Attention Repair: Why Detox Fades and Defaults Work (2026)

The short answer to whether attention damage can be undone is yes, partly, and mostly not by willpower. The best randomised evidence shows that cutting smartphone and social media use produces measurable, real improvements in focus and mood within days ◈ Strong Evidence [1] [2]. The same evidence shows that those improvements are modest in size, that most participants fail to sustain the change, and that use drifts back once the experiment ends [1] [3]. The decisive finding of the past five years is economic rather than psychological. Many users would prefer that the platforms they use did not exist, but no individual can act on that preference alone, because the value of staying comes from everyone else staying too ◈ Strong Evidence [4] [5]. That is the definition of a collective trap, and collective traps are not solved by personal discipline. They are solved by changing defaults, rules and costs for everyone at once. The regulatory turn of 2025 and 2026, from Brussels to Canberra, is a bet on that diagnosis [6] [7].

The scale of the problem being repaired is not in doubt. Meta alone reported advertising revenue of $196.2 billion in 2025, from 3.58 billion people who used at least one of its apps every day ✓ Established [8]. DataReportal counts 5.66 billion social media user identities worldwide, equal to 68.7% of the global population, and estimates that the typical user spends well over two and a half hours a day on social and video platforms combined ✓ Established [9]. Among American adolescents, Gallup measured an average of 4.8 hours a day on social media in 2023, with 51% above four hours [10]. Pew found in late 2025 that four in ten US teenagers describe themselves as online almost constantly, against 24% a decade earlier [11]. Every one of those hours is inventory sold to advertisers. Any repair strategy is therefore competing with a business model whose revenue depends on the damage continuing.

What exactly has been damaged is narrower than the popular account suggests, and the distinction matters for repair. Gloria Mark at the University of California, Irvine, has tracked how long office workers stay on one screen before switching, and found the average fell from about two and a half minutes in 2003 to 75 seconds in 2012 and 47 seconds in recent measurements ✓ Established [12]. That is a measure of behaviour, not of cognitive capacity. Laboratory evidence that the mere presence of a phone drains working memory, once widely cited, has not survived direct replication ⚖ Contested [13] [14]. The OECD found in PISA 2022 that 65% of students across member countries were distracted by their own devices in at least some maths lessons, and that frequent distraction was associated with scores 15 points lower ✓ Established [15]. The damage, in other words, is concentrated in habits, environments and time allocation, which is precisely the territory where repair is possible.

This report examines both routes to repair. The individual route covers deactivation, blocking software, friction apps, notification batching, mindfulness training and the revival of basic phones, and it has the stronger experimental record [3] [16] [17]. The structural route covers school phone bans, minimum-age laws, design regulation and litigation, and it has the larger potential reach but a thinner and more self-reported evidence base ⚖ Contested [18] [19]. The two routes are usually presented as rivals, with one side insisting that parents and users must take responsibility and the other insisting that only regulation can work. The evidence supports neither position cleanly. Individual tools work while they are used, and the main reason they stop being used is that the surrounding environment pulls people back [1]. Structural rules change that environment, but the early data from Australia and English schools show that rules which are easy to evade or which leave the phone in the pocket change less than their supporters claim [19] [18].

The economic framing clarifies what a successful repair would look like. Economists Leonardo Bursztyn, Benjamin Handel, Rafael Jiménez-Durán and Christopher Roth found that college students demanded about $59 to deactivate their own TikTok account for four weeks, yet would pay $28 to have TikTok deactivated for everyone, themselves included ◈ Strong Evidence [5]. On their estimates, 64% of active TikTok users and 48% of Instagram users derive negative net welfare from the platforms once the effects on others are counted [5]. A product that people pay to use and would also pay to abolish is not one that markets can be relied upon to correct. The rest of this report asks which interventions move the whole group at once, what they cost, and what the evidence shows about whether the attention they return to people stays returned.

The strongest causal evidence on attention repair comes from economists who paid people to change their behaviour and then measured what happened. In the largest early study, Hunt Allcott, Luca Braghieri, Sarah Eichmeyer and Matthew Gentzkow recruited 2,743 American Facebook users before the 2018 midterm elections and randomly assigned some to deactivate their accounts for four weeks ✓ Established [2]. Deactivation freed about 60 minutes a day. Subjective well-being rose by 0.09 standard deviations, an effect the authors put at roughly 25 to 40 percent of the effect of standard psychological interventions [2]. News knowledge fell by 0.19 standard deviations and political polarisation by 0.16. After the experiment, the treated group used Facebook about 22% less than before, and 80% said deactivation had been good for them [2]. Most strikingly, the amount participants demanded to deactivate again fell by up to 14%, suggesting that some of the value users place on the platform is habit rather than benefit [2].

A follow-up by Allcott, Gentzkow and Lena Song tested whether people know they overuse their phones and whether tools help. About 2,000 Android users were offered either a bonus of $2.50 for every hour they cut from a set of apps, or the ability to set binding limits on their own use ✓ Established [3]. The bonus reduced use by 56 minutes a day, or 39%, and use remained 19 minutes a day lower in the three weeks after the payments ended and 12 minutes lower in the three weeks after that [3]. Self-set limits cut use by 22 minutes a day, or 16%, over twelve weeks, and 78% of participants chose limits that actually bound [3]. The authors’ model attributes 31% of social media use to self-control problems, meaning use that people themselves would not choose on reflection ◈ Strong Evidence [3]. That number is the economic justification for every commitment device, from app timers to lockable phone pouches.

The most direct test of attention itself came in 2025. Noah Castelo, Adrian Ward and colleagues asked 467 American iPhone users to install software that blocked all mobile internet for two weeks while leaving calls and texts working ✓ Established [1]. Average screen time fell from 314 to 161 minutes a day. Sustained attention, measured with a standard laboratory task, improved with an effect size of 0.24, which the authors compare to reversing about ten years of age-related decline [1]. Mental health improved with an effect size of 0.57 and subjective well-being with 0.46, and 91% of participants improved on at least one outcome [1]. Ward summarised the result bluntly: the data suggest that people are not adapted to constant connection to everything all the time [20]. The benefits remained above baseline two weeks after the block ended, but screen time had already climbed back to 265 minutes a day [1].

Lighter interventions also show effects. The one sec app inserts a short breathing pause before a chosen app opens. In a six-week study of 280 users published in PNAS, participants abandoned 36% of their attempts to open the target app, attempts themselves fell by 37%, and total openings dropped by 57% ✓ Established [16]. Notification design matters too. Nick Fitz and colleagues randomised 237 smartphone users in India to receive notifications as usual, in batches three times a day, or not at all; batching improved attention, productivity, mood and feelings of control, while receiving no notifications raised anxiety and fear of missing out [21]. The pattern across these studies is consistent. Friction and batching work because they lower the frequency of the cue rather than asking people to resist it, which is also why they are the features platforms are least inclined to offer by default.

• Full deactivation (four weeks): well-being up 0.09 standard deviations, 60 minutes a day freed, lasting reduction of about 22% in later use [2].
• Paid reduction: 39% less use while paid, 12 minutes a day less six weeks after payments ended [3].
• Mobile internet blocked (two weeks): sustained attention effect size 0.24, mental health 0.57, but only 25.5% fully complied [1].
• Friction pause before opening: 57% fewer app openings over six weeks [16].
• Batched notifications: better attention and mood than either constant or no notifications [21].
• Mindfulness training: attention and working-memory gains of g = 0.19 to 0.39 against active controls across 111 trials [17].

Some of the oldest repair techniques also hold up. A meta-analysis of 111 randomised trials with 9,538 participants by Zainal and Newman found that mindfulness training improved executive attention, sustained attention and working memory, with effect sizes between 0.19 and 0.39 against active control groups ◈ Strong Evidence [17]. Effects were larger in face-to-face formats and in people with symptoms, and absent for processing speed. A systematic review of attention restoration theory found that exposure to natural environments improved working memory and cognitive flexibility, while effects on attentional control were less reliable [22]. The market has responded with hardware. Counterpoint Research estimates that basic feature phones still made up about 15% of handsets sold in 2024, nearly 210 million units, though almost all outside rich countries; North America bought just 1.7 million [23]. The so-called dumbphone revival among young Western buyers is real but small, a niche of intentional products rather than a mass exit [24] [23].

The individual evidence therefore establishes three things. First, the damage is at least partly reversible on a timescale of days, which rules out the most fatalistic accounts of permanent rewiring ◈ Strong Evidence [1] [2]. Second, people want help and use it when offered, as the 78% uptake of binding limits shows [3]. Third, the effect of every tool depends on continued use, and continued use depends on the environment. Only 119 of the 467 people enrolled in the blocking study, or 25.5%, kept the block on for at least ten of fourteen days [1]. The next section examines why the gains so often fade, and why that fading is the central fact for policy.

Single trials can mislead, so the decisive question is what happens when the studies are pooled. The answer depends on what is being measured and how the intervention is designed ⚖ Contested [25] [26]. A 2025 meta-analysis in Scientific Reports by Lemahieu and colleagues combined ten studies with 4,674 participants and found no significant effect of social media abstinence on positive affect, negative affect or life satisfaction, and no evidence that longer breaks did better [25]. A separate 2025 meta-analysis of ten randomised trials with 1,491 participants by May, Malouff and Meynadier found that reducing social media use lowered depressive symptoms with an effect size of 0.25 after adjusting for publication bias [26]. Reduction alone produced 0.33; complete abstinence did not reach significance [26]. Taken together, these results suggest that cutting back helps more reliably than cutting out.

A larger 2025 meta-analysis of 32 randomised trials by Burnell and colleagues, reviewed by Thimm-Kaiser and Keyes in SSM Mental Health, found small but significant improvements in depression, anxiety, stress, self-esteem and self-rated well-being after social media restriction, and no effect on loneliness, happiness or life satisfaction ◈ Strong Evidence [27]. The reviewers noted that 69% of the trials lasted only one or two weeks, which means the literature mostly measures the first days of a change rather than its durability [27]. Christopher Ferguson’s 2024 meta-analysis of 27 experiments found an average effect of 0.088 with a confidence interval that included zero; Jonathan Haidt and Zach Rausch dispute his inclusion choices and his pooling of laboratory exposure studies with real-world restriction trials ⚖ Contested [28]. A 2022 systematic review by Radtke and colleagues reached a similar verdict on digital detox in general: mixed results across health, relationships and performance, with uncertain durability [29].

The election-period experiments funded through Meta’s research partnership add a further caution about size. In a 2024 PNAS study, 19,857 Facebook users and 15,585 Instagram users were paid to deactivate for six weeks before the 2020 US election; effects on polarisation, perceived election legitimacy and turnout were precisely estimated and close to zero ✓ Established [30]. A 2025 working paper from the same project found that Facebook deactivation improved an index of happiness, depression and anxiety by 0.060 standard deviations and Instagram deactivation by 0.041, with the Instagram effect concentrated among women under 25 [31]. These are real effects in samples of tens of thousands. They are also small. A policy promising that quitting social media transforms well-being for the average adult is not supported by them, while a policy targeting young women on image-based platforms has more support [31] [32].

Rebound is the most consistent pattern in the data. In the Castelo blocking study, screen time fell by roughly half during the intervention and recovered to 265 minutes a day, about 84% of its starting level, within two weeks of the block ending ✓ Established [1]. In the Allcott paid-reduction study, the 56-minute daily cut shrank to 19 minutes and then 12 minutes over the following six weeks [3]. After Facebook deactivation, use settled about 22% below its earlier level, a lasting change but far from the 100% reduction achieved during the experiment [2]. The direction of the long-run effect is encouraging. Its size makes clear that most of the change is held in place by the intervention rather than by a new habit. When the intervention is removed, the environment that produced the original habit reasserts itself.

The research on what damage exists is subject to the same correction. Adrian Ward’s 2017 brain drain study, which reported that the mere presence of a smartphone on the desk reduced available cognitive capacity, became the most cited paper in the history of its journal [33]. A pre-registered direct replication by Ruiz Pardo and Minda in 2022 found no effect, and a Bern study of 302 participants found no effect on short-term or prospective memory ⚖ Contested [13] [14]. A 2023 meta-analysis of 22 studies found a small overall effect of minus 0.14, driven by memory tasks, with no significant effect on attention specifically [34]. The implication for repair is reassuring. If the phone in the pocket does not by itself sap capacity, then the problem is what people do with it and how often they are interrupted, both of which can be changed.

The fading problem therefore has a precise shape. Individual tools produce effects of roughly 0.1 to 0.6 standard deviations depending on outcome and intensity [2] [1] [26]. They decay by half or more within weeks [3] [1]. They work best for reduction rather than abstinence [26], for groups with symptoms [17], and for young women on image-based platforms [31]. Most of the literature measures a fortnight [27]. A rational policymaker reading this evidence would not conclude that individual tools are useless. They would conclude that the tools need to be built into the environment so that their effect does not depend on daily acts of will, and that is an economic and regulatory problem rather than a psychological one.

Standard economics treats what people pay, or demand to be paid, as a measure of what a product is worth to them. Erik Brynjolfsson, Avinash Collis and Felix Eggers used large online choice experiments to estimate that the median American would need $48 in 2016 and $38 in 2017 to give up Facebook for a month, and valued social media in general at $322 a year ✓ Established [35]. Allcott and colleagues found a median of $100 to deactivate for four weeks, implying about $31 billion of consumer surplus across US users for that period [2]. By this measure, social media is enormously valuable. The problem is that the measure assumes each person’s choice is independent of everyone else’s. For a network whose whole point is that other people are on it, that assumption fails, and when it fails the willingness to pay stops measuring welfare.

The 2025 American Economic Review paper by Bursztyn, Handel, Jiménez-Durán and Roth tested the assumption directly. Among about 1,000 US college students, the average demanded to deactivate their own TikTok account for four weeks was about $59, and for Instagram $47 ◈ Strong Evidence [5]. But when asked what they would pay to have the platform deactivated for everyone, including themselves, TikTok users offered $28 and Instagram users $10 [5]. Non-users would pay $67 and $39 respectively. Users valued the platforms 33% and 24% more when their peers were on them [5]. Handel described the result in a 2024 interview: on average, people would have to be paid about $50 to deactivate on their own, yet would pay about $30 to deactivate everybody [36]. The authors call this a product market trap, in which active users of a platform prefer it not to exist [4].

The trap explains the rebound pattern in the previous section. A person who blocks mobile internet loses access to group chats, invitations, homework threads and the shared culture of their peers, and the cost of that loss rises with every day the block lasts [5] [1]. The Allcott finding that willingness to accept payment for a second deactivation fell by up to 14% shows that people revise their valuation once they experience life off the platform, but they revise it within an environment where everyone else remains on [2]. The rational individual response is to return. The welfare-improving collective response is for many to leave together, which is precisely what school bans, minimum-age laws and design defaults attempt to engineer. The economic case for structural intervention rests on this finding more than on any claim about brain damage.

The costs extend into working life, where attention is a factor of production. Gloria Mark’s early field research found that 81.9% of interrupted tasks were resumed the same day, but on average only after 23 minutes and 15 seconds, and that 44% of interruptions were self-initiated rather than imposed ✓ Established [37]. Microsoft’s 2025 telemetry found that the most heavily pinged 20% of users of its workplace software were interrupted by a meeting, email or notification roughly every two minutes, about 275 times a day, and 48% of surveyed employees said their work felt chaotic and fragmented [38]. RescueTime’s data on more than 50,000 users found that workers checked communication tools every six minutes on average [39]. None of these sources produces a credible single figure for the national cost of distraction, and estimates that do should be treated with caution. What they establish is that the interruption economy is not confined to teenagers and social media. It runs through the working day.

The trap also explains why platforms have not competed on attention-friendly design. Apple introduced Screen Time dashboards and app limits in 2018 [40], and the European Commission found in 2026 that Instagram’s and Facebook’s time-management tools for teenagers can be easily dismissed and do not lead to a meaningful reduction in use ✓ Established [41]. A platform that unilaterally removed infinite scroll or autoplay would lose engagement to rivals without any guarantee that users would reward it. The same logic that keeps individuals on the platform keeps platforms in the engagement race. When internal TikTok documents released in a Kentucky lawsuit indicated that a user could become hooked after about 35 minutes, they described a product working as designed rather than malfunctioning [42].

The policy implication is uncomfortable for both sides of the usual debate. To those who insist that attention is a matter of personal responsibility, the Bursztyn result shows that responsible individuals rationally stay on platforms they would vote to abolish ◈ Strong Evidence [5]. To those who argue that the platforms are simply addictive poison, the same data show substantial value for many users and a minority who benefit even after accounting for the trap [5] [35]. The defensible conclusion is that attention repair is a coordination problem. It requires interventions that change what everyone does at once, in a defined setting, at a defined time. The school is the most obvious such setting, and it is where governments moved first.

School phone bans are the purest test of the collective-trap logic, because they remove the device from everyone in a defined setting at the same time. The policy has spread faster than its evidence. Ballotpedia counted 33 US states with K-12 phone bans by March 2026, and 41 with some law on phones in schools ✓ Established [43]. The Netherlands introduced a national classroom agreement for secondary schools from 1 January 2024 [44]. France piloted a pause numérique, in which pupils hand over phones for the whole day, in about 180 collèges from September 2024 and extended it to all public collèges from the 2025 school year [45]. Brazil’s Law 15.100, signed on 13 January 2025, bans personal devices in class and during breaks across basic education [46]. South Korea’s National Assembly passed a classroom ban by 115 votes to 31 in August 2025, in force from March 2026 [47]. Sweden has announced a whole-day ban up to grade 9 from August 2026 [48].

• United States: 33 states with K-12 bans by March 2026; New York bell-to-bell since September 2025 [43] [49].
• Netherlands: national classroom agreement since January 2024; 75% of 317 surveyed secondary schools report better concentration [44] [50].
• France: whole-day pause numérique in all public collèges from 2025, extended to lycées by the July 2026 law [45] [51].
• Brazil: Law 15.100 of January 2025 covers class time and breaks, with health and accessibility exceptions [46].
• South Korea: classroom ban in force from March 2026, without penalties [47].
• Australia (New South Wales): 87% of about 1,000 principals report less distraction after the state ban [52].

The strongest positive evidence comes from economists exploiting the staggered timing of bans. Louis-Philippe Beland and Richard Murphy studied 91 schools in four English cities and found that exam results rose after schools banned phones, with the gains concentrated among the lowest-achieving pupils; they described bans as a low-cost policy to reduce educational inequalities ◈ Strong Evidence [53]. Sara Abrahamsson’s study of 477 Norwegian middle schools between 2010 and 2018 found that bans raised girls’ grade point averages and externally graded maths scores by about 0.22 standard deviations and reduced their visits to health services for psychological symptoms by two to three a year, with larger effects for girls from low-income families [54]. Critics have questioned how the Norwegian study coded what counted as a ban [55]. A 2025 Florida study by David Figlio and Umut Özek found test scores improved significantly in the second year of a statewide ban, partly because unexcused absences fell [56].

The Florida study also recorded a cost that advocates rarely mention. Suspensions rose significantly in the first year of the ban, especially among Black students, before fading after the first year ✓ Established [56]. Enforcement is not free, and its burden falls unevenly. The Florida evidence suggests that a ban initially converts phone use into disciplinary incidents, and that the learning benefits arrive only once the norm has settled. That sequence matters for evaluation, because a ban judged after one year may look like a failure on discipline and a success on nothing, while the same ban judged after two years shows gains in achievement [56]. It also matters for design, because schools that pair a ban with consistent, low-drama enforcement are likely to avoid the worst of the first-year spike.

The SMART Schools study, published in The Lancet Regional Health Europe in February 2025, is the most cited evidence against bans ⚖ Contested [18]. Its authors concluded that there was no evidence that restrictive school policies were associated with lower overall phone and social media use or with better adolescent wellbeing [18]. Haidt, Rausch and McClean responded that most of the restrictive schools simply required phones to stay in bags rather than locked away, that the design was cross-sectional, that academic performance was measured with a single teacher rating, and that 10 to 15% of pupils reported more social media time than total phone time [57]. Both points have force. The study shows that a classroom-only ban with phones in backpacks does not change total use or wellbeing. It does not test a whole-day ban with phones physically removed, which is the model France, Sweden and New York have adopted [45] [48] [49].

Most of the positive evidence on recent national bans is self-reported by the institutions that implemented them. A government-commissioned survey of 317 Dutch secondary schools found that 75% reported better concentration, nearly two-thirds a better social climate and about a third better academic performance ◈ Strong Evidence [50]. In New South Wales, 87% of about 1,000 principals said students were less distracted and 81% said learning had improved [52]. A New York survey of 585 educators reported that 80% saw positive results and 60% less bullying, and Governor Kathy Hochul said the results showed the policy was working [49]. In France, 78% of pilot schools rated the pause numérique positively [45]. None of these are outcome studies. They measure what teachers and principals perceive, and staff who have had to enforce a policy have reasons to believe it worked.

The OECD data suggest why bans help where they are enforced and fail where they are not. PISA 2022 found that 65% of students across OECD countries were distracted by their own devices in at least some maths lessons and 59% by other students’ devices, and that students distracted by others’ devices scored 15 points lower, about three-quarters of a year of learning ✓ Established [15]. In schools that formally banned phones, about 29 to 30% of students still reported using them several times a day [15]. A rule that is not enforced is not a collective intervention. It is an individual one with an official label. The school evidence, read as a whole, supports bans that remove phones for the whole day, are enforced consistently, and are evaluated on attainment over at least two years. It does not support claims that classroom rules alone will reverse adolescent mental health trends.

The structural turn began with access and is moving towards design. Australia’s Online Safety Amendment, in force since 10 December 2025, requires platforms including Facebook, Instagram, TikTok, Snapchat, X and YouTube to take reasonable steps to prevent under-16s from holding accounts, with fines of up to A$49.5 million ✓ Established [7]. By mid-January 2026 platforms reported restricting 4.7 million accounts [7] [58]. The United Kingdom announced in June 2026 an under-16 ban covering the same platforms, excluding messaging services such as WhatsApp, with the aim of bringing it into force in spring 2027 [59]. France’s parliament passed a ban for under-15s on 21 July 2026 by 279 votes to 81, blocking new accounts from 1 September 2026 and existing ones from January 2027 [51]. The European Parliament called in November 2025 for an EU-wide minimum age of 16, by 483 votes to 92, in a non-binding resolution [60].

Account counts are not attention counts. A BMJ study of 408 adolescents aged 12 to 17 in New South Wales, surveyed before the law and three months after, found that more than 85% of under-16s were still using restricted platforms, between 54% and 67% on their own accounts, and that 15 to 19% used fake accounts ◈ Strong Evidence [19]. Daily use among 14 and 15-year-olds fell from 78% to 69% [19]. That is a real reduction of nine percentage points, but it is far from the exit the law intended. The eSafety Commissioner has itself cautioned that accounts are not users, and opened investigations in 2026 into several platforms’ compliance [58]. In the United Kingdom, Ofcom data reported in May 2026 found that 84% of children aged 8 to 12 used major platforms despite their age-13 minimum, which indicates how porous existing age rules already are [61].

Design regulation goes after the mechanism rather than the user. In its February 2026 preliminary findings against TikTok, the European Commission named infinite scroll, autoplay, push notifications and the personalised recommender as features that shift users’ brains into what it called autopilot mode, and proposed that TikTok disable infinite scroll over time and introduce effective screen-time breaks ✓ Established [6]. TikTok called the findings categorically false and entirely meritless [6]. In July 2026 the Commission reached similar preliminary findings against Instagram and Facebook, concluding that their teen time-management tools could be easily dismissed and seeking infinite scroll and autoplay switched off by default [41]. Executive Vice-President Henna Virkkunen said that protecting the physical and mental health of Europeans must be a priority for social media platforms [41]. Neither case has yet produced a non-compliance decision or fine, which under the DSA can reach 6% of global turnover [6].

The European approach is about to widen. The European Parliamentary Research Service reported in October 2026 that the Commission’s Digital Fairness Act proposal, expected in the fourth quarter, will target autoplay, infinite scroll, notifications and rewards for continued use as unfair commercial practices across the consumer economy, not only on the largest platforms ✓ Established [69]. In the United Kingdom, Ofcom’s Protection of Children Codes took effect on 25 July 2025 and require highly effective age assurance and recommender systems that filter harmful content out of children’s feeds, but they target content rather than time spent [70]. In the United States, the Ninth Circuit in September 2025 upheld the parts of California’s SB 976 that require parental consent for algorithmic feeds for minors and private-by-default accounts, while blocking a provision that hid like counts [71]. New York’s SAFE for Kids Act rules, finalised in July 2026, will bar algorithmic feeds and overnight notifications for under-18s without parental consent from 25 January 2027 [72].

• In force: Australia’s under-16 minimum age; UK children’s codes; California SB 976 feed-consent rules (litigation continuing); national school bans in the Netherlands, France, Brazil and South Korea [7] [70] [71].
• Legislated, not yet in force: France’s under-15 ban for existing accounts (January 2027); New York SAFE for Kids Act rules (January 2027) [51] [72].
• Announced: UK under-16 ban, targeted for spring 2027 [59].
• Preliminary findings only: EU DSA cases against TikTok, Instagram and Facebook; no fines yet [6] [41].
• Proposed: EU Digital Fairness Act, expected late 2026; EU minimum age of 16 [69] [60].
• Paused: Minnesota’s warning-label law, effective July 2026 but not enforced pending a NetChoice challenge [73].

There is direct evidence that design changes alter time spent. In the 2020 Facebook and Instagram election research, Andrew Guess and colleagues switched more than 20,000 users per platform to a reverse-chronological feed for three months ✓ Established [74]. Chronological-feed users spent markedly less time on the platforms: on Facebook, their usage was 37% above the average US user against 73% for the algorithmic group, and on Instagram 84% against 107% [75]. Removing algorithmic ranking, in other words, cut engagement substantially. It also increased exposure to political and untrustworthy content and did not change polarisation [74]. That trade-off is a warning that design mandates have side effects. China offers a second warning. Its gaming limits for minors, in force since 2019 and tightened in 2021, were followed by no credible reduction in heavy play in a telemetry study covering more than 7 billion hours [63]. Rules that fight the product rather than reshape it invite evasion.

Asia shows a spectrum of approaches. China’s Cyberspace Administration issued guidelines in November 2024 for a minor mode with default daily limits of 40 minutes for children under eight, one hour for those aged 8 to 15 and two hours for 16 to 18, with no service between 10pm and 6am ✓ Established [76]. South Korea’s science ministry found in its 2025 survey that 43% of adolescents fell into the at-risk group for smartphone overdependence, against 22.7% of the population as a whole [77]. In Japan, the city of Toyoake passed an ordinance by 12 votes to 7 in September 2025 recommending no more than two hours a day of leisure smartphone use, with no penalties; its mayor described it as a guideline rather than a blanket rule [78]. The range, from binding state defaults to voluntary civic norms, gives researchers a natural experiment that has barely begun to be exploited.

Litigation has become the fastest-moving channel of structural repair in the United States, because it converts disputed harm into a price. In March 2026 a Los Angeles jury in K.G.M. v. Meta found Meta and YouTube liable for negligent design and failure to warn, awarding a 20-year-old plaintiff $3 million in compensatory damages and a further $3 million in punitive damages after more than 40 hours of deliberation ✓ Established [68]. Snap and TikTok had settled before trial, and post-trial motions were denied in June 2026, with appeals pending [79]. The same month, a New Mexico jury found that Meta had violated the state’s consumer-protection laws and imposed $375 million in civil penalties, against $2 billion sought by the attorney general [80]. Thousands of similar cases are pending in state and federal courts.

The largest outcome came in August 2026. The multistate suit filed in October 2023 went to trial in federal court on 18 August and settled eight days later, with Meta agreeing to pay $17 billion over ten years to fund youth online safety initiatives and to add child-safety measures to Instagram and Facebook ✓ Established [65] [66]. The settlement resolves claims filed by 47 states, of which 29 had sued in 2023 [66]. Reported terms include daily time caps and limits on notifications during school hours for young users, age assurance and an independent auditor [65]. Virginia’s attorney general, Jay Jones, said Meta had for years intentionally deceived the public about addictive and harmful design features [65]. Measured against Meta’s $201 billion in 2025 revenue, the payment is large but not existential [66] [8].

The science underlying these cases remains contested, and honesty about that is part of the evidence ⚖ Contested [81] [82]. The US National Academies concluded in December 2023 that there was not enough evidence to say that social media causes changes in adolescent health at the population level [81]. Amy Orben and Andrew Przybylski found in 2019 that digital technology use explained about 0.4% of the variation in adolescent well-being, a smaller association than wearing glasses [83]. A 2023 Oxford study of 2 million people aged 15 to 89 in 168 countries found no smoking gun linking internet adoption and well-being [84]. Candice Odgers, reviewing Jonathan Haidt’s The Anxious Generation in Nature, wrote that her own research had produced a mix of no, small and mixed associations [82]. Haidt replied that most experiments find evidence of negative effects and very few show benefits [85].

The causation dispute matters less for attention repair than it appears. The question the courts and regulators are answering is not whether social media causes depression across a population. It is whether specific design features were built to maximise compulsive use, and whether companies knew it ◈ Strong Evidence [6] [42] [64]. That is a narrower claim, and the experimental evidence reviewed earlier supports it: people use these products more than they would choose to on reflection, by an estimated 31% [3], and many would collectively prefer the products gone [5]. Amy Orben has warned that concerns about new technologies, from novels to radios to smartphones, recur through history and that research restarts with each new technology [86]. The difference this time is that the economic evidence of a trap does not depend on proving brain damage.

The two columns are less opposed than they look. Every item on the cautious side is an argument about the size and reliability of effects, or about the design of a particular rule ⚖ Contested [25] [19] [18]. Every item on the structural side is an argument about mechanism and incentive [3] [5]. A policy that accepts both would target the specific design features with the clearest evidence of compulsive effect, enforce rules in settings where coordination is possible, measure outcomes rather than compliance counts, and avoid promising mental health miracles. The settlement terms and the EU’s proposed remedies, which focus on defaults, time caps and notifications rather than on content or on banning platforms outright, come closer to that position than either the panic or the dismissal [65] [41].

Three findings anchor the conclusion. First, the damage is reversible. Two weeks without mobile internet improved sustained attention with an effect size of 0.24 and mental health with 0.57 ◈ Strong Evidence [1], four weeks off Facebook raised well-being and freed an hour a day [2], and the claim that phones permanently drain cognitive capacity has not survived replication [13] [34]. Second, the repair does not last on its own. Screen time recovered to 84% of its starting level within two weeks of a block ending, paid reductions decayed by four-fifths within six weeks, and only 25.5% of people offered a full block kept it [1] [3]. Third, the reason it does not last is economic. Users are caught in a trap in which leaving alone is costly even when leaving together would make most of them better off [5] [4].

Those findings imply a hierarchy of interventions. The most effective are those that change defaults for everyone in a defined setting: whole-day school phone removal with consistent enforcement, notification batching and friction built into operating systems, and design rules that turn off infinite scroll and autoplay unless a user chooses them [56] [16] [21] [41]. Next come access rules such as minimum ages, which coordinate exit but leak heavily without robust age assurance [19]. Individual tools such as app timers, blockers and mindfulness training work for those who use them and should be supported, but they cannot carry the burden alone [3] [17]. Litigation changes the price of harmful design and, as the Meta settlement shows, can impose design changes that regulation has not yet achieved [65].

The measurement problem is the largest gap in the evidence. Most evaluations of school bans are surveys of the staff who implemented them [50] [52] [49]. Australia reports accounts removed rather than minutes of attention regained [7]. The EU’s design cases have not yet produced a decision, let alone an evaluation [6] [41]. Most individual-level trials last two weeks [27]. If governments want to know whether structural repair works, they need to measure the outcomes that matter: sustained attention, sleep, attainment, reading and time use, at the population level, over years rather than weeks ◈ Strong Evidence [15] [87]. The decline of reading for pleasure is one such indicator. In the United Kingdom, 32.7% of 8 to 18-year-olds said they enjoyed reading in 2025, the lowest in 20 years, and in the United States daily reading for pleasure fell from 28% of adults in 2004 to 16% in 2023 [87] [88].

• Whole-day phone removal in schools, enforced consistently and judged on attainment over at least two years [56] [54].
• Engagement features off by default, including infinite scroll, autoplay and night-time notifications, as the EU and New York now seek [41] [72].
• Friction and batching at operating-system level, so that the tools that work in trials do not depend on daily willpower [16] [21].
• Age rules backed by real verification, with independent behavioural evaluation rather than removal counts [19].
• Population-level measurement of attention, reading and sleep, published on a regular schedule [15] [88].

The economics also clarify who should pay. A product whose revenue rose 22% in 2025 to $196.2 billion in advertising, and whose own time-management tools regulators found easy to dismiss, can absorb the cost of attention-friendly defaults [8] [41]. The $17 billion settlement shows the scale of liability companies will now accept to avoid juries [65]. Shifting the cost of repair from users, who pay in time, and from schools, which pay in enforcement, to the platforms that profit from engagement is not a punitive idea. It is the standard economic remedy for an externality, applied to a market that the evidence shows is a trap ◈ Strong Evidence [5] [4].

Can the damage be undone, then. The evidence says that much of it can, and quickly, for anyone whose environment changes. It also says that it will not be undone at scale by asking individuals to swim against a current designed to carry them back. The repair that lasts is the repair that changes the current. The first serious attempts to do that, in European design cases, Australian and French age laws, school bans from Seoul to São Paulo and a $17 billion settlement in California, are now under way [6] [51] [47] [46] [65]. Whether they work will depend less on the rules themselves than on whether anyone measures the attention they return.