The loss of focus. Holding an idea for half an hour is now countercultural

Today we're talking about attention, a subject on which everybody has very confident opinions and very few measurements. My thesis, up front, is that holding an idea for half an hour has become rare, and that AI turns up to cash in on that rarity, not to cause it. But are we really more distracted, or do we just look it because our surroundings tug at our sleeve every two minutes? I'm not sure, and while preparing this text I found that the part of the science arguing against me was more solid than I'd expected. I'm writing this with three tabs open and my phone face down, which is the polite way of having it face up. Let's look at it, starting with a fish.

Eight seconds and a goldfish

In 2015 a marketing report from Microsoft Canada put out a figure that became a slogan: human attention had supposedly dropped from twelve seconds in 2000 to eight in 2013, one less than a goldfish. The journalist Simon Maybin, of the BBC, tracked the figure down in 2017. Microsoft hadn't measured it; it was quoting a website called Statistic Brain, and the sources that site claimed to use couldn't find any study behind it. Nor is there any evidence that goldfish have a particularly short attention span. The number was smoke.

We believed it because what was plausible was the trend, not the number. And the trend, in one specific sense, has been measured.

What Gloria Mark measured on screens

Gloria Mark, a psychologist at the University of California, Irvine, has spent about twenty years installing loggers on the computers of real office workers to see how long they stay on one screen before switching. The data, gathered in her book Attention Span (2023), trace a clean curve. In 2004 the average was about two and a half minutes. In 2012, seventy-five seconds. By the end of the 2010s, forty-seven seconds, with a median of forty: half of all attention periods on a screen ended before reaching a good half-minute.

That curve measures how long someone sticks with one computer window while they work, not whether they're capable of concentrating. The press tends to mix the two up, and the difference matters for what comes next.

Mark has another piece of work, less cited, that I find more revealing. In 2008, with Daniela Gudith and Ulrich Klocke, she interrupted volunteers while they were writing emails. The interrupted ones finished sooner and with the same quality. The price showed up in another column: more stress, more frustration, more sense of hurry and effort. We make up for interruption by speeding up, and we pay for it on the inside.

Attention's defender has good arguments

Here's the part that made me rewrite half the article. If the ability to concentrate were collapsing, lab tests ought to notice. And they don't.

In 2024, Andrzejewski, Zeilinger and Pietschnig published in Personality and Individual Differences a meta-analysis of results on the d2 test, a classic test of concentration under pressure. They pulled together around three hundred samples, more than twenty thousand people from thirty-two countries, between 1990 and 2021. Children scored the same across those thirty years. Adults, slightly better.

The cognitive psychologist Daniel Willingham, who has spent years taking apart educational myths, summed up the question in spring 2026 for the American Federation of Teachers' magazine: devices don't seem to degrade the capacity to pay attention. What changes, he suggests, is the decision to pay attention. A student with something more entertaining than the class in their pocket decides sooner that they're bored, because they're comparing without realising it. They can do it; they choose not to.

And AI, according to some data, doesn't even make the picture worse. A field experiment with 7,137 workers at 66 companies (Dillon, Jaffe, Immorlica and Stanton, NBER, 2025) randomly gave part of the workforce access to Microsoft 365 Copilot. Those who used it spent about two hours less a week on email and did less work outside hours. If email was the great source of interruptions, an assistant that dispatches it gives back continuous time.

It's a serious position: the problem would lie in the conditions, not in the brain. Fix the conditions and attention is still there.

My reply, which is a concession with caveats

I grant it the main point and keep mine on the one that matters. Well, I don't know, but what I do know is that «you can concentrate on a ten-minute test if someone asks you to» and «you concentrate for half an hour when nobody asks you to» are different capacities, and the second is the one at stake. The d2 measures the first. The real day, the second.

Willingham's own argument, read slowly, confirms it. If the capacity is intact but we choose to exercise it less and less, the problem doesn't disappear; it moves. A capacity you choose not to exercise for years stops being available when you need it. Sorry, I mean that's what I suspect, not what's been measured: there's no longitudinal study following the same person for twenty years. I'm marking it as a hypothesis.

And the Copilot experiment saves hours of email, but says nothing about what's done with them. Which is where another piece of data comes in, pointing the opposite way.

Two hundred and seventy-five interruptions a day

Microsoft's Work Trend Index special report of June 2025, Breaking Down the Infinite Workday, analysed Microsoft 365 usage signals. The average worker gets an interruption (meeting, email or chat message) every two minutes during their core hours: about 275 a day. It's a corporate report, from a company that sells the solution to the problem it describes, but it squares with Mark's numbers.

More revealing is the work of Aruna Ranganathan and Xingqi Maggie Ye, of the University of California, Berkeley, published in Harvard Business Review in February 2026. For eight months they followed a tech company of around two hundred people where AI use was voluntary. AI didn't reduce the work: it intensified it. Tasks expanded, the boundary between work and home blurred and multitasking became constant. The hours the assistant frees up get filled with more assistant.

One experiment says email gets saved; an ethnography, that the time evaporates into more fragmented tasks. Probably both are right.

AI as the collector, not the cause

Fragmentation wasn't invented by ChatGPT in 2022. The smartphone started it around 2007, and notifications, infinite feeds and the persuasive design of the following decade sped it up. Nicholas Carr framed it in The Shallows (2010) in terms of plasticity: the brain reinforces the circuits it uses. The d2 meta-analysis forces that thesis down a notch: if raw concentration were eroding, it would show up in the tests.

What AI adds is something else. It appears exactly at the moment you were about to let go of the oar. Autocomplete arrives when the sentence was resisting; the chatbot's suggestion fills the gap that doubt was about to leave. It doesn't interrupt: it offers a way out. And the way out has a very precise shape, that of dodging the uncomfortable minutes in which an idea doesn't have a form yet and you have to hold it up without help.

Those uncomfortable minutes are, I suspect, where the attention the d2 doesn't measure gets trained.

Switching off as a countercultural gesture

Cal Newport sold it in Deep Work (2016) as a market advantage: if fewer and fewer people know how to concentrate, whoever does is worth more. Johann Hari, in Stolen Focus (2022), puts it the other way round, as a collective resource extracted by an industry that lives off extracting it. Hari has a journalistic record with stains on it (in 2011 he admitted to presenting in his interviews, as said to him, quotes taken from other sources), so I read him with a magnifying glass; even so, his framework has one virtue. It takes the question out of individual guilt and hands it back to the infrastructure.

My position is in between and somewhat uncomfortable. If Willingham is right and attention is intact but we don't choose it, individual discipline works. If Hari is right and there's an industry optimised so that you don't choose it, individual discipline is rowing against a designed current. Probably it's the first thing inside the second, which is the worst possible combination: you have the capacity and you have the responsibility, but you don't have the environment.

Sitting in front of a blank screen with no assistant, no suggestions and no notifications, and holding an idea for half an hour until it takes shape, is today a gesture you have to decide on every fifteen minutes. And the bill nobody presents in the productivity reports is exactly that one: what the task was training without you knowing.

Definitions

• Sustained attention: the capacity to keep focus on the same task or content for an extended period without shifting it to competing stimuli.
• Screen switch: in Gloria Mark's research, an event logged by software when the person changes the active window or tab on the computer.
• d2 test: standardised test of selective attention and concentration in which specific symbols have to be found quickly among similar distractors.
• Deep work: Cal Newport's term for cognitively demanding work done in distraction-free concentration.

References

• Mark, G. — Attention Span. A Groundbreaking Way to Restore Balance, Happiness and Productivity (Hanover Square Press, 2023).
• Mark, G., Gudith, D. & Klocke, U. — The cost of interrupted work. More speed and stress. Proceedings of CHI 2008, 107-110.
• Andrzejewski, Zeilinger & Pietschnig — Is there a Flynn effect for attention? Cross-temporal meta-analytical evidence for better test performance (1990–2021). Personality and Individual Differences 216, 112417 (2024).
• Willingham, D. — Ask the Cognitive Scientist. Do Today's Kids Have Reduced Attention Spans? American Educator, spring 2026.
• Dillon, E. W., Jaffe, S., Immorlica, N. & Stanton, C. T. — Shifting Work Patterns with Generative AI. NBER Working Paper 33795 (2025, revised November 2025).
• Microsoft WorkLab — Breaking Down the Infinite Workday, Work Trend Index Special Report (June 2025). Corporate report, not peer-reviewed.
• Ranganathan, A. & Ye, X. M. — AI Doesn't Reduce Work—It Intensifies It. Harvard Business Review, 9 February 2026.
• Maybin, S. — Busting the attention span myth. BBC News (2017).
• Carr, N. — The Shallows. What the Internet Is Doing to Our Brains (Norton, 2010).
• Newport, C. — Deep Work. Rules for Focused Success in a Distracted World (Grand Central, 2016).
• Hari, J. — Stolen Focus. Why You Can't Pay Attention (Crown, 2022).

Further reading

• Crawford, M. B. — The World Beyond Your Head. On Becoming an Individual in an Age of Distraction (Farrar, Straus and Giroux, 2015). Treats attention as a practical relationship with a world of things that demand rigour, not as an internal resource. A useful counterweight to Hari's economistic reading.

## You might also like - The erosion of one's own judgement - The cost of thinking less - Total information saturation - Inflated productivity - The design of dependence

## Elsewhere - Scientific American — Are attention spans really shrinking? What the science says - American Educator — Daniel Willingham on children's attention - Harvard Business Review — AI Doesn't Reduce Work—It Intensifies It