Bionic Reading is a typographic technique that sets the opening letters of every word in bold and leaves the remainder in normal weight. Under the method’s default rule, a word with an even number of characters has its first half bolded, and a word with an odd number has the first half plus one additional character bolded. The rationale attached to it is that the darker openings tell the eye where to land, so a reader can take in the salient first letters and let the rest of each word resolve without a separate look at it.

A registered report published in Royal Society Open Science on 2 September has now put that rationale in front of an eye tracker, and the bolding did one thing. It pulled the reader’s first fixation closer to the start of long words. It did not make them read faster, skip more words or understand more.

Adam Parker and Hyunjun Lee of University College London, Haibei Wang of Royal Holloway, University of London, and Holly Joseph of the University of Reading recorded the eye movements of 90 skilled adult readers working through single sentences in three typefaces. The method they were testing is distributed as a set of apps for phones, desktops and the Chrome browser, along with an API the company licenses to other products. The researchers note that uptake is large enough to register as more than 100,000 downloads on the Google Play Store.

Earlier tests of the same idea

This is not the first attempt, and the results have been converging for a while. A web experiment run by the reading app Readwise timed more than 2,000 people and found them 2.6 words per minute slower with the bolding, a difference that was not statistically significant. The new paper adds a caveat of its own to that one: the sample’s reported reading speeds ran from 1 to 2,000 words per minute, which makes it hard to judge how engaged the readers were.

A 2024 paper in Acta Psychologica by Joshua Snell, titled “No, Bionic Reading does not work” (doi:10.1016/j.actpsy.2024.104304), had 32 participants read 100 paragraphs and found no reliable advantage among fast and slow readers alike.

Three eye-tracking studies preceded this one. Spear and colleagues, examining boldface letters during paragraph reading in 2025, found that start-bolded text produced shorter individual fixations but more of them, leaving reading times on individual words unchanged while the overall rate came out lower in words per minute. They found no landing-position differences between bolding conditions.

A 2025 study by T. R. Beelders, tracking gaze under the Bionic font, is cited by the new paper for a null effect of bolding the first few characters of a word. Možina and colleagues, testing the technique across different media, reported the same shorter-but-more-numerous fixation pattern Spear had.

Against that sits an unpublished study of 12 readers conducted through a Swiss institute, which reported largely positive effects. The new paper notes that it had not been shared as a preprint or a peer-reviewed article, that it found no benefit for one subgroup, and that it is not clear whether its gains were objective or simply how the readers felt.

The pattern is familiar from other attempts to make fluent reading faster by changing how text looks. Rapid serial presentation tools, which flash words one at a time in a fixed spot, did increase speed but cost comprehension in the process, and specialist dyslexia typefaces have repeatedly failed to outperform ordinary ones.

The control condition that separates two explanations

Each participant read sentences with a four-letter or an eight-letter target word embedded in them. Three conditions were compared: ordinary unbolded text, text with the first half of each word bolded in the Bionic Reading style, and a control condition with the middle of each word bolded instead.

That third condition is the design’s real work. If only the start-bolded text changed behaviour, the effect belongs to the initial letters. If any bolding changed it, the effect is general visual disruption rather than anything to do with where a word begins.

The 90 target-word pairs were matched on word frequency, the sentences averaged about 13 words, and everything appeared in 18-point Courier New on a single line, which is a narrower reading situation than a page of continuous prose. Eye movements were recorded with an EyeLink 1000 Plus sampling a thousand times a second, with participants’ heads steadied on a chin and forehead rest.

The recruitment plan was registered in advance, and it is what makes the null result readable. The team committed to an initial 42 readers and to adding participants in batches of 12 until the Bayes factors on every effect became decisive, stopping at a ceiling of 90. The evidence never became decisive on every effect, so the study ran to that ceiling. Ninety-three readers were tested, three were dropped for poor tracking during calibration and validation, and 90 remained, with a mean age of 22.9 years.

The first fixation moved back on eight-letter words

Readers tend to land slightly left of a word’s centre, at what reading researchers call the preferred viewing location. The long-standing explanation is that they aim at the centre and fall short, though whether that left-of-centre preference is a targeting error or a target in its own right is still disputed. Word length shifts the landing point: the longer the word, the further right the first fixation falls.

That length effect is where the bolding showed up. Comparing unbolded and start-bolded text, the interaction between word length and typeface was decisive, with the length effect weaker under bolding. In plain terms, bolding the opening letters dragged the first fixation back towards the beginning of eight-letter words, and did less to four-letter words.

The precision here matters. The paper’s simple comparison of landing positions between start-bolded and unbolded text did not settle anything on its own: its Bayes factor sat inside the indecisive band. The finding is specifically about long words.

The middle-bolded control is unsettled in the same way. Its landing-position comparison against unbolded text returned a Bayes factor of 0.52, also inside the indecisive band, so it neither confirms nor rules out a shift of its own. What was decisive there was the absence of any interaction with word length.

No gain in speed, skipping or comprehension

Skilled readers skip roughly a third of the words on a first pass through a text, and the words they skip tend to be short, common and predictable. If the bolding were doing its advertised job of cueing the eye, skipping rates ought to move. They did not: comparing start-bolded and unbolded text, the evidence pointed to no difference.

Nor did anything move on the timing measures. Gaze duration, which is how long the eye spends on a target word on its first pass, came back the same. So did total reading time on the target word, and so did the time taken over the whole sentence. Comprehension, tested with a true-or-false question after every sentence, ran at about 95 per cent correct across the board, and the start-bolded and unbolded conditions were indistinguishable.

Every one of those comparisons produced Bayes factors below one third, the threshold the authors registered in advance as evidence for the absence of a difference rather than a mere failure to find one.

That threshold is worth holding at its actual strength. A Bayes factor below one third is positive evidence for a null, not proof of one, and the study is open about where its evidence is thinner. Several comparisons, mostly involving the middle-bolded control, came back indecisive rather than clean, including the one on word skipping. The comprehension scores sat close to ceiling, which leaves little room for any effect to appear at all, so a comprehension benefit is better described as unsupported, and in the authors’ view unlikely, than as ruled out.

Plain text topped the readability ratings

Participants rated each typeface for readability on a ten-point scale, a check the authors added because perceived readability is often taken as a benefit in its own right even where objective speed does not move.

Of the 90 readers, 43 named the unbolded text the most readable, 31 chose the start-bolded text and 16 chose the middle-bolded version. That is a plurality rather than a majority, though the paper’s own discussion calls it the majority and its conclusion says most participants. Mean ratings ran 7.48 for unbolded, 6.76 for start-bolded and 5.43 for middle-bolded.

The researchers then checked whether a reader who happened to like a typeface read faster in it. No such relationship appeared on any of the reading-time or eye-movement measures.

One further limit bounds what any of this settles. Bionic Reading now lets users adjust how much of each word is bolded and in which font, and none of those individualised settings was tested here. Practice is a possibility the authors raise and then rate as less likely: Spear and colleagues found the effect did not grow over the course of their experiment.

The sample is the limit worth holding onto, because it is where the interest in the technique actually sits. These were skilled adults aged 18 to 40 with no history of reading difficulty or neurodevelopmental conditions, which is the group with the least room to improve. University students with dyslexia or broader reading difficulties are a group for whom a low-cost, easily implemented reading support would be genuinely welcome, and this experiment cannot speak to them.

The authors are explicit on that point, and on a second one behind it. Theoretical considerations counsel caution about expecting substantial benefits there, because the causes of reading difficulty are unlikely to be reached by increasing the visual salience of text. The paper names phonological processing, vocabulary and higher-level language skills among them.

Whether individualised bolding settings or other fonts change the picture is testable, and untested. What this study settles is narrower and still useful: for skilled adults reading single sentences in Courier New, bolding the front of every word moved the first fixation on the long words and produced no measurable change in anything that happened afterwards.