Hyperlinks: The Cost That Skilled Readers Turned Into a Signpost

Every blue word in a sentence is a small interruption. Before you can carry on, some part of you has to decide not to click.

That was the standard account of what hyperlinks do to reading, and it had a good theory behind it: each link imposes a decision, decisions consume working memory, and working memory is what comprehension runs on. It is the reason a generation of writing advice told you to put links at the end of paragraphs.

It is also, for competent adult readers, mostly not what happens. This is a story about a finding that got overturned by better instruments, and about who still pays the cost the theory predicted.

The orthodoxy, and it was a serious one

DeStefano and LeFevre reviewed 38 studies published in refereed journals and edited books between 1990 and 2004 and concluded that hypertext features imposed a real cost:

“The increased demands of decision-making and visual processing in hypertext impaired reading performance.”

“Hypertext features that introduced novel structures or required additional processing decreased readers’ comprehension relative to linear texts.”

And the prediction that matters most:

“For readers with low working memory capacity, available mental resources will be quickly exceeded by the additional demands of hypertext.”

One important qualification about the review itself: it is a narrative review, not a meta-analysis, and the authors say so explicitly. They were “unable to conduct statistical meta-analyses” because the studies varied too much in their manipulations and because “meta-analysis requires consistent reporting of effect sizes and variances.”

They also found a condition under which hypertext helped: “Hypertexts that were structured to capitalize on the inherent organization of the domain usually resulted in better comprehension, memory, and navigation.” Structure that matched the subject was an aid. Structure that did not was a tax.

Then someone watched people’s eyes

The 2007 review rested on comprehension scores and navigation logs. It could not see the moment of reading. Eye-tracking can, and a group at Southampton spent several years pointing it at exactly this question.

Experiment one. Fitzsimmons, Weal and Drieghe ran three eye-tracking experiments (n = 30, 36 and 32) on Wikipedia articles, with target words presented as blue hyperlinks or as ordinary black text, crossed with word frequency.

The blanket cost did not appear:

“there were no statistically significant differences in any of the fixation time measures across the conditions except when the target word was grey.”

The cost appeared in one specific place:

“The low frequency, linked words had significantly longer go-past times and total reading times on the target word compared to the other three conditions.”

Their conclusion: “coloured text in itself does not hinder reading, but also that coloured hyperlinks can cause the participant to reread previous content if the word is a low frequent/difficult word.”

So the tax is real but narrow. It is charged when a link sits on a hard word, where the reader is already uncertain, and the link adds a second thing to resolve.

Experiment two, and the reversal. The same group then compared careful reading against skimming (two experiments, n = 32 each). When participants skimmed, a word-frequency effect appeared only for linked words (b = 8.71, SE = 3.77, t = 3.03) and not for unlinked ones (b = -2.12, SE = 3.37, t = -0.63).

Read that the other way round. When skimming, readers were still doing careful processing on the linked words and skating over the rest. Their conclusion was that readers “use hyperlinks as markers to suggest important information and use them to navigate through the text in an efficient and effective way.”

Experiment three confirms why. Jayes and colleagues asked 50 people to rate sentence importance without any eye-tracking, and found that readers “rated sentences without hyperlinks as being of less importance than sentences that did feature hyperlinks.”

A link is not just a door. To an experienced reader it is the writer’s own annotation saying this part matters, and readers use it that way whether or not they ever click.

Except for the readers who have not learned that yet

The reversal has a limit, and it is a developmental one.

Krenca, Taylor and Deacon tested 75 English-speaking children in Grades 3 to 5 (mean age 9.90) on standardised passages from the Gates-MacGinitie Reading Tests, in four conditions: plain, scrolling, hovering hyperlinks that revealed a word definition, and both.

“there was a significant main effect of condition on children’s performance on the reading comprehension questions, with higher scores for the condition with no digital features compared with the conditions with hovering hyperlinks and both scrolling and hovering hyperlinks.”

Their summary of which feature did the damage: “hyperlinks providing word level information interfering and scrolling having no negative impacts.”

Two things stand out. The hyperlinks in that study were helpful ones: they supplied definitions, the sort of scaffolding a well-designed children’s e-book adds on purpose. They still hurt comprehension. And scrolling, the feature that carries most of the screen-versus-paper effect in adults, cost these children nothing.

So the developmental picture inverts the adult one. For adults, scrolling is the expensive part and links are a cue. For nine-year-olds, scrolling is free and links are the interference.

What is actually going on

The 2007 theory was not wrong about the mechanism. A link does impose a decision, and decisions do cost something. What the theory got wrong was assuming the cost stays a cost.

Skilled adult readers appear to have converted the interruption into information. They have learned that a link marks something the writer considered load-bearing, and they now allocate attention accordingly, spending more on linked words, less on the surrounding text, and doing this even while skimming. That is not the absence of an effect. It is the same effect, repurposed.

Children in Grades 3 to 5 have not made that conversion. For them the link is what the theory said it was: an extra thing to resolve, at a moment when resolving the sentence is already taking everything they have.

And in both groups the cost concentrates in the same place, on difficult words. For an adult that means the occasional low-frequency term. For a nine-year-old reading at the edge of their fluency, it means much of the page.

What this does not say

It does not say hyperlinks are free. They are free-ish for practised adults reading ordinary vocabulary, and they measurably are not free on hard words, or for developing readers.

It does not say the 2007 review was wrong. It reviewed 38 studies and reported what they showed, on the instruments available. It also predicted the moderator that has held up best: that the cost falls hardest on readers with the least spare capacity.

It is not a meta-analysis, because there isn’t one. We searched for a hyperlink-specific meta-analysis and found none. There is a 2021 meta-analysis of interactive e-text features reporting a substantial benefit (g = .66 across 26 studies), but its three feature types are questions with feedback, digital glossaries and collaborative tools, not hyperlinks. It is sometimes cited as though it were. It should not be.

The modern evidence is narrow. Four experiments, three of them from one research group, with samples of 30 to 50, on Wikipedia-style expository text with undergraduate participants. That is a thin base for a general claim, and the children’s study, the one that finds the opposite, is a single study of 75.

Nothing here is about clicking. Every study above concerns reading text that contains links. What happens to comprehension when you actually follow one, and what a session of jumping between pages does to what you retain, is a different question that this evidence does not answer.

What to take from it

If you are a practised reader, the blue words are probably not costing you what you have been told. You have most likely learned to read them as emphasis, and the evidence suggests you are extracting real signal from where a writer chose to put them.

Where the cost does show up is exactly where you would expect if you thought about it: on words you were already going to struggle with. If a link sits on an unfamiliar term, expect to reread the sentence, and notice that the interface has just made a hard word harder.

And if you are choosing reading material for a child, the finding to carry is the uncomfortable one. The definition-on-hover feature that looks like helpful scaffolding measurably reduced comprehension in nine- to eleven-year-olds. Scrolling did not. The features that help a fluent reader and the features that help a developing one are not the same features, and the evidence suggests we have been guessing.

The companion question, whether the medium itself changes comprehension, and how much of that turns out to be scrolling, is in reading on a screen versus on paper. For the wider mechanism of using the environment to hold information for you, see the cognitive offloading guide.