Hypertext
Hypertext is text with embedded links to other text, a term Ted Nelson coined in 1965. Literary theorists of the 1980s and 1990s treated it as liberation from the linear authority of the printed page. Comprehension studies found the opposite. Erping Zhu's 1999 work varied only the number of links in otherwise identical passages, and comprehension fell as link count rose, along with readers' confidence in their own understanding. Hypertext's cost is the load of deciding, repeatedly, whether to click.
What it is
The mechanical definition is dull and worth stating anyway, because the argument that followed depends on it. Hypertext is writing in which some words are also doors. A reader can proceed along the sentence or leave it, and the choice is available at every marked word. Ted Nelson coined the term in 1965, well before the web made it the ordinary condition of reading.
What made hypertext interesting for two decades was not the mechanism but the claim built on top of it. Literary theorists in the 1980s and 1990s treated the link as a democratic device. Print, on this account, imposed an order: the author decided what came next and the reader submitted to it. Hypertext handed that decision to the reader, dissolving the author's authority into a network the reader could traverse however they chose. It was a genuinely attractive idea, and it was argued for on aesthetic and political grounds rather than empirical ones.
The empirical work went the other way, and it went the other way consistently. Erping Zhu's 1999 study in the Journal of Educational Multimedia and Hypermedia is the cleanest version because of what it holds constant. Zhu varied only the number of links in otherwise identical passages, so any difference in outcome could not be attributed to different content, different authors or different subject matter. As the link count rose, comprehension scores fell. The second half of the result is the one worth carrying: readers' confidence in their own understanding fell too. They were not simply doing worse, they had some sense that they were doing worse.
DeStefano and LeFevre's 2007 review in Computers in Human Behavior examined thirty-eight studies and reached the general conclusion. Their account of the mechanism is that continually evaluating links imposes a cost, in disorientation and in load, that interferes with the task the reader is actually trying to do. Across the literature they surveyed, hypertext did not consistently beat linear presentation. The promised gain from reader control did not show up in the measurements.
The most instructive finding in this literature is a divergence rather than a deficit. In a comparison of linear pages against hypertext versions of the same lecture material, the linear group comprehended more and the hypertext group found the material more interesting. Neither group was wrong about its own experience. Engagement and understanding came apart, and the version that felt better was the version that worked worse. That gap between how absorbing something feels and how much of it survives is the recurring result here, and it recurs across neighbouring literatures too, which is a reason to take it more seriously than any single study.
Where this argument ended up is screen inferiority, a smaller and much better bounded claim: a modest advantage for paper over screen on comprehension of expository text under time pressure, and no such advantage for narrative text. That is the version of the hypertext argument that survived contact with meta-analysis, and it is more useful for being narrower. The mechanism proposed underneath both is cognitive load: every link is a decision, and a decision about navigation is attention not spent on the argument.