Human attention is not expendable.
We regard TikTok’s endless feed as a catastrophe for human attention. A product that can occupy years of a person’s finite life deserves to be judged on that scale.
Our argument begins with something every user possesses and can never replace: time. The feed makes continuing effortless. Stopping requires another decision, then another, while the next video is already waiting. TikTok made this format its defining product; Reels and Shorts have carried the same basic pattern across competing platforms. [18][19]
At 90 minutes a day, viewing between the 13th and 25th birthdays adds up to three-quarters of a calendar year. Continue at that rate to 80 and the total reaches 4.19 years—or 9.4% of waking time during that interval, assuming 16 waking hours a day. These are explicit time-use scenarios, not a forecast of anyone’s habits or a reduction in life expectancy.
Human attention is part of human life. Absorbing it at scale is a public concern.
The research gives this argument substance. A 2025 meta-analysis covering 71 studies and 98,299 participants found that greater short-video use was associated with poorer cognition, with attention and inhibitory control showing the strongest associations. Those findings are serious; they are associations, not proof of permanent damage. [3]
Our position is that an endless feed should answer to the life of the person using it. The question is what people would choose for their time if the product helped them stop, reflect and decide. That is the standard we apply throughout this article.
The feed adds up to years.
A session feels small. A lifetime of sessions is not. Use the calculator to see how daily viewing occupies the years between two birthdays.
calendar years
spent scrolling
Age → · Constant daily exposure, no age-related change assumed.
Calendar years = daily minutes × (ending age − starting age) ÷ 1,440. Days = years × 365.25. Waking share = daily minutes ÷ (16 × 60), using a fixed assumption of 16 waking hours per day. The ending birthday is excluded. The three reference intervals retain ages 13–25, 13–40 and 13–80 regardless of the controls above. An 80-year lifespan includes 13 unexposed years at the default start.
This measures time allocation. It assumes use every day at a constant dose, without discounting, mortality adjustment or replacement behavior. It does not measure shortened life expectancy.
Ages 13–25 matter.
Education, first jobs, friendships and habits compete for the same finite hours. A calendar year of accumulated viewing is spread over thousands of ordinary days; it is not a continuous year that could simply be reclaimed. The marginal alternative matters more than the dramatic total.
A trajectory is not a destiny.
Use can rise, taper, stop or move to another platform. An 80-year lifespan is a fixed modeling horizon, not a demographic forecast. The comparison below separates remaining exposure from past behavior.
The use is substantial. The source matters.
The scale of viewing is central to the case against the feed. We use the published figures on their own terms: whole-app time, short-video time and different user populations cannot be mixed.
| Platform | Population | Minutes | Denominator / scope | Period | Source |
|---|---|---|---|---|---|
| TikTok | UK, ages 15–17 | 50 | Internet users | May 2025 | [1] |
| UK, ages 15–17 | 36 | Internet users; whole app | May 2025 | [1] | |
| TikTok | UK, ages 18–34 | 49 | Monthly visitors ÷ days | May 2025 | [1] |
| TikTok | UK, ages 18+ | 28 | Monthly visitors ÷ days | May 2025 | [1] |
| UK, ages 18+ | 20 | Monthly visitors ÷ days; whole app | May 2025 | [1] | |
| TikTok | Monitored children, ages 4–18 | 132 | App users; selected families | 2025 | [2] |
| Monitored children, ages 4–18 | 79 | App users; whole app | 2025 | [2] |
Ofcom’s adult figures divide monthly visitor time by calendar days; they are not daily-active-user averages. Its teen figures use an internet-user base. Whole-app Instagram time includes messaging, photos and Stories, so it cannot be relabeled Reels time. [1]
Qustodio records substantially heavier use in selected, monitored families. Its trend is a repeated sample, not a single cohort followed as it ages. It should not be treated as a representative estimate for all children or all countries. [2]
More time in TikTok. A different denominator.
View chart data
| Year | TikTok | |
|---|---|---|
| 2021 | 91 | 41 |
| 2022 | 107 | 45 |
| 2023 | 112 | 63 |
| 2024 | 120 | 79 |
| 2025 | 132 | 79 |
National panels and monitored-family datasets answer different questions. Neither supplies a representative worldwide joint distribution of TikTok, Reels and Shorts viewing. Usage intensity also varies within age groups, by geography, access and household rules. A mean conceals light users and a much heavier tail.
Instagram is not synonymous with Reels, and YouTube is not synonymous with Shorts. Total-app time is an upper bound for any one feature within the same observed app-time budget. The report uses app-level measurements as context and keeps the feature-specific global model explicitly hypothetical.
The scale makes this a public question.
Multiply an ordinary daily habit across a vast population and the scale changes. In a scenario with 1.5 billion distinct daily viewers watching one combined hour of short-form video, viewing occupies 62.5 million person-years every year.
of viewing every year
1.5 billion unique daily viewers × 60 minutes a day ÷ 1,440 minutes
This figure illustrates the reach of the argument. Even a small disadvantage relative to how people would prefer to spend their time could matter enormously at that scale. The welfare section below makes that value judgment explicit.
The audience is counted once across all platforms, with one combined daily time budget per person. The calculation does not add platform user counts or assign a full hour separately to TikTok, Reels and Shorts. Published reach and app-level minutes do not establish a current worldwide short-video total.
Young people have the most time at stake.
A teenager encountering the feed has decades of potential exposure ahead. The same daily habit occupies more years when it begins earlier.
Fixed daily use and survival to age 80; past viewing is excluded. This compares exposure windows, not measured generational differences or predicted habits.
Those years include education, first jobs, friendships and the development of habits. They deserve particular attention in research and product design. The figures isolate the length of the exposure window; identifying whether young people retain, reduce or increase their use requires repeated measurement over time.
The warning signs deserve action.
The evidence gives ample reason to challenge the product. Serious criticism also needs to distinguish an association, which identifies a pattern, from an experiment, which tests an intervention.
Problematic use measures ask about loss of control and interference with everyday life. Duration measures ask how much someone watches. Mixing the two can make ordinary viewing appear equivalent to functional impairment. The dashboard preserves that distinction and reports uncertainty where it was available in the reviewed source.
Problematic use and routine use tell different stories.
Tang et al. (2026). Mixed short-video platforms and ages. Positive r means greater symptoms with greater use; it does not express a percentage increase. Routine-use anxiety includes zero. Confidence intervals describe pooled correlations, not causal uncertainty. No new pooling is performed. [4]
View the estimates as a table
| Outcome | Exposure | r | Lower CI | Upper CI |
|---|---|---|---|---|
| Depressive symptoms | All measures | 0.24 | 0.15 | 0.33 |
| Anxiety symptoms | All measures | 0.26 | 0.17 | 0.35 |
| Anxiety symptoms | Problematic use | 0.38 | 0.33 | 0.43 |
| Anxiety symptoms | Routine use | -0.02 | -0.18 | 0.14 |
| Loneliness | Problematic use | 0.33 | 0.25 | 0.41 |
| Boredom | Problematic use | 0.42 | 0.29 | 0.53 |
The evidence, by outcome
9 evidence summariesMeta-analysis · Short-form videoAttention & inhibitory control
Attention r = −0.38; inhibition r = −0.41; overall cognition r = −0.34. [3]
Confidence: Association: moderate confidence. Causality: low.
Limit: Heterogeneous exposures include problematic use. These correlations are not percentage reductions in attention. CIs were not extracted from the accessible abstract; no whiskers are invented.
Meta-analysis · Short-form videoDepression & anxiety
Depression r = 0.24 [0.15, 0.33]; anxiety r = 0.26 [0.17, 0.35]. [4]
Confidence: Association: moderate confidence. Causality: low.
Limit: Baseline correlations; routine and problematic use differ. Symptom overlap and shared self-report can inflate associations. The plot above shows reported 95% CIs.
Experiment · TikTokProspective memory
N = 60; TikTok condition performed worse on remembering intended actions. [6]
Confidence: Immediate task effect: limited experimental evidence.
Limit: Small, brief laboratory intervention bundles content, format and switching. Does not establish permanent memory loss. No standardized effect or CI is transcribed here.
Randomized trial · Broader technologySustained attention & well-being
Two-week mobile internet block improved measured attention and reported well-being. [7]
Confidence: Causal for this intervention; indirect for short video.
Limit: Only 119 of 467 committed participants met the preregistered compliance threshold. Volunteers and follow-up limits constrain generalization; no Reels-specific treatment.
Meta-analysis · Broader technologySleep quantity
Bedtime device use: inadequate sleep OR = 2.17 [1.42, 3.32]. [15]
Confidence: Association: moderate confidence. Causality: low.
Limit: Odds ratio, not risk ratio or mortality hazard. Older studies of portable devices cannot identify the effect of TikTok or Reels.
Longitudinal · Broader technologyAdolescent brain development
Habitual checking was associated with changing neural responses to social anticipation. [12]
Confidence: Longitudinal association; causal direction unresolved.
Limit: Functional activation differences do not demonstrate brain damage, arrested maturation, or a loss of intelligence.
Natural experiment · Broader technologyAcademic functioning
Early Facebook rollout increased reported academic impairment attributed to poor mental health. [9]
Confidence: Quasi-experimental, subject to rollout assumptions.
Limit: Early Facebook and college populations; not an estimate of test-score or earnings losses from short video. Reels-specific causal academic effects remain a gap.
Randomized trial · Broader technologySubjective well-being
Four-week Facebook deactivation improved well-being but reduced factual news knowledge. [8]
Confidence: Causal for the studied deactivation and sample.
Limit: Some replacement time went to television. A benefit from deactivation does not make every hour of use harmful.
Systematic review · Short-form videoExecutive function & emotional regulation
17 studies in young adults; mixed methods and heterogeneous findings. [5]
Confidence: Preliminary; insufficient for a pooled causal effect.
Limit: Age focus 18–29; largely observational, self-reported exposure. Publication in 2026 does not mean the underlying search included 2026 studies.
Confidence labels are editorial assessments, not a formal GRADE evaluation. The reviews overlap in their underlying studies; their participant counts must not be added. Correlations, odds ratios, laboratory outcomes and randomized intervention effects are kept separate. A pooled correlation cannot identify what would happen if an adolescent reduced viewing by an hour.
The 2025 synthesis reports the strongest cognitive associations for attention and inhibition; it does not establish irreversible decline. The 2026 young-adult review finds heterogeneous evidence and bidirectional explanations. Reels-specific causal estimates remain particularly limited in the selected literature. [3][5]
What stronger designs add
Randomization can test the effect of an actual intervention; natural experiments can strengthen causal inference under explicit assumptions. Facebook deactivation and early Facebook rollout provide evidence that social products can change well-being. Neither identifies the marginal effect of Reels, infinite scroll or a minute of TikTok on today’s adolescents. [8][9]
Brief experiments on switching and remembering intentions get closer to specific mechanisms. They address immediate performance under a task, not the permanent cognitive capacity of a generation. A stronger test would change one design feature while holding content, social contact and viewing duration constant, then follow meaningful outcomes. [6][11]
Reverse causality is part of the question.
Anxiety, boredom, loneliness, ADHD symptoms or poor sleep can lead people toward a feed. Use may then reinforce difficulties, or provide coping and connection. Family stress, socioeconomic conditions and offline opportunities can affect both use and health. Longitudinal adjustment helps, but does not automatically remove these pathways.
Development also matters. Observational work finds age-dependent and bidirectional associations between social media use and life satisfaction. A longitudinal brain-imaging association is not a diagnosis of injury or proof that a platform interrupts maturation. Effects may differ greatly between people, and adult findings should not be imposed uniformly on ages 13–25. [13][12]
Evidence that pushes against a sweeping conclusion
A large 2019 analysis of broad digital technology use found small associations with adolescent well-being, explaining at most 0.4% of variation. Its exposure predates much current short video, so it neither settles the new question nor permits the older evidence to be ignored. [14]
In Pew’s 2026 report, roughly seven in ten teen users described mostly positive experiences on each of TikTok, Instagram and Snapchat. Most said the platforms neither helped nor hurt their mental health; about three in ten TikTok users reported spending too much time there. Subjective reports can coexist with clinically important harm in a minority. They also make it untenable to count every viewing hour as a loss. [17]
A loop with no natural ending.
An endless feed removes a natural stopping point. Personalized selection and rapid novelty add reasons to continue. This design deserves scrutiny on its own terms: which features make it harder to stop, and what changes when users regain control? A platform-wide correlation cannot assign a separate causal effect to each component.
No stopping cue
An endless feed removes the page ending or episode boundary that could otherwise prompt disengagement. Automatic playback further reduces the action needed to start viewing; it is distinct from automatically advancing to another clip. Longer sessions are a plausible consequence, but the independent lifetime effect of these features has not been identified here. [5]
Select a step. Arrows represent a conceptual sequence, not estimated causal coefficients. Platforms also use safety, negative feedback and relevance signals; engagement is not their only stated objective.
| Product / format | Shared design features | Evidence boundary |
|---|---|---|
| TikTok | Personalized video feed; continuous discovery | Platform-specific experiments exist, but long-term causal evidence is limited. |
| Instagram Reels | Personalized short-video stream inside a broader social app | App-level Instagram research and minutes do not isolate Reels. |
| YouTube Shorts | Short-video stream within a long-form video ecosystem | Whole-YouTube comparisons must not be relabeled Shorts results. |
| Earlier social media | Social rewards, ranking, repeated checking; sometimes infinite feeds | Useful adjacent experiments, with different content and social contexts. |
A claim about the underlying technology is more credible when it survives changes in platform, content and population. The selected evidence supports investigating common mechanisms; it does not yet establish that the same magnitude of harm transfers between products. Educational clips, deliberate search and reciprocal interaction can have different consequences from compulsive passive viewing.
The cost is what the feed displaces.
The standard is whether the feed serves the life a person wants to live. Its cost depends on what their actual alternative would have been, after counting the entertainment and connection it provides.
Observed use expresses a preference at the moment of choice. It need not match a person’s considered preference about tomorrow’s behavior. Experiments with incentives and self-imposed limits support a role for habit and self-control; one economic model attributes 31% of social media use in its sample to self-control problems. That estimate is model-dependent and is not a universal short-video harm fraction. [10]
| Assumed value gap versus the actual alternative | Hypothetical annual opportunity cost |
|---|---|
| 0% | No net loss |
| 5% | 3.13m equivalent years lost |
| 20% | 12.50m equivalent years lost |
| 50% | 31.25m equivalent years lost |
| -10% | 6.25m equivalent years gained |
These fixed examples hold the audience and viewing time constant. A 20% gap means an hour of viewing is assumed to deliver 0.2 fewer equivalent hours of value than its replacement. No reviewed study supplies that global value gap; the table shows how much the judgment matters. Negative gaps allow viewing to be a net benefit. These are opportunity-value units, not deaths, QALYs or years of biological life.
Turning all viewing time into hypothetical wages would obscure the point. Sleep, friendship, study, exercise and leisure can matter without becoming paid work. The case for control over attention does not depend on pricing every hour.
Facebook deactivation also changed how participants valued the service afterward. Yet the replacement included watching television, and news knowledge declined. An honest welfare account must allow that saved scrolling time can be spent on something equally sedentary or less useful, as well as something more valuable. [8]
The case goes beyond a death count.
Sleep disruption, displaced activity and worsening mental health are plausible pathways to health loss. To quantify a short-video mortality effect, however, one must establish the incremental exposure caused by the design, the resulting health change, the appropriate causal risk function and the relevant baseline risk.
activity or distress
over time
Dashed arrows indicate links requiring identification. Depression, sleep and inactivity can overlap; multiplying associated risks would double count shared pathways.
For example, a prospective-cohort meta-analysis reports an all-cause mortality relative risk of 1.06 [95% CI 1.04–1.07] per hour less sleep below seven hours. This is an adult observational sleep association, not a TikTok hazard ratio. It cannot be transported to a teenager’s scrolling hour. The same distinction applies to established risks associated with insufficient activity. [16][25]
Short-video-attributable mortality is not established.
The reviewed research supplies no validated platform-specific death count, life-expectancy decrement or global QALY/DALY total. That gap does not establish an absence of harm.
QALYs
Quality-adjusted life-years combine survival with health-related quality of life. A credible calculation needs an attributable health state, its utility difference and duration. An hour of enjoyable entertainment does not automatically reduce a QALY.
DALYs
Disability-adjusted life-years combine years of life lost with years lived with disability. The necessary attributable incidence, disability weights and mortality estimates are not supplied by a correlation between viewing and symptoms.
There is no validated conversion here from scrolling hours to deaths, biological years lost or health-adjusted burden. A credible health-burden estimate requires measured causal inputs. Welfare-equivalent opportunity time stays in a separate account.
Destruction does not have to look like an explosion.
The argument about TikTok is an argument about reach, repetition and a human lifetime. A technology’s consequences deserve scrutiny even when they arrive a minute at a time.
The atomic-bomb comparison forces a question about what counts as destruction. Hiroshima and Nagasaki caused immediate deaths and lasting radiation-related illness. Those consequences are documented in a way that a global short-video health burden is not. [26]
Calling TikTok the most destructive technology in history is our editorial argument, not an established scientific ranking. The evidence reviewed here cannot support a claim that it is thousands of times more harmful than the atom bomb. Such a multiplier would require a shared measure of harm, including a credible counterfactual for both technologies. Counting viewing years does not supply it.
Established global mortality burdens
Millions of global deaths per year. Tobacco’s bar is a lower-bound marker; lead is approximate. Different reference periods and attribution methods prevent a precise contemporaneous ranking. No scroll-time or welfare-equivalent years appear on this axis. [20][21][22]
| Exposure | Annual global mortality | Basis |
|---|---|---|
| Tobacco | >7m/year; WHO 2026 fact sheet | [20] |
| Alcohol | 2.6m deaths in 2019 | [21] |
| All lead exposure | ≈1.5m/year; WHO 2024 statement | [22] |
| Gambling | No comparable global total used | [23] |
| Short-form video | Not established in the selected evidence | No estimated zero and no plotted bar |
UNEP estimated that ending leaded petrol would prevent more than 1.2 million premature deaths each year. That is a modeled benefit of phaseout, not a current mortality count from petrol. WHO’s all-lead figure above includes other sources of lead and cannot be assigned wholly to gasoline. [24][22]
Gambling offers a useful analogy for compulsive behavior and harms beyond the user, but shared design concepts do not establish equal harm. Tobacco and alcohol have causal mortality evidence that short-video research has not supplied. These benchmarks test the proposed ranking; they do not establish it. [23][20][21]
Our stance rests on a broader judgment: the organization of human attention is consequential enough to demand limits and accountability. Social spillovers, family life, cultural production, political information and network benefits belong in that assessment. A public debate that counts only deaths would miss much of what makes the feed matter.
Human attention deserves a line of defense.
Our position is clear: people should control the feed, and the product should help them stop. Human attention deserves protection from systems organized around keeping people watching.
The scale of exposure makes this a public-interest question. Concerning associations, experimental evidence on related digital interventions and the sheer amount of time at stake justify action alongside better measurement. Waiting for a defensible global death count would impose the wrong threshold on a problem that includes autonomy, opportunity and the shape of everyday life.
What we should demand
- A stopping point people can choose. Finite sessions, visible boundaries and tools that help users honor their own decisions about time.
- Meaningful control over recommendations. Understandable choices about how a feed is selected, with control that persists beyond the next session.
- Independent accountability. Privacy-preserving access to feature-level viewing data, joint platform use and age-specific outcomes, so claims can be checked outside the companies making them.
- Serious tests of harm and benefit. Longer-term, preregistered experiments that change design features and measure sleep, cognition, daily functioning and what people do instead.
- Protection that works for young people. Evaluate age-appropriate defaults and commitment tools against well-being and user goals, with unequal effects and unintended costs reported.
Entertainment, creativity and connection have value. So do the hours that can be spent building a life beyond the feed. The time belongs to the person. The product should answer to them.
Every assumption, in the open.
This article combines FLDI’s editorial position, a critical evidence review and a lifetime accounting calculator. The opening historical superlative is an argument, not a finding of the reviewed studies. This is not a preregistered systematic review or a new clinical study. It has not undergone independent peer review. Sources were checked through 9 November 2026; the figures will not update automatically.
Search and selection
Targeted web searches located primary articles indexed in PubMed, publisher pages, regulator reports, original telemetry reports and WHO/UN publications. Search concepts included short-form video, TikTok, Reels, attention, inhibitory control, memory, anxiety, depression, sleep, longitudinal, randomized, meta-analysis, time spent, mortality and digital addiction. Reference trails connected recent syntheses to experiments and broader social media studies.
Priority was given to systematic syntheses, experiments, longitudinal research and natural experiments, with adolescent and young-adult relevance recorded. Primary usage and public-health sources were used for quantitative benchmarks. Platform documentation informs system description only. This targeted English-language search is not exhaustive; evidence outside indexed and accessible sources may be missed. No PRISMA screening count is claimed.
Extraction, confidence and uncertainty
Bibliographic details, population scope, design, numerical estimates and available intervals were recorded in the source register and dashboard. Full text was used where accessible; the 2025 cognitive meta-analysis is summarized from its accessible abstract, so its CIs are not transcribed. No treatment effect has been inferred from a correlation, no new meta-analysis has been performed, and overlapping review samples are not summed.
Reported CIs remain source-specific. They do not capture reverse causation, selection, unmeasured confounding, transport between ages or countries, measurement error or publication bias. The scenario range is a set of choices, not a sampling interval. Confidence labels are interpretive, not formal risk-of-bias scores.
Model definitions and assumptions
| Output | Formula / definition |
|---|---|
| Lifetime calendar years | daily minutes × exposure years ÷ 1,440 |
| Waking-life share | daily minutes ÷ (16 × 60), using 16 waking hours per day |
| Annual global person-years | unique daily viewers × combined daily minutes ÷ 1,440 |
| Platform allocation | total minutes × platform weight ÷ sum of weights |
| Equivalent opportunity years | attention person-years × assumed net value gap |
| Annual economic value | annual hours × displaced-work share × realized-output share × value/hour |
| Conditional QALYs | viewers × assumed causal excess affected share × utility decrement × health-state duration |
| Remaining exposure with changing use | sum of initial daily minutes × (1 + annual change)^year ÷ 1,440; dose capped at 1,440/day |
The lifetime calculator is the article’s only calculator. Global scale, remaining exposure and welfare illustrations use fixed assumptions. Supplemental allocation, economic and conditional health formulas remain available in the source and data downloads.
One year is 365.25 days or 8,766 hours; one day is 24 hours. All age windows exclude the ending birthday. Lifetime models assume fixed survival to the selected ending age and exclusive attention. Monetary lifetime valuation alone supports discounting; raw time is undiscounted. Relative time values, audience, platform weights and economic assumptions are not fitted from the cited studies.
The default low/central/high assumptions are respectively 1/1.5/2 billion unique daily viewers, 30/60/90 minutes and 5/20/50% net value gaps. These yield 20.83/62.50/125.00 million attention person-years and 1.04/12.50/62.50 million equivalent opportunity years annually. They are illustrative corner cases. Negative gaps are supported to represent benefit.
Institutional context and competing interests
FLDI operates a product lab through Dopaminning, which develops phones intended to support deliberate technology use. That commercial interest is relevant to this report’s subject. None of the cited studies constitutes an evaluation or endorsement of those products. The report makes no claim that an FLDI product prevents or treats a medical condition.
Inspect. Recalculate. Reuse.
Source register, reported estimates and model definitions.
Sources & reading
- Ofcom (2025). Online Nation 2025.
UK communications regulator; Figures 22, 24 and 77.
UK app and website use; whole platforms, not Reels or Shorts separately.
- Qustodio (2026). Lost in the Scroll: Annual Data Report 2025.
Original family device telemetry; social media section, p. 43.
Children aged 4–18 in monitored families; commercially collected, nonrepresentative sample.
- Nguyen et al. (2025). Feeds, feelings, and focus: A systematic review and meta-analysis examining the cognitive and mental health correlates of short-form video use.
Psychological Bulletin 151:1125–1146; doi:10.1037/bul0000498.
71 studies; 98,299 participants; pooled associations across short-form video platforms.
- Tang et al. (2026). Association Between Short-Form Video Use and Mental Health: Systematic Review and Meta-Analysis.
Journal of Medical Internet Research 28:e82503; doi:10.2196/82503.
58 included studies; 53 pooled; 96,676 participants overall. Baseline correlations, not treatment effects.
- Arouch et al. (2026). Short-form video, cognitive and mental health outcomes: a systematic review.
BMJ Public Health 4:e004086; doi:10.1136/bmjph-2025-004086.
17 studies focused on ages 18–29; searched publications through November 2024; narrative synthesis.
- Chiossi et al. (2023). Short-Form Videos Degrade Our Capacity to Retain Intentions: Effect of Context Switching On Prospective Memory.
CHI 2023; author manuscript, arXiv:2302.03714.
Between-group experiment, N=60; short-term prospective memory task, including TikTok.
- Castelo et al. (2025). Blocking mobile internet on smartphones improves sustained attention, mental health, and subjective well-being.
PNAS Nexus 4:pgaf017; doi:10.1093/pnasnexus/pgaf017.
Randomized two-week mobile internet block; 467 committed participants; broad intervention.
- Allcott, Braghieri, Eichmeyer & Gentzkow (2020). The Welfare Effects of Social Media.
American Economic Review 110:629–676; doi:10.1257/aer.20190658.
Randomized Facebook deactivation for four weeks; adult volunteers before the 2018 US midterms.
- Braghieri, Levy & Makarin (2022). Social Media and Mental Health.
American Economic Review 112:3660–3693; doi:10.1257/aer.20211218.
Staggered Facebook introduction at US colleges; quasi-experimental, early social media.
- Allcott, Gentzkow & Song (2022). Digital Addiction.
American Economic Review 112:2424–2463; doi:10.1257/aer.20210867.
Randomized incentives and limits plus a structural economic model; social media generally.
- Tam & Inzlicht (2024). Fast-forward to boredom: How switching behavior on digital media makes people more bored.
Journal of Experimental Psychology: General; publisher research summary; doi:10.1037/xge0001639.
Seven experiments; short-term video switching and boredom, not clinical impairment.
- Maza et al. (2023). Association of Habitual Checking Behaviors on Social Media With Longitudinal Functional Brain Development.
JAMA Pediatrics 177:160–167; doi:10.1001/jamapediatrics.2022.4924.
Adolescent longitudinal neuroimaging; observational social media checking.
- Orben et al. (2022). Windows of developmental sensitivity to social media.
Nature Communications 13:1649; doi:10.1038/s41467-022-29296-3.
UK observational datasets; age-dependent associations and bidirectionality.
- Orben & Przybylski (2019). The association between adolescent well-being and digital technology use.
Nature Human Behaviour 3:173–182; doi:10.1038/s41562-018-0506-1.
Specification curve analysis; broad digital technology use, largely before Reels and Shorts.
- Carter et al. (2016). Association Between Portable Screen-Based Media Device Access or Use and Sleep Outcomes: A Systematic Review and Meta-analysis.
JAMA Pediatrics 170:1202–1208; doi:10.1001/jamapediatrics.2016.2341.
Bedtime portable devices and sleep in children; not specific to short-form video.
- Yin et al. (2017). Relationship of Sleep Duration With All-Cause Mortality and Cardiovascular Events: A Systematic Review and Dose-Response Meta-Analysis of Prospective Cohort Studies.
Journal of the American Heart Association 6:e005947; doi:10.1161/JAHA.117.005947.
Adult prospective sleep cohorts; observational mortality associations, not a scrolling risk estimate.
- Faverio, Park & Gottfried / Pew Research Center (2026). Teens’ Experiences on TikTok, Instagram and Snapchat.
US survey of 1,458 teenagers aged 13–17 and their parents; fieldwork September–October 2025.
Self-reported experiences, reasons for use, benefits and harms; not a clinical causal study.
- TikTok (2026). How TikTok recommends content.
TikTok Help Center; accessed 9 November 2026.
Platform description of recommendation signals; not independent validation of welfare effects.
- Meta (2023). How AI Influences What You See on Facebook and Instagram.
Meta Newsroom; system cards and user controls.
Platform explanation of personalized ranking; not independent health evidence.
- World Health Organization (2026). Tobacco and nicotine.
Fact sheet, 26 June 2026.
More than 7 million annual global deaths; includes second-hand smoke. Annual burden, not a 2026 death count.
- World Health Organization (2024). Alcohol.
Fact sheet, 28 June 2024; 2019 burden.
2.6 million global alcohol-attributable deaths in 2019.
- World Health Organization (2024). Global call to action on lead poisoning prevention.
WHO statement, 18 October 2024.
Approximately 1.5 million annual deaths associated with all lead exposure; not all attributable to petrol.
- World Health Organization (2024). Gambling.
Fact sheet, 2 December 2024.
Health, financial and family harms; limited standardized global harm estimation.
- UN Environment Programme (2021). Era of leaded petrol over, eliminating a major threat to human and planetary health.
UNEP, 30 August 2021.
Phaseout benefit estimated at more than 1.2 million avoided premature deaths annually; a different counterfactual.
- World Health Organization (2026). Physical activity.
Fact sheet, 4 November 2026.
Physical inactivity and health; establishes neither the amount displaced by scrolling nor a platform-specific risk.
- International Committee of the Red Cross & Japanese Red Cross Society (2025). 80 years after the atomic bombings of Hiroshima and Nagasaki: A call to eliminate nuclear weapons.
Joint statement, 5 August 2025.
Immediate and enduring humanitarian consequences of nuclear weapons; no comparison with digital-media harm.