Why do I scroll through articles and remember none of them?
Because the text never stops moving and never ends. Reading that requires scrolling measures worse than reading a fixed page, and paging through a text produces better integrated understanding than scrolling through it. Then there is the missing ending, the point where an argument gets assembled, which never arrives, so nothing gets filed.
Here is the part the papers only describe technically: infinite scroll did not steal your attention, it removed the moment when you could have decided to stop. Turning a page made a gap, one second to ask whether to continue, and that gap was optimised away as a design goal.
You are not weak for failing to stop inside something built to have no stop. The fix cannot be willpower inside the feed: the decision has to happen somewhere else.
Is the cost in the screen or in the scrolling?
In the scrolling. When a text can only be read by scrolling, paper is notably better for comprehension than digital devices, with an effect around a third to half a standard deviation. When scrolling is not required, that difference collapses to almost nothing and stops being reliable.
The newest evidence is a network meta-analysis across computers, tablets, e-readers, smartphones and paper: 56 studies, 79 effect sizes. With scrolling required, paper beat digital devices at Hedges' g of 0.35 to 0.48. Without scrolling, the gap fell to 0.03 to 0.12 and was not reliable, and a tablet read without scrolling was the best digital format in the network. The screen was never the villain, the moving text was. See the meta-analysis.
Direct experiments agree. Sanchez and Wiley found scrolling reduced understanding of complex texts, most of all for readers with less working memory to spare, while paging produced better essay scores. A later study had Norwegian students read on a phone or tablet: paging tended to produce better integrated understanding and more strategic backtracking, the behaviour a feed never invites. Read that study.
Two limits travel with it. That paging advantage was a tendency rather than a clear effect, and scrolling on a tablet felt better than paging. The format that measures worse is the format that feels better.
What do you lose when a page stops being a map?
A retrieval cue. Readers encode where information sits on a page without being told to, and that incidental location memory helps them find a claim again. A fixed page with headings is a map with landmarks. A feed has none, because everything moves.
Rothkopf showed it in 1971: readers pick up a good deal about where content sat on a page without being asked to, and later work on headings as landmarks agrees.
The newest test gave 128 German students in grades 4 to 6 the same text in print and on screen. Recall of where information sat was high in both, but significantly poorer on screen, and it did not depend on how well they had understood the text. Spatial working memory predicted that recall; heavy reading of non-fiction, comics and blogs on screen was associated with weaker recall. That study is correlational on the habits question, so it does not show the feed doing damage.
The counterweight matters. Therriault and Raney found place-on-page precision unrelated to comprehension scores, and Cataldo and Oakhill found good and poor comprehenders remembered locations about equally well while searching differently. The fair claim is narrow: a feed removes a cue a fixed page supplies. A lost tool, not a damaged memory.
Why does an ending carry so much weight?
Because the end of a text is where the reader does the assembly work. Eye-tracking work found that readers spend extra time on the last words of a clause and a sentence, which is when the clause is integrated and the model of the text gets updated. No ending means no integration point.
This is the wrap-up effect, described by Just and Carpenter and observed repeatedly since. It is the best-measured account for alphabetic reading, not a law of brains: eye-tracking work in Chinese reading found the opposite, sentence-final words read faster, modulated by reading scenario, punctuation and word position, so the universal version is wrong.
What survives is modest. In English the end of a sentence does work, and a feed is built to deny you endings. You read four openings, get four half-built models, and none of them closes.
How do I give one piece of reading an ending?
Choose surfaces that end. Anything you intend to finish belongs where the text stays put or arrives in pages, since the cost belongs to scrolling rather than to glass. Open it in a reader, a PDF or print: that restores the map and the ending, and a page break is a built-in chance to ask whether to continue.
Then add one deliberate stop. A click, a page turn, standing up, walking to another room. The point is not virtue, it is making the continue decision happen somewhere other than the consuming.
Then close the loop. Three lines in your own words before you swipe away: what the piece claimed, what it gave you, what it did not answer. That is the integration the missing ending would have done, in under a minute.
One concession: plenty of articles are padded to hold you on the page, and filler complaints run as hot as memory ones. So run a ten-second test: skim the first screen, jump to the last third, look for a number, a claim or a mechanism. If there is one, read the middle. If not, the writing was the problem.
If you want a short daily rep at finishing something with a real end, see Absorb's daily reading and analysis practice on the App Store. A session runs 10 to 12 minutes and scores reading, listening and analysis.
The format was designed, so the person holding the phone can redesign it: one piece a day, read where the text stops moving, with one line at the end.