For some readers, yes, though the evidence is mixed rather than settled. Pinning each word to one screen position removes the need to visually track a moving line, which is a specific kind of effort some readers find disproportionately hard. It isn't a treatment, and it doesn't work the same way for everyone who tries it — a persistent, significant reading difficulty is worth raising with someone qualified to assess it, not solving with a settings toggle.
Does it work on a scanned PDF?
Not directly. RSVP needs actual text to tokenise and display, and a scanned page is a picture of words with nothing for the reader to extract. Run it through OCR first to add a real text layer, and the speed reader — along with copy-paste, search, and everything else that depends on text — works normally.
Is there a real ceiling on how fast a person can read?
Lower than the marketing suggests. The review most focused on this exact question concluded that reading rates above 1,000 words per minute aren't achievable without a severe loss of comprehension (Rayner et al., 2016) — which rules out treating the very high figures some apps still advertise as literal reading speeds, rather than very fast skimming relabeled as reading.
Does OxygenPDF's speed reader upload my file anywhere?
No. Text extraction, tokenising, and playback all happen in your browser; nothing is sent to a server. That matters more for this tool than most, since the documents people reach for a speed reader with — a backlog of drafts, reports, or a reading list — are often exactly the ones they'd rather not hand to a third party just to get through faster.
The point of any of this is not to hit a number. It is to spend your slow, expensive attention on the parts that deserve it, and move briskly through the parts that do not.
Try the speed reader — your file stays in the browser.
