47 Seconds: Short-Form Content and Your Attention Span
What the feed is actually training you to do, several thousand times a night
In 2004, researchers shadowing office workers with stopwatches found they stayed on one screen for about two and a half minutes before switching. By the early 2020s, the same research program measured the average at 47 seconds (Mark, 2023). That number gets quoted as "the human attention span," which it isn't — it's the measured attention span of people using screens built to interrupt them. The distinction is the whole story, and it's better news than the headline.
A trait you're born with can't be fixed. A behavior you've practiced can be practiced differently.
What a swipe actually is
Consider what one short-form session consists of, mechanically. A video starts. Within a few seconds you make an evaluation — rewarding enough? — and either stay or swipe. Next video, next evaluation. A half-hour session is a few hundred of these; a heavy evening is thousands.
Each one is a rep. The skill being drilled is rapid disengagement: assess in under five seconds, abandon anything that doesn't pay immediately, next. The feed's recommendation engine handles the other half, making sure the payoffs land on a variable schedule — sometimes the third video is great, sometimes the ninth — which is the schedule that makes lever-pulling durable, in pigeons and everyone else.
Practice effects go where the practice goes. Nobody runs thousands of five-second disengagement reps a night and expects to sit differently with a textbook. The textbook pays on a fifteen-minute schedule. You've been training on a five-second one, and the training says: this isn't paying, swipe. There's no swipe on page 214, so the hand finds the phone, which is the swipe.
What the evidence shows, honestly
Time for this blog's standing discipline. The experimental record on short-form video specifically is young, and you should know its shape:
Heavy short-form use correlates with self-reported attention problems, academic procrastination, and poorer performance on attention tasks across a growing number of studies. Correlation runs both directions, and probably does here: people who struggle to sustain attention plausibly prefer the format that never asks them to. The clean longitudinal experiments — randomize short-form dosage, follow attention performance for a year — largely don't exist yet.
What is established: the interruption cost (attention residue), the presence cost (the brain-drain effect), the sleep pathway, and the 47-second measurement itself. And in the other direction, attention performance improves within weeks when the reps change (two weeks of attention training moved working-memory and reading-comprehension scores in a randomized trial — Mrazek et al., 2013). Whatever the feed has or hasn't done to you, the system demonstrably adapts to new conditions, fast.
So the defensible claim is not "TikTok destroyed your brain." It's narrower and more actionable: you are practicing disengagement at industrial volume, the practice is optimized by professionals, and the only lever you hold is the volume.
Why willpower loses this one on schedule
Resolving to "scroll less" pits a resolution — a fixed intention, made once, at full strength in the morning — against a variable-ratio schedule that renegotiates with you every five seconds, preferentially at night when you're depleted. That's not a fair fight and was never meant to be. The interventions that work in trials don't strengthen the resolution; they change the transaction. A pause before the app opens cut openings 57% in six weeks. Friction beats resolve because friction doesn't get tired.
Re-pricing the feed
HabitApp's version: the short-form apps you choose go behind a shield, and minutes inside them are bought with progress on goals you wrote — the study block, the run, the pages. The session still exists. It's just no longer free, unbounded, or accidental: it has a price you paid on purpose, a length you declared at the door, and a hard floor under it, because when the balance runs out the next session costs another check-in.
Notice what this does to the reps. The disengagement drills still happen — nobody's pretending you'll swipe more slowly — but the sessions become bounded and chosen, the pickups-per-day number gets charted and tends to fall, and the protected blocks on the earning side are where the long-schedule reps quietly come back. Limiting beat quitting in the trials anyway. The feed doesn't need to die. It needs to cost something, and earning internet points spikes habitapp besides.
Forty-seven seconds is a measurement, not a sentence. It was practiced down. The equipment to practice it back up is the same phone.
Sources: Mark, G. (2023), "Attention Span" — screen-attention measurements, 2.5 min (2004) to 47 s. Mrazek et al. (2013), Psychological Science. Leroy (2009), attention residue. Ward et al. (2017), brain drain. Six-week RCT on pre-app pauses, 57%. May, Malouff & Meynadier (2025), limiting vs. abstinence. Short-form correlational literature: emerging studies on short-video use, attention, and procrastination — correlational, direction unresolved, and this article says so.