Infinite scroll erases location and a sense of ending
Aliases: lost in the stream · no-page feed
What it is
Infinite scroll appends the next batch automatically near the bottom, so the list never shows a last screen. It removes two things: location — there is no way to say which stretch of the whole set this is — and a sense of ending — there is no “I have finished” signal at which to stop. That is not whether the footer can be reached, and not whether place is restored after leaving.
Why it happens
Continuous append turns the result space into an unmarked tube. With no ordinal, “that stretch just now” cannot be named, and remainder cannot be estimated. With no bound, the stopping rule changes from “I saw the end” to “I decided I was tired” — a vague internal state, easily postponed by the next batch that just appeared. Feeds lean on that postponement to extend dwell: it is not that people cannot find an exit, it is that an exit never appears in the form of done.
Scroll position is expressed in pixels, and pixels reshuffle after an append, so the same item does not keep a stable coordinate. People then orient by visual landmarks (an image, a headline); once a landmark leaves the viewport, the only remaining action is “further down”. Over time, landmarks in working memory are overwritten by new items, and the cost of looking back spikes. That is not a failed load. A coordinate system was never built.
Studying it
Use stop and locate: give a finite set dressed as infinite, and watch when people stop, whether they can say “about how much I have seen”, and whether they can return to an earlier item from a description. Independent variable: presence of page numbers / progress / an explicit end. Dependent variables: time-to-stop, error in location estimates, scroll distance to re-find the target.
Do not turn the control into “pagination is always better”. The measurement is whether location and ending exist, not what the chunking chrome looks like. If “loaded n of N” is added and infinite scroll remains, location can partly return — which shows that coordinates do the work, not scrolling itself.
Where it stops holding
Content with no natural whole set (live chat, monitoring events, logs still being written) has no ending; fake pagination would invent a bound. In casual, discovery-oriented feeds, losing the ending is sometimes intentional, at the admitted cost that people struggle to stop and to cite. In professional search, comparison, audit, and mail — tasks that need exhaustion or citation — that cost is usually unacceptable. A known total that is still appended without coordinates is throwing away a coordinate that was available.
Applying it
- Ask whether the task needs “finish” or “return to a stretch”. If either is yes, do not use unmarked infinite append; at least provide loaded count, a total, or a shareable locator.
- If automatic append stays, put a place expression beside the list that does not drift with pixels (date breaks, items seen, range labels) so people can say where they are.
- When the set is actually exhausted, show an explicit end, not an endless spinner waiting for another batch.
- How to check: after a stretch of viewing, ask “where in the whole set are you, how much is left, what is the place-name of that earlier item”. An answer of “I just kept scrolling” is a location failure. Watch also for people who report they cannot stop and cannot say why they are still scrolling — that is the missing sense of ending.