The bold-prefix family

Somewhere around the early 2020s a large fraction of the internet discovered that if you bold the first few letters of every word, text feels different to read. Screenshots travelled. Browser extensions appeared within days. It became, very quickly, one of the most widely adopted reading interventions ever released — and one of the least supported by controlled measurement. Both of those facts are true at once, and neither cancels the other.

The best-known instance is Bionic Reading, a method developed by a Swiss typographic designer and made widely known through an app and a licensable service. It has since acquired a large family of relatives: open-source scripts, reader-mode toggles, e-reader settings and a great many one-afternoon browser extensions that do approximately the same thing. The rendering differs in detail — how many letters are emphasised, whether the weight is graded, whether short words are treated separately — but the idea underneath is single and clear.

We have an interest here, and it is worth stating early. We are not in this family; we are in the word-stream family, and the two are frequently confused because both do something to the shape of a word before it reaches your eye. We also use a mechanism that sounds like a cousin of bold prefixing and is not, which is the second half of this article.

What the bolding is for

The reasoning behind bold prefixes is not silly. Reading is saccadic: the eye fixates for roughly 200–250 milliseconds, jumps seven to nine characters, and fixates again, and where it lands is chosen fast and largely without conscious control. Word recognition also leans heavily on the opening letters — the beginning of a word carries more identifying information than the middle, which is why a prefix is often enough to trigger recognition before the rest of the word has been fully processed.

Put those together and you get the proposal. If the opening letters are visually louder, they become artificial anchors: the eye has an obvious target for each jump, the saccade programme has less work to do, and recognition may fire from the emphasised fragment while the remainder is filled in. Advocates usually describe three effects, and it is worth separating them because they would need different evidence.

• Guidance. The eye is given somewhere to land, so fewer fixations are wasted on the gaps and edges between words.
• Early recognition. The emphasised prefix is meant to be sufficient to identify the word, letting the reader move on before the tail is fully read.
• Reduced regression. With clearer anchors, the reader is said to lose their place less often, and therefore to go back less often.

What happened when it was tested

The controlled work that has appeared since the method went viral has been, on the whole, unkind to it. Studies comparing bold-prefix rendering against ordinary text on reading rate and comprehension have largely failed to find a reliable benefit on either. Several found no difference at all. Some found a small cost in speed, which is not surprising if you think about what has been done to the page: heavier type in every word changes word spacing, alters the visual rhythm of the line, and gives the reader a new pattern to habituate to.

There is a second reason the guidance argument is shakier than it sounds. Fluent readers do not need help finding the next word. Saccade targeting in reading is fast, automatic and already well calibrated — the eye lands close to the useful part of a word without being asked, and it does so on unfamiliar text in an unfamiliar font. An intervention that offers a solution to a problem the visual system solved a long time ago is unlikely to buy much, however sensible it sounds written down.

The eye did not need to be told where the next word begins. It has always known.

Why the guidance argument struggles

Easier is not the same as better

Here is where the debunking usually overreaches. A null result on speed and comprehension is a null result on speed and comprehension. It is not a finding that the thing does nothing, and it is certainly not evidence that the millions of people who report that bolded text feels easier are imagining it. Subjective ease is a real and measurable property of an interface, and in reading it has consequences that a fifteen-minute laboratory passage is badly designed to capture.

Reading that feels easier is reading that gets finished. Effort has a compounding effect over a chapter, an evening, a term. If a rendering makes a reader stay with a long document for forty minutes instead of abandoning it at twelve, that reader has read more, and no measurement of words per minute will show it. We would not build our product on that argument, because it is a hypothesis about behaviour over weeks rather than a demonstrated effect. But dismissing it because a short trial came back flat is the same category of mistake the speed-reading industry has always made, run in the opposite direction.

The same caution applies in the other direction, and it matters most where the stakes are highest. Bold prefixing is frequently recommended to dyslexic readers, and here the honest position is that we do not know. Dyslexia is a difficulty with decoding rather than with thinking, and it is not obvious that adding weight to the front of every word helps decoding at all — it may simply change what the page looks like. The decoding literature is clear that these are different problems. Anyone claiming a benefit for that group, in either direction, is going beyond what has been shown.

The pivot letter

The optimal recognition point is a separate idea, and it gets conflated with bold prefixing constantly because both end up marking part of a word. In natural reading, recognition is fastest and most accurate when the eye lands slightly left of a word's centre — roughly a third of the way in. Land elsewhere and identification takes measurably longer. This is one of the sturdier findings in the eye-movement literature, and in ordinary reading the eye targets that position without instruction.

In a word stream the situation inverts. There is no saccade to aim, because the eye never moves; the words move instead. That means the reader cannot land on the right part of the word — the word has to arrive with the right part already under the eye. If words are simply centred, the recognition point drifts left and right with every change in word length, and the reader makes small corrective movements nobody asked for, at every single word, several times a second.

So the fix is geometric rather than typographic. Each word is offset horizontally so that its pivot character sits exactly on the fixed centre rail, and that character is marked in amber so the eye has a stable landmark to hold. The mark is not there to speed up recognition of the word. It is there to stop the eye wandering across a stream of words of wildly different lengths. Bolding a prefix inside a paragraph and pivoting a word inside a stream are answers to two different questions.

Our own pivot, and what we will not say about it

We regard the pivot as a small, real, mechanical win — the same category as removing the return sweep, not the same category as the second channel. It is housekeeping. It removes an irritation that a fixed-point display creates and that ordinary reading does not have. Nothing about our argument for the product depends on it, and if a study tomorrow found that centring words works just as well, we would change the default and lose nothing we care about.

The concession is straightforward, and it is the same one we owe the bold-prefix family. We cannot show you a controlled trial of our pivot either. We have the general finding about where recognition is fastest, and an argument that it should transfer to a display in which the eye is stationary. What we do not have is an experiment comparing our pivoted stream against a centred one, with recall measured on both. Until someone runs it, that is reasoning rather than evidence, and it is filed that way on the science page.

Which is the whole point of writing this article about somebody else's feature. It would be easy to spend a thousand words on the null results in the bold-prefix literature and then quietly present our own unmeasured typographic decision as settled science. The two things are in the same position. One is popular and untested; the other is obscure and untested. If you want to know whether either is helping you specifically, the recall test at the end of a passage in the app is a better instrument than anybody's confidence, including ours.