There is a number you have heard about this: eight seconds, less than a goldfish. It is not real. No study anywhere produced it. What replaced it in the actual research is better measured, more specific, and considerably harder to shrug off — and one of the most repeated pieces of advice about fixing your focus turns out to be the one thing a proper trial showed does not work.
Let us clear the myth out of the way first, because it poisons everything downstream.
The claim — human attention fell from twelve seconds to eight, while a goldfish manages nine — comes from a Microsoft Canada consumer-insights report in spring 2015. Microsoft did do real work for that report: two thousand Canadians surveyed, a review of more than a hundred EEG studies. But the eight-second figure was not in any of it. It sat in an infographic whose entire source line read: “Source: Statistic Brain.”
Statistic Brain is a commercial statistics aggregator. When the BBC’s Simon Maybin contacted them in 2017, they could not produce any supporting research. A later trace found the figure resting on two sources, one of which was a 2008 web-analytics report about twenty-five people leaving websites.
There is also no study establishing a nine-second goldfish attention span. Goldfish memory research measures retention over days and months.
And the researchers who actually measure human attention say the opposite. Edward Vogel at the University of Chicago: he has been measuring college students for twenty years, and it has been remarkably stable across decades. Michael Posner, whose name is on the founding work in this field, says there is no real evidence attention has changed since it was first reported on in the late 1800s.
The nostalgic version of the claim — you used to sit through whole films — is also weaker than it feels. Average film runtimes have risen, not fallen.
So if the capacity has not shrunk, what has?
Gloria Mark’s lab at UC Irvine has been measuring how long people stay on one screen before switching, using objective background logging rather than asking them. In 2004: about two and a half minutes. By 2012: around seventy-five seconds. In a 2016 study: a mean of 47 seconds, median 40. Independently replicated at 44 and 50 seconds.
This is not a measure of how long you are able to concentrate. It is a measure of how long you actually do, in front of a working computer, in a real workplace. Capacity unchanged; behaviour transformed. That is a far more interesting claim than a shrinking brain, and it has instrumentation behind it.
This is the fact that makes the whole subject honest, and it is the one most articles leave out.
Mark’s summary of her own data: we are just as likely to interrupt ourselves as to be interrupted by a notification.
That matters, because the comfortable story is entirely external — they built it, they did this to you. Half of it, roughly, is you reaching for the tab yourself with nothing prompting you. Which is also the reason that silencing your phone does not fix as much as you would expect.
You will have seen it: after an interruption it takes twenty-three minutes and fifteen seconds to refocus. Sometimes twenty minutes, sometimes twenty-five.
The precise 23:15 version appears in no peer-reviewed publication at all. The word does not appear in the paper it is usually credited to. It traces to interviews — a 2006 magazine interview, a business-press feature — and then circulated independently of any study.
There is a real number, and it means something quite different. Mark, González and Harris shadowed twenty-four information workers for over seven hundred hours. When someone was pulled off a task, the average elapsed time before they came back to it was 25 minutes 26 seconds. But during that gap they deliberately worked on an average of 2.26 other things. It is a measure of how work queues up in a fragmented day, not of cognitive recovery time, and certainly not the price of one glance at a notification.
The finding from the same lab that should be famous is stranger. In a controlled experiment, interrupted participants finished the task faster — about 20.3 and 20.6 minutes against 22.8 for the uninterrupted group — while reporting significantly more stress, frustration, time pressure and effort.
So interruption does not obviously cost you output. It costs you something else. You compress, you hurry, you finish, and you pay in depletion. That is much closer to how a fragmented day actually feels than “you got less done.”
The standard dramatic line is that every switch strengthens a switching circuit, and your brain is literally remodelling itself for distraction. There is no study demonstrating this. Hebbian plasticity — the principle behind “neurons that fire together wire together” — is a synaptic learning rule, not a demonstrated mechanism for distractibility. (That four-word phrase is Carla Shatz’s 1992 paraphrase, not Hebb’s own words, and specialists point out it is technically misleading: what matters is temporal order, not simultaneity.)
The study people lean on here is Ophir, Nass and Wagner’s 2009 finding that heavy media multitaskers filter irrelevant information worse. It is correlational, and it has largely failed to replicate: the change-detection result failed in at least three subsequent studies, the switch-cost result failed in four, and one study found heavy multitaskers switched better. A meta-analysis and a ten-year retrospective both come out unimpressed.
Nor is there causal evidence of structural harm. The study most often waved at — brain activation during internet searching, 2009 — had twenty-four subjects over five days and showed changed activation, which is what learning anything looks like. A 2026 scoping review concludes that observed cognitive decline is often mediated by the displacement of other activities rather than direct technological toxicity.
Mark’s own model is not injury. It is resources: a tank that leaks, because reorienting to a new task costs effort every single time. Nothing is broken. Something is being spent.
None of the above means the environment is innocent. It means the mechanism is motivational rather than neurological.
Variable-ratio reinforcement is the real thing underneath the slot-machine comparison. Reward an animal after an unpredictable number of responses and you get a high, steady response rate that is extraordinarily resistant to extinction — it keeps going long after the reward stops. Ferster and Skinner documented the schedules in 1957. (Small correction to the way this usually gets told: pigeons peck keys, rats press levers. And the technical term is resistance to extinction, not addiction.)
Tristan Harris, who was Google’s design ethicist — not an engineer — put it in 2016 like this: when we pull our phone out of our pocket, we are playing a slot machine to see what notifications we got. The popular “slot machine in your pocket” is a headline paraphrase of that sentence.
Infinite scroll was designed in 2006 by Aza Raskin, then working at a user-interface consultancy called Humanized — not, as is usually said, at Mozilla, which came later. His verbatim words to BBC Panorama in 2018 were that it is as if they are taking behavioural cocaine and sprinkling it all over your interface, and that behind every screen on your phone there are generally literally a thousand engineers who have worked on making the thing maximally addicting. The sentence “I regret inventing infinite scroll,” which gets put in quotation marks everywhere, is the BBC’s own summary — they wrote that he had not set out to addict people and now felt guilty about it. He said the first thing. He did not say the second.
The cleanest designer regret actually on record belongs to Loren Brichter, who built pull-to-refresh. In 2017: “I have two kids now and I regret every minute that I’m not paying attention to them because my smartphone has sucked me in.” And: “Pull-to-refresh is addictive. Twitter is addictive. These are not good things. When I was working on them, it was not something I was mature enough to think about.”
Sandy Parakilas, formerly of Facebook, on the record: there was definitely an awareness that the product was habit-forming and addictive. Facebook, in the same piece, on the record too: products are designed to bring people closer to their friends, and at no stage does wanting something to be addictive factor into that process. Both belong in an honest version of this.
This is the most concrete documentation available, and it comes from the company itself.
YouTube’s engineering blog, August 2012: their discovery features had previously been designed to drive views, which “rewarded videos that were successful at attracting clicks, rather than the videos that actually kept viewers engaged. (Cleavage thumbnails, anyone?)” The stated replacement: how a video contributes to a longer viewing session.
Four years later, three Google researchers published the recommender architecture in a peer-reviewed paper. Their words: the final ranking objective is generally a simple function of expected watch time per impression, because ranking by click-through rate often promotes deceptive videos while watch time better captures engagement. An entire section is headed “Modeling Expected Watch Time.”
It is not a leaked memo. It is a published paper. Time is the thing being optimised, and they wrote it down.
On the auction: it is real, but it is worth being exact about who bids. Advertisers bid; apps do not bid for your attention. Real-time bidding puts an individual ad impression up for auction, which the industry’s own standards body compares to how financial markets operate. The honest formulation: your attention is captured by the app, then auctioned to advertisers — and the more time you spend, the more inventory there is to sell.
Herbert Simon saw the shape of this in 1971, and the quote is exact: “a wealth of information creates a poverty of attention and a need to allocate that attention efficiently among the overabundance of information sources that might consume it.” One correction, since it is always got wrong: Simon did not coin the phrase “attention economy.” In that volume the words appear once, in someone else’s discussion of his paper.
And the line everyone half-remembers is worth getting right, because the accurate version is sharper: “The best minds of my generation are thinking about how to make people click ads. That sucks.” That is Jeff Hammerbacher, an early Facebook data scientist, in Bloomberg Businessweek in April 2011.
Two things I am not going to assert, because they are repeated everywhere and I could not confirm them. That notification timing is deliberately randomised — no document shows it; unpredictability mostly falls out of other people posting at unpredictable times and ranking amplifying it. And any story about a specific meeting where autoplay was approved and somebody got promoted. That is how features probably survive. It is not something anyone has documented.
This is the part that surprised me most, and it is in the same 2016 essay that everyone cites to prove the conspiracy.
Harris, verbatim: “But in other cases, slot machines emerge by accident. For example, there is no malicious corporation behind all of email who consciously chose to make it a slot machine. No one profits when millions check their email and nothing’s there. Neither did Apple and Google’s designers want phones to work like slot machines. It emerged by accident.”
So the strongest available version of the claim is not that someone set out to break your attention. It is that a measurable objective was chosen, features that moved it survived, features that did not were deleted, and your fragmented day was the expected and acceptable by-product. Nobody had to intend it. That is worse, and it is documented.
Here is the part that changed my mind while researching this.
“Turn notifications off — not on silent, off” is the standard prescription. In the best randomised field experiment on the question, the all-notifications-off condition produced no attention benefit at all — the effects on inattention and concentration were small and non-significant — while measurably raising anxiety and fear of missing out.
What worked in the same experiment was batching: notifications delivered three times a day. Inattention fell substantially, concentration rose, phone unlocks dropped from 57 a day to 37, stress fell. Hourly batching did essentially nothing — too frequent to be a batch.
Caveats belong here: self-report measures, twenty-eight outcomes with uncorrected p-values, 31% attrition. But it is the strongest direct test that exists, and it points the opposite way from the advice.
Three at a time. Not zero.
Two weeks without mobile internet. A randomised trial blocked mobile internet while leaving calls and texts working: 467 enrolled, 313 completed. Sustained attention improved with an effect size the authors benchmark as roughly reversing a decade of age-related decline. Mental health improved more than the meta-analytic effect of antidepressants; wellbeing improved substantially; 91% got better on at least one outcome. The mediators are the interesting part: more in-person socialising, more exercise, more time outdoors. Honest caveats — only about a quarter met the compliance criterion, most participants already wanted to cut their phone use, and everyone knew what was being studied.
Nature. The most consistent restoration evidence of anything in this article. Meta-analysed across 42 articles and 49 outcome measures: working memory, cognitive flexibility and attentional inhibition all improved after exposure to natural environments, at low-to-moderate effect sizes, with real exposure beating simulated.
Rote, undemanding activity. Mark’s own recommendation, and the better-sourced version of “be bored”: simple mechanical tasks let limited attentional resources replenish. The popular “boredom makes you creative” claim is thinner than it sounds — the 2014 study found gains mainly in convergent thinking, and the mind-wandering incubation study helped only on problems participants had already seen.
Meta-awareness. Catching the urge to switch before acting on it. Given that half the interruptions are self-generated, this is the only intervention that addresses that half at all. Also on Mark’s list: write unfinished tasks down before sleep, and learn your own daily rhythm instead of fighting it.
School phone bans, for some students. Across 91 English schools, test scores after a ban rose by about 6.4% of a standard deviation overall — but 14.2% for the lowest-achieving quartile, and nothing for the top quartile, and nothing at all where the ban was not enforced. Balance it, though: a 2026 national US study found academic effects consistently close to zero, and a thirty-school English study found no wellbeing difference by phone policy.
The phone face-down on the desk does not drain your working memory. The 2017 “brain drain” study is one of the most shared findings in this whole area. It failed a pre-registered direct replication in 2022. The best-powered meta-analysis — 53 studies, 166 effect sizes, 4,368 people — found an effect of d = −0.02, flatly null. A second meta-analysis found a small memory effect and no significant attention effect.
“Digital detox” as a genre does not hold up. A preregistered meta-analysis of social-media abstinence — 10 studies, 4,674 people, 38 effect sizes — found no significant effect on positive affect, negative affect or life satisfaction, and no dose-response with duration. A 21-study systematic review calls the literature mixed, with positive, null and negative results.
Which is exactly why the specific interventions above are worth more than the general resolution. Abstinence is not the mechanism. What you replace the time with is.
And on the hopeful end: the claim that focus comes roaring back in four or five days is invented. Nobody has measured a four-day threshold or a day-five switch. The nearest real timeline is two weeks for a measurable sustained-attention gain, and a 2026 pilot that found early improvements in sleep, sleepiness, mood and stress after 21 days of abstinence in adolescents — which were not maintained at follow-up.
Your attention span is not eight seconds, and a goldfish’s is not nine. Your capacity has not measurably shrunk in twenty years. Your brain is not being remodelled for distraction, and the research that made people believe it was mostly did not replicate.
What is true is narrower and harder to dismiss. You now last about forty-seven seconds on a screen before switching, down from two and a half minutes. Roughly half those switches are your own. The cost is not lost output — it is stress and a resource that leaks every time you reorient. The environment was genuinely engineered, using a reinforcement schedule documented in 1957, optimised against a metric its own engineers published in a paper. And the single most-repeated fix — notifications off — is the one condition a proper trial found did nothing except make people more anxious.
Batch them three times a day. Go outside. Do something boring with your hands. Notice the urge before you obey it.
Nothing is broken. Something is just being spent faster than it comes back.
Short detours through the parts of being human that nobody explains.
Subscribe on YouTube