Gloria Mark's workplace logs put average screen attention at about 47 seconds, down from 2.5 minutes in 2004. Notification systems, algorithmic feeds, and always-on connectivity are the tax. The brain did not get a matching upgrade.
Tap to enlargeThe measured decline
Dr. Gloria Mark, a professor of informatics at UC Irvine and author of Attention Span, has conducted the most rigorous longitudinal research on digital attention. In a 2023 APA interview she restated the series: 2.5 minutes in 2004, 75 seconds around 2012, about 47 seconds in the last several years. The median of those later observations is 40 seconds. The data comes from shadowing and then computer logging in real workplaces, not a lab goldfish test.
Reviews.org's 2025 survey of 1,000 US adults put daily phone checks at 205, up from 144 in their 2023 wave. That is self-report against screen-time dashboards, not a carrier telemetry census. The same survey found 76% check within five minutes of a notification. Older vendor surveys often cited ~46-80 pushes per day. Those figures are a range, not a census.
Gloria Mark's CHI work is often summarized as 23 minutes and 15 seconds to return to the interrupted task. That is recovery of the original thread, not 23 extra minutes of calendar time per ping. Notifications overlap. Multiplying 46 alerts by 23 minutes and calling it a lost workday is bad arithmetic. The real cost is a workday that never stays on one thread.
The structural incompatibility
The information environment is structurally incompatible with the cognitive architecture that evolved over 300,000 years of Homo sapiens existence.
Working memory, the cognitive workspace where information is held temporarily for processing, can maintain approximately 4±1 items simultaneously (a refinement of George Miller's original "7±2" estimate from 1956). This is a hard biological constraint, and it has not increased in the smartphone era. What has increased is the rate at which new items demand entry to working memory.
A pre-digital knowledge worker in 1990 might face a dozen interruptions per day: a phone call, a colleague dropping by, a memo in the in-tray. A digital knowledge worker in 2025 faces hundreds: emails, Slack messages, calendar notifications, text messages, app alerts, news push notifications, social media pings. Each one demands a decision, respond now, defer, or ignore, and each decision consumes a fraction of the limited daily budget for executive function.
The average attention span on a screen has dropped from 2.5 minutes to 47 seconds in twenty years. The drop measures the environment: what constant interruption does to biological cognition over time.
Total screen time now averages 6 hours and 45 minutes per day across all devices (phone, computer, tablet, TV), with smartphone-specific time at approximately 4 hours and 30 minutes. Gen Z reports up to 9 hours. The load is fragmented engagement: switching between apps, feeds, and conversations dozens of times per hour.
The compounding effect is what researchers call "attention residue." When you switch from Task A to Task B, part of your cognitive resources remain allocated to Task A for minutes afterward. If you switch tasks frequently, residue accumulates. The result: you are nominally working on the current task but cognitively processing fragments of several previous ones. Productivity drops. Error rates increase. The subjective experience is fatigue, scattered thinking, and a persistent sense of being behind.
The design problem
The attention crisis is the output of a design philosophy optimized for engagement.
Every major content platform, Meta, TikTok, YouTube, X, employs algorithms that select content for emotional salience, not informational value. Infinite scroll eliminates natural stopping points. Autoplay removes the need for deliberate choice. Variable-ratio reinforcement schedules (the same mechanism that makes slot machines addictive) drive users to check feeds repeatedly because the timing and nature of the "reward" (an interesting post, a like, a reply) is unpredictable.
A meta-analysis of nearly 100,000 participants found that frequent consumption of short-form, low-effort content (TikTok, Reels, Shorts) is correlated with poorer sustained attention and reduced inhibitory control. The brain adapts to the tempo of its inputs. A brain habituated to 15-second videos finds a 15-minute article physiologically uncomfortable. That discomfort is the measured response of a neural system trained for a different cadence.
Platform companies have no incentive to reduce cognitive load. Their revenue, $201 billion for Meta alone in 2024, is directly proportional to time-on-platform. Every design change that would reduce cognitive load (batched notifications, removal of infinite scroll, elimination of autoplay) would also reduce engagement metrics and advertising revenue. The companies profiting from the cognitive load crisis have no structural reason to fix it. The incentive must come from outside the platform.
The organizational cost
The cognitive load crisis carries a direct, quantifiable cost to organizations.
Microsoft's 2023 Work Trend Index, based on aggregated data from Microsoft 365 usage across millions of users, found that employees spend 57% of their work time in communication apps: meetings, email, and chat. Only 43% goes to creation apps like documents, spreadsheets, and design tools. The average meeting-heavy employee now spends 7.5 hours per week in meetings alone, a figure that doubled between 2020 and 2023. In the median knowledge-worker's day, more time goes to talking about work than doing work.
Atlassian has circulated a figure of one interruption every 3.5 minutes for knowledge workers. That is vendor research, not a census. Not every interruption is external; many are self-initiated (checking email, glancing at Slack). The cognitive cost is the same regardless of origin. Each switch drains the limited pool of executive function, and the pool does not refill until the next sleep cycle.
McKinsey Global Institute estimated that improving collaboration and communication through reduced interruptions could raise knowledge-worker productivity by 20 to 25%. Applied to the US knowledge-worker population of roughly 60 million, and using average compensation figures, multiple analyses place the annual cost of unnecessary context switching at $450 billion to $650 billion in lost productivity. That range exceeds the entire GDP of countries like Norway or Austria. It dwarfs the annual revenue of any single technology company responsible for producing the interruption tools.
The organizational response has been, for the most part, to add more tools rather than reduce interruption load. Companies layer project management software on top of chat on top of email on top of document collaboration on top of video conferencing. Each tool adds a new notification channel. Each channel adds a new source of context switches. The result is a productivity stack that generates the very fragmentation it was purchased to solve. Okta's 2023 Business at Work report found that the average large enterprise in its sample deploys 211 SaaS applications. That is Okta's customer sample, not every company. Each app still competes for attention via its own badge, banner, or push notification.
The individual toolkit
Individual responses help one person. Platform incentives and after-hours email norms sit outside that person's control. The evidence for specific interventions is still strong.
Notification pruning. Disabling all non-essential notifications eliminates the primary interrupt vector. Gloria Mark's ~23 minutes is time to get back on task after an interruption, not 23 minutes of calendar saved per silenced badge. Turning notifications off still cuts the interrupt stream.
Time blocking. Cal Newport, the author of Deep Work, advocates scheduling specific blocks of uninterrupted time for cognitively demanding tasks. The evidence supports this: sustained attention improves dramatically when the environment is structured to prevent interruption for 60-90 minute blocks.
Device separation. Using different devices for different functions, a laptop for work, a phone for communication only, a tablet for reading, reduces the cue-triggered switching that occurs when work tools and distraction tools share the same screen. The physical separation creates a cognitive boundary.
Mindfulness training. Controlled trials find that even brief mindfulness meditation practice (8-12 weeks) improves the ability to notice attention wandering and redirect it voluntarily. The mechanism is ordinary: measured cognitive training with documented effects on prefrontal cortex activity.
Digital sabbaticals. A 2022 study published in Cyberpsychology, Behavior, and Social Networking tested the effects of a one-week break from social media on 154 participants aged 18 to 72. Participants who abstained for seven days reported significant improvements in well-being, depression scores, and anxiety scores compared to the control group that continued normal use. The effects were strongest among heavy users, those who had been spending the most time on platforms before the intervention. A one-week break did not produce permanent change, but it showed that the cognitive and emotional costs of constant platform engagement are reversible. The brain retains the capacity for sustained attention; it needs the interruption stream to stop long enough to recover. Newport's concept of a "digital declutter," a 30-day period of stripping back all optional technology followed by selective reintroduction, builds on this finding. The principle: you cannot evaluate whether a tool serves you while you are addicted to it. Distance is a prerequisite for honest assessment.
The structural responses
Individual discipline cannot scale against industrial-grade attention engineering. Structural responses have to hit the platform, the employer, and the law.
Platform regulation. The EU Digital Services Act and proposed "humane technology" legislation aim to restrict the most manipulative design patterns: infinite scroll, autoplay, notification defaults, and algorithmic amplification of emotionally extreme content. The precedent is food labeling: companies can sell what they want, but they must disclose what is in it and are prohibited from the most harmful additives.
Legislative protection of off-hours cognition. Some governments have moved beyond platform regulation to protect workers' attention directly. In 2017, France enacted its droit à la déconnexion (right to disconnect), requiring companies with more than 50 employees to negotiate policies establishing hours during which staff are not obligated to send or respond to work emails. The law treats an employee who is "available" by email from 6 AM to midnight as working an 18-hour day with intermittent breaks. The cognitive load of perpetual availability degrades the quality of their actual working hours. Portugal went further in November 2021, passing legislation that made it illegal for employers to contact workers outside working hours except in emergencies. Violations carry fines. Portugal's law also prohibited employers from monitoring employee productivity through surveillance software at home, a provision that directly addressed the remote-work surveillance surge that followed COVID-19. These are binding legal constraints that treat cognitive load as a workplace safety issue, analogous to restrictions on physical labor hours that emerged during the industrial revolution. The UK, Belgium, and Ontario have introduced or are debating similar legislation.
Employer responsibility. Organizations that require employees to be perpetually reachable via Slack, email, and phone impose a continuous cognitive tax that reduces the quality of the work those employees produce. Companies like Basecamp and Doist (makers of Todoist) have implemented policies restricting after-hours messaging and reducing internal notification volume. The early data suggests these policies improve both output quality and employee retention. Shopify took a different approach in January 2023, deleting 12,000 recurring meetings from employee calendars and instituting "No Meeting Wednesdays" company-wide. CEO Tobi Lütke framed the decision explicitly as cognitive load: reclaiming uninterrupted time for the deep work that generates business value.
AI-mediated filtering. The most scalable response may be AI agents that sit between the user and the information stream, filtering, prioritizing, and batching inputs according to the user's stated goals rather than the platform's engagement metrics. This is the "attentional prosthesis": an AI that protects attention rather than harvesting it. The architecture described in The Text Field is the New Dashboard provides the interface layer for this kind of agent.
Portugal made after-hours contact illegal except in emergencies. Shopify deleted 12,000 recurring meetings in January 2023. The average large enterprise still runs 211 SaaS apps, each with a badge. An agent that batches the 46 daily pings to your stated goals is a filter. An agent that optimizes for time-on-platform is another slot machine. Skip the seventh chat tool.