TheAttentionCollapse

The Decline Is Real, and It’s Universal

The skeptic’s first objection to Mark’s data is usually generational. “Of course Gen Z has a short attention span — they grew up on phones.” The implication is that the decline is a youth problem, a cohort effect, something that older professionals don’t need to worry about.

The data doesn’t support that reading. Mark’s forty-seven-second figure is a single population mean: forty participants, logged at their own desks, reported as one average and never broken out by cohort. So the study cannot tell us that every generation fragments at the same rate. What it can tell us is more useful than that. The fragmenting was measured in working adults, not in teenagers, which is the opposite of where the generational story expects to find it. The decline reads as an environmental response. Put any human being inside the digital environment, with its infinite tabs, its notification layer and its node-and-link architecture, and their attention fragments. Not because they’re undisciplined. Because the environment produces exactly this behavior, whether it was designed to or not.

This distinction matters for creative professionals. If declining attention were a generational problem, you could segment your way around it: design differently for younger audiences, keep the old model for older ones. It isn’t generational. It’s a medium problem. The scroll affects everyone. The forty-seven-second ceiling applies to the CMO reviewing your work on her phone just as it applies to the twenty-two-year-old scrolling Instagram on the train.

Nanyang Technological University published a study in 2025 with Research Network and Listen Labs. It was not a survey but a set of AI-moderated voice interviews with five hundred and eighty-three young people aged thirteen to twenty-five in Singapore and Australia, coded thematically. Sixty-eight percent of them said social media had harmed their ability to focus. The sample was screened rather than random, and the paper carries no DOI and no peer review, so that number is not a population estimate. It is something else, and it is worth having. It’s a self-report from an entire generation saying the environment they live in is degrading their capacity for sustained attention. They can feel it, and they’re telling us.

The volume behind this shift is hard to picture. Researchers Martin Hilbert, of the Annenberg School for Communication at the University of Southern California, and Priscila López, of the Open University of Catalonia, calculated that the total information reaching the average person rose from the equivalent of forty newspapers per day in 1986 to one hundred and seventy-four newspapers per day by 2007 (Hilbert and López, “The World’s Technological Capacity to Store, Communicate, and Compute Information,” Science 332, no. 6025 (2011): 60-65). That was before the smartphone became ubiquitous. Before TikTok. Before the algorithmic feed reached its current sophistication. The number today is almost certainly higher. The human brain has not evolved to match this pace. It has adapted instead: switching faster, sampling more, committing less deeply to any single source.

The question for designers is not whether the decline is real. It is. The question is what we do with it, as practitioners who make things that have to function inside this reality rather than as wellness advocates.

Why We Switch: The Science of Self-Interruption

The most unsettling finding in Mark’s research isn’t the forty-seven-second number. It’s what causes the switching.

The popular narrative blames notifications. The phone pings, you look, focus breaks. That narrative has the appeal of naming a villain: the technology companies, the algorithms, the attention-hijacking design patterns. And it’s partially true. Notifications do interrupt. Algorithms are built to capture and retain attention. These are real forces.

But Mark’s data shows something more complicated: people are nearly as likely to interrupt themselves as to be interrupted by something outside. We switch because a thought surfaced, or a curiosity fired, or a habit kicked in, with nothing pinging at all. We remember we need to check something. We wonder what’s happening on the other tab. We feel a pull toward something easier or simply different.

Mark describes her own experience to illustrate the point: while reading an article on AI, a random thought suddenly entered her mind about whether it was safe to eat nonorganic strawberries. She couldn’t get the question out of her head, switched to her browser, searched for strawberries and pesticides, and spent a substantial chunk of time reading about the topic before returning to the article. No notification triggered this. No algorithm pushed it. The interruption came from within.

Jing Jin and Laura Dabbish at Carnegie Mellon studied self-interruption directly. They shadowed workers and found people interrupted themselves for all kinds of reasons: to change their environment, to do something less boring, to seek information, to handle something they suddenly remembered, or to fill time while waiting. Sometimes the task itself cued the interruption, like reading an email that reminded them to check another email. Sometimes it was nothing identifiable at all.

Mark gave this pattern a name: kinetic attention. She defines it as a dynamic state marked by rapid shifts between applications, sites and screens. It isn’t focused attention, which is sustained and deep, and it isn’t mind-wandering, which is unfocused but internal. Kinetic attention is active, external and fast: screen to screen, tab to tab, app to app, with a restlessness that has become the default mode of digital behavior.

The word kinetic comes from physics: dynamic, in motion, marked by vigorous activity. Mark chose it because it captures something the word “distracted” misses. Kinetic attention isn’t a failure. In many ways it’s an adaptive response. Faced with more information than you can possibly process, rapid switching is a reasonable strategy, a way of sampling and triaging. The problem isn’t the switching. It’s the cost.

And the cost is steep.

The Four Costs of the Switch

Johann Hari interviewed Earl Miller, a neuroscientist at MIT who studies attention and working memory (as reported in Hari’s Stolen Focus: Why You Can’t Pay Attention, Crown, 2022). Miller’s research has produced a finding that should reframe how the creative industry thinks about its audience: what we call multitasking is actually rapid switching, and every switch carries a measurable cost. Not one cost. Four.

The switch cost effect. When you shift from one task to another (even a glance at a text message, even five seconds) your brain has to reconfigure. It has to let go of the cognitive context of the first task and load the context of the second. Then, when you switch back, it has to do it again. Each reconfiguration takes time and energy. You’re slower. Your performance drops. And the accumulated cost is far larger than the time spent on the interruption itself. If someone’s screen time shows four hours of phone use, the actual attention cost is much higher, because the switching throughout those four hours degrades attention during the non-phone hours as well.

Mark’s data quantifies the return lag: when people are interrupted or interrupt themselves, they switch to at least two other tasks first, with an average lag of over twenty-five minutes before returning to the interrupted work. Twenty-five minutes. The idea that someone sees your ad, gets distracted, and then comes back to think about it is optimistic to the point of being disconnected from how attention actually works. They’re already three contexts away, and the cognitive residue of your work has been overwritten by two entirely different tasks.

The error effect. This is the one that did not survive checking. The story goes that switching introduces mistakes: the brain backtracks, picks up where it left off, and spends its resources on error-correction and reorientation instead of on comprehension. Gloria Mark ran that experiment. Forty-eight people answered blocks of email in a simulated office while a “supervisor” interrupted them by phone or by instant message, and the interrupted blocks came out faster than the uninterrupted ones, 20.3 and 20.6 minutes against 22.8, with no significant difference in errors, in message length or in politeness (Mark, Gudith and Klocke, “The Cost of Interrupted Work: More Speed and Stress,” CHI 2008). What did rise, all of it significantly, was stress, frustration, time pressure and effort. Her own reading is that people compensate for interruption by working faster and writing less. So the switch has a cost and the cost is not accuracy. It is strain. A chapter built on Mark’s data does not get to skip the experiment of hers that cuts against it.

The creativity drain. New ideas come from the brain forming novel connections between things it has absorbed. That takes undistracted time, mental space for associative thinking to happen below conscious awareness. If the processing is constantly interrupted, the connections don’t form and the ideas don’t emerge. You’re not just less productive; you’re less capable of original thought. That applies to the creative teams producing the work as much as to the audience seeing it. The same environment that fragments the audience’s attention fragments the designer’s.

The changed-memory effect. The study everyone cites here is usually told backwards. Foerde, Knowlton and Poldrack had one group learn a probabilistic task with and without a second task running (“Modulation of Competing Memory Systems by Distraction,” Proceedings of the National Academy of Sciences 103(31), 2006, 11778-11783). Their finding, in their own words: the dual-task condition did not reduce accuracy, but it did reduce how much declarative knowledge about the task was acquired. Distraction didn’t make people learn less. It changed which memory system did the learning, moving it from flexible declarative memory toward habit learning that stays tied to the conditions it was acquired in. For anyone building brand memory that is the worse result, not the milder one: a distracted viewer can absorb your work and still fail to recognize it anywhere but where they first saw it.

The costs are cumulative rather than dramatic, and one of them is smaller than the industry believes. The claim that people who switch between media a lot end up with worse cognitive control has been tested for more than a decade and it has not held. Parry and le Roux pooled 118 assessments and got an association of z = .138; among the performance-based measures, correcting for the small studies that never got published took it to z = .032 and out of significance, and their verdict is that ten years on “we are no closer to understanding ‘cognitive control in media multitaskers’” (Cyberpsychology 15(2), 2021). Wiradhany and Nieuwenstein had reached the same place through two replications and a meta-analysis (Attention, Perception, & Psychophysics, 2017); Wiradhany and colleagues found Bayesian evidence favoring the absence of an effect in 261 people (2020); Alzahabi and Becker found the association running the other way (Journal of Experimental Psychology: Human Perception and Performance, 2013); and Johannes and colleagues, preregistered, found that a visible phone makes people feel distracted while their performance does not move (Journal of Media Psychology, 2019). So: each switch spends time and attention, and at forty-seven seconds a screen the environment asks for a great many of them. What the evidence does not support is an audience left permanently degraded. Our creative work competes for whatever capacity is left after the switching tax has been paid, and that capacity is intact.

What the Brain Sees First

The attention research tells us how long we have. The neuroscience tells us what happens inside that window.

Vision does not hand you a picture in the order you would build one. It hands you almost all of it at once. When Howe compared identifying a whole natural scene against judging one feature of one line, he found no difference between them. Hegdé, reviewing the field, goes further: as soon as you detect that an object is there, you know what it is. Detection and recognition are one event, not two. So the useful division is not between a scene and the things inside it, which was the version in this book’s first draft. It is between everything the eye takes in at a glance and everything that has to be read. Two machines. The second one is slow, and it only runs if the viewer agrees.

The order the design industry has been repeating, mine included, does not survive contact with a direct test. Piers Howe, writing in Attention, Perception, & Psychophysics in 2017, measured the shortest exposure at which people could still identify a natural scene, and compared it against the shortest exposure needed to judge the orientation or the color of a single isolated line. His finding, and his title, is that there was no difference. A whole scene is identified as rapidly as one feature of one line. Which means there is no ladder of features to climb: no stage at which color has arrived and shape has not. That idea appeared in the first draft of this book, and it is gone.

And the sequence is not a clean climb from simple to complex. In one influential study of nine people, Moshe Bar and colleagues found that recognition-related activity appears in orbitofrontal cortex about fifty milliseconds ahead of the temporal regions that do object recognition (PNAS, 2006). In a separate experiment, with different people and different images, that same region responded to coarse, low-detail versions at around one hundred fifteen milliseconds. Their reading is that the rough whole is handed forward as a prediction, and they are careful about it: their analysis cannot show which way the information flows, and they say plainly that recognition does not depend on this pathway. A ladder is the wrong picture for this. It is closer to a fast sketch that the rest of the system then argues with.

For designers, this isn’t an abstraction. It’s a design specification you can build around.

Because the picture arrives whole and coarse, your coarse properties are your signal. A dominant color field, a silhouette, the overall arrangement of masses: these are properties of the whole rather than elements inside it, so they are in the glance by default. They land in peripheral vision while the thumb is still moving. If they are not distinctive, there is nothing to recognize before the decision to keep scrolling has already been made. This is not a claim that one attribute is faster than another. It is a claim about what survives a look that short.

What Howe does give us is a floor. With proper masking, the minimum exposure came out around thirty-five milliseconds, and the separate tasks all landed within a few milliseconds of each other. It is worth knowing how fragile a number like that is. Maguire and Howe had reproduced the classic result a year earlier using sequences built only from natural scenes, and once the displays were masked properly the effect vanished at twenty-seven milliseconds and only held at fifty-three (Attention, Perception, & Psychophysics, 2016). And one piece of language matters more than any of the digits. These are minimum exposure times, not processing times. The picture only has to be present that long. The brain keeps working on it after it is gone.

Thorpe, Fize, and Marlot published in Nature in 1996 showing that complex visual categorization (identifying an animal in a natural scene, for example) happens in under one hundred and fifty milliseconds. The brain performs that analysis in roughly the time it takes to blink.

This is the timeline your creative works on. Not thirty seconds. Not five. Tens of milliseconds of exposure are enough for the scene and for what is in it, which arrive together; what takes longer is what any of it means. The decision to pause or keep scrolling is made by systems running below conscious thought. By the time a viewer “decides” to look at your work, the brain already has a rough read on the whole and is working out the parts. The conscious decision ratifies a judgment that was already made.

The direct implication for design practice: whatever lands in the glance must carry the most signal. The picture arrives. The words wait. Signal before message.

Most creative work inverts this hierarchy. It leads with the message (the headline, the copy, the rational argument) and treats the visual signal as decoration or support. But the visual signal is the only part that does not need the viewer’s consent. In a scroll environment, it may be the only thing that gets processed before the content leaves the viewport. The message arrives second, if it arrives at all.

The Four Myths We Need to Release

Mark’s book identifies four myths about attention that the creative industry has largely absorbed without examination. Releasing them is essential to designing effectively in the scrolling economy.

Myth one: we should always be focused. Mark’s research shows that focused attention ebbs and flows. It follows rhythms. Long stretches of sustained focus without breaks are unnatural for most people, and they come with higher stress and degraded performance. For creative teams that means work demanding sustained focus from the audience is work designed against human nature. Build for the natural rhythm of attention instead.

Myth two: flow is the ideal state. Flow (Csikszentmihalyi’s concept of total, joyful immersion) is real, but it’s rare in digital environments. It occurs for musicians, athletes, coders working on complex problems. It almost never occurs for someone scrolling a feed, checking email, or browsing a website. The default state of digital attention isn’t flow. It’s kinetic. Designing for flow in a feed is designing for a state the audience isn’t in.

Myth three: distractions are primarily caused by notifications and lack of discipline. As we’ve seen, self-interruption is nearly as common as external interruption. The forces driving distraction extend well beyond notifications: the associative structure of the internet, personality traits, social pressures, cultural habits formed by media consumption, and the simple fact that the brain is wired to seek novelty. Blaming the audience for not paying attention to your work is blaming them for being human inside an environment that produces exactly this behavior.

Myth four: mindless digital activity has no value. Mark’s research shows that rote, undemanding activity (scrolling, browsing, playing simple games) helps people replenish cognitive resources. It works as a recovery mechanism. People are happiest using their attention for easy, engaging activity that isn’t challenging or stressful. So the scroll isn’t a waste of the audience’s time. It’s a cognitive state, low-effort and restorative, and your creative work shows up inside it. Design work that demands a jump from that resting state to a high-effort one and you’re asking for a gear change the audience may not make. The work that succeeds in the scroll meets people where they are.

Separating Signal from Noise in Attention Research

Not all attention research is created equal. As this topic has moved from academic journals to mainstream media, a secondary market of attention “facts” has emerged: statistics that get cited, reshared, and built into marketing decks without anyone checking the source.

The most famous example is the claim that humans now have an attention span shorter than a goldfish, eight seconds versus nine. The claim has been attributed to a Microsoft study from 2015. The problem is that the study never compared human attention to goldfish attention, the eight-second number isn’t supported by the methodology described, and the original source appears to be a marketing report rather than a peer-reviewed study. The goldfish comparison is a meme, not a finding.

This matters because the creative industry makes real decisions on this research, and the failure mode is not always a fabricated statistic. Sometimes it is a real study that acquires numbers it never contained. Drèze and Hussherr’s 2003 banner study is cited across the attention business as a source for how long people look at ads. The copy I checked is the authors’ own July 2003 manuscript, self-archived by Drèze; I could not get the published article by any open route. It reports no duration figure anyone could quote as time spent looking at an ad. Forty-nine subjects, eight banners each, and one binary finding: a 0.49 probability that a given banner was fixated. Anyone quoting seconds from it is quoting something that is not in it. And the 0.49 needs a guardrail of its own. Higgins, Leinenger and Rayner’s review of eye movements in advertising puts the share of display ads that get fixated anywhere from 11.7 percent, in Burke and colleagues’ banner-blindness experiments (ACM Transactions on Computer-Human Interaction 12(4), 2005), to close to half (Frontiers in Psychology 5:210, 2014). There is no universal fixation rate, and Drèze’s figure sits at the top of that range rather than in the middle of it. There is enough legitimate, well-designed research available that we don’t need marketing memes or borrowed decimal points.

What survived checking: Gloria Mark’s longitudinal studies, built on direct observation and digital tracking across several studies and years. Thorpe, Fize and Marlot’s finding that the brain categorizes complex natural scenes in under 150 milliseconds, in Nature. Hegdé’s synthesis of the coarse-to-fine sequence in Progress in Neurobiology. Howe’s measurement of the shortest exposure at which each of those judgments is still possible. In every case I can tell you what was measured and what the number is a number of.

What is less well-supported: precise claims about “the average attention span” as a single number, comparisons between human and animal attention, claims about exact percentage declines since specific years, and most statistics that circulate on social media without a primary source. There is a good reason to distrust the self-reported ones in particular. Brasel and Gips (Cyberpsychology, Behavior, and Social Networking, 2011) coded 42 people frame by frame at 30 frames per second while they used a laptop with a television on, and counted 120 shifts of gaze in 27.5 minutes. Asked afterward, the participants estimated 14.8. They reported about twelve percent of their own behavior. Nobody knows how much they switch, which is why asking them produces a number and not a measurement.

What did not survive: several findings this book leaned on in its first draft. The famous thirteen-millisecond figure was the shortest exposure one experiment happened to test, not the time comprehension takes. A perceptual-asynchrony figure, used here as evidence that color is processed faster than form, could not be verified at the value quoted, and the interpretation behind it has been challenged in print (Nishida and Johnston, Current Biology, 2002). When someone finally ran the direct comparison, they found no difference at all between identifying a whole scene and judging one feature of one line, which takes the whole ladder down rather than reordering it. And a four-step millisecond timetable that ran through twenty-six figures rested on two rungs this book has now withdrawn and two that never had a source. All of it is out of this edition. The principle is still the one I started with: cite what’s rigorous, acknowledge what’s suggestive, discard what’s clickbait. What changed is the test. It is not whether a source is prestigious. It is whether a reader can check it and land where you did. Several of mine failed that, and saying so costs less than the alternative.

What the Collapse Means for Design

Here is what the attention research, taken as a whole, tells us about the environment creative work now inhabits:

The average screen holds attention for under a minute, and the average piece of content inside that screen holds it for a fraction of that. The cause isn’t only external interruption. People interrupt themselves nearly as often, driven by habit, curiosity and the associative structure of the digital environment. Every switch carries a measurable cost in speed, accuracy, creativity and memory. The brain takes in the whole picture at once, coarse before fine, at speeds measured in milliseconds and well before conscious awareness engages; there is no strict hierarchy of features and no stage at which color has arrived and shape has not. And the decision to pause or scroll is made by perceptual systems operating beneath conscious thought.

This is not an abstraction. This is the operating environment.

For a creative director briefing a team, the brief should name the micro-moment, the one-to-two-second window, as the primary design constraint rather than an afterthought. For a designer building a social post or a display ad, the coarse signal, color and silhouette, is the first consideration and not the last. And for a brand strategist evaluating a visual system, the question is what survives the glance: blur it, shrink it, push it to the edge of vision and see what is still recognizable. That test belongs in the evaluation criteria.

And for everyone involved in approving creative work: the presentation room is lying to you. It’s showing you what the work looks like with thirty minutes of focused attention. The feed will give it less than two seconds. If you want to know whether your work will perform, you need to evaluate it at scroll speed, on a phone, in a feed full of competing signals.

Attention is kinetic now: rapid, shifting, ruthlessly selective. The work that survives an environment like that is not the work that demands attention. It is the work that earns it, at the right speed, with the right signal.

The next chapter maps the economic logic that governs this environment: how attention functions as currency, how creative functions as product, and how the feed functions as a marketplace with its own rules of supply, demand, and exchange.

The attention didn’t collapse. It reorganized. And it’s waiting for creative work that understands the new pattern.