Better Internet Movement

Evidence Review

A synthesis of the major claims about social media's effects on individuals and society. For each claim, we present the strongest evidence on both sides, rate the overall strength of the evidence, and try to identify what an honest reading of the research actually supports.

Strong evidenceModerate evidenceContestedWeak evidence

Social media is driving the teen mental health crisis

Moderate evidence
+

There is a real and well-documented rise in anxiety, depression, and self-harm among adolescents beginning around 2012. Social media is a plausible contributing cause, but average effect sizes are small and the causal mechanism is debated.

Evidence for

Teen mental health declined sharply starting around 2012, correlating with smartphone adoption. The effect is strongest among girls and heavy users (3+ hours/day), who face double the risk of depression and anxiety. Facebook’s own internal research found Instagram made body image worse for 1 in 3 teen girls who already had body image issues. Multi-week reduction experiments consistently show mental health improvements (d ≈ 0.16–0.20). The U.S. Surgeon General issued a formal advisory in 2023.

Evidence against

Meta-analyses (Ferguson, 2024) find average effect sizes near zero. The crisis is concentrated in the Anglosphere and Nordics, not universal to smartphone adoption, which weakens a global causal claim. Tyler Cowen argues that modern schooling may be a larger driver of youth mental health problems. Some researchers (Orben, Przybylski) argue the effects are no larger than those of wearing glasses or eating potatoes.

How to think about this

The averages debate may miss the point. The relationship is nonlinear — a J-shaped dose-response curve means light use may be harmless while heavy use is significantly harmful. The strongest evidence is for specific vulnerable subgroups (girls, heavy users, those with pre-existing conditions) rather than all teens uniformly. This is analogous to alcohol: small average population effects, but serious harm concentrated in a subset.

Key sources

Social media displaces in-person socializing

Contested
+

In-person socializing among young people has declined dramatically, but the best time-use evidence suggests social media mostly displaces TV watching, not face-to-face contact. The decline in socializing predates smartphones and is driven partly by economic and structural factors.

Evidence for

Teens spending time with friends in person “almost every day” dropped from 44% in 2010 to 32% in 2022. Young adults (15–25) lost 140 hours per year of in-person social interaction between 2003 and 2017. The steepest drop began around 2010, coinciding with smartphone adoption. For teens specifically, the inflection point is sharper than for adults.

Evidence against

A University of Kansas study examining time-use data across the US, UK, and Australia found the decline in face-to-face time began well before social media and has been a steady, uniform trend since the mid-1990s. Social media time mostly displaces other screen time (especially TV), not socializing. The primary drivers of reduced in-person contact appear to be work hours, commuting, and the economics of “third places.” Some studies find social media stimulates in-person contact — more active online users also see friends more.

How to think about this

The most honest reading is that multiple forces are driving the decline simultaneously. Economic and structural factors (car-dependent suburbs, expensive social venues, longer work hours) eroded in-person socializing starting in the 1990s. Smartphones then accelerated an existing trend by offering a low-friction substitute. The displacement is real but smartphones are an accelerant, not the original cause.

Key sources

Social media fragments attention and lowers cognitive ability

Contested
+

IQ scores are declining in multiple developed countries and the steepest drops are among young heavy smartphone users. The causal link to social media specifically is plausible but circumstantial — competing explanations include educational changes and reduced deep reading from all causes.

Evidence for

IQ scores have been declining since the 1990s–2000s in Norway, Denmark, UK, Finland, France, and others. A Norwegian sibling study rules out genetics — the decline is environmental. The hardest-hit demographic (18–22 year olds) is the heaviest smartphone user group. Eye-tracking studies confirm people skim online rather than reading deeply. Gloria Mark’s research shows average focus on a single screen dropped from ~2.5 minutes (2004) to ~47 seconds (2020). Having a phone nearby impairs cognitive performance even when unused (Ward et al., 2017).

Evidence against

The IQ decline began in the mid-1990s in some countries — before smartphones and social media. Other explanations include changes in education (less memorization, less demanding curricula), environmental factors, and the possibility that IQ tests measure skills that are less culturally practiced rather than actual declining intelligence. Lab studies measuring social media’s direct impact on cognitive performance find small effect sizes. People who stop heavy use may recover cognitive skills — we lack good longitudinal data.

How to think about this

The strongest link in the chain is: less deep reading → weaker verbal and reasoning skills → lower IQ scores. But the displacement of deep reading may be driven by digital entertainment broadly (streaming, gaming, short-form video), not social media specifically. The theoretical framework (McLuhan, Postman, Carr) is compelling — the medium shapes cognition — but the specific prediction of permanent cognitive decline goes beyond what the evidence strictly supports.

Key sources

Recommendation algorithms radicalize and polarize

Moderate evidence
+

There is strong investigative evidence that recommendation algorithms amplify inflammatory and extreme content, and credible case studies of real-world radicalization. But the population-level evidence on whether social media increases political polarization overall is mixed.

Evidence for

Max Fisher’s investigative reporting documents cases where Facebook’s algorithm amplified extremist content in Myanmar, the Philippines, and elsewhere, with measurable real-world consequences. Facebook’s own internal research found that its recommendation systems directed users toward increasingly extreme content. Scott Alexander’s “Toxoplasma of Rage” provides a compelling theoretical mechanism: controversial content spreads precisely because it’s controversial, selecting for the most divisive framings of any issue.

Evidence against

Some research suggests political polarization has increased most among demographics that use social media least (older Americans). Filter bubbles may be less hermetic than feared — people encounter more cross-cutting content online than offline. A large-scale Facebook experiment during the 2020 election (Guess et al., 2023) found that removing algorithmic curation reduced exposure to political content but did not measurably change political attitudes or polarization over the study period.

How to think about this

There may be a difference between radicalization of individuals (strong evidence in specific cases) and polarization of populations (weaker evidence at scale). The algorithm doesn’t create extremists from nothing, but it can accelerate the journey for people who are already susceptible. The strongest evidence is for amplification of outrage and emotional content, not necessarily for shifting people’s political positions.

Key sources

The internet centralizes power despite its decentralized architecture

Strong evidence
+

This is one of the best-evidenced claims on the list. Every major layer of the internet — search, social, commerce, cloud infrastructure, app distribution — is dominated by a handful of companies. The pattern is consistent with historical precedent across all information technologies.

Evidence for

Google controls ~90% of search. Meta controls the dominant social platforms. Apple and Google control mobile app distribution. Amazon controls ~40% of US e-commerce and a third of cloud computing. Tim Wu’s historical analysis shows the same open-to-closed cycle occurred with telephone, radio, film, and television. Samo Burja argues this isn’t a corruption of the internet but a feature of the technology itself — network effects and economies of scale make centralization inevitable.

Evidence against

The decentralized web still exists (email, RSS, the open web, Mastodon, AT Protocol). New platforms do emerge and displace incumbents (TikTok disrupting Meta’s dominance in short-form video). Regulatory action (EU’s Digital Markets Act) may force structural changes. The internet is more competitive than telephone or broadcast monopolies were.

How to think about this

The question isn’t whether centralization happened — it clearly did — but whether it’s reversible. The optimistic case is that protocols (like AT Protocol/Bluesky) can re-decentralize at the application layer even if infrastructure remains concentrated. The pessimistic case (Burja) is that decentralization is unstable and will always re-centralize.

Key sources

Social media undermines democratic deliberation

Moderate evidence
+

The theoretical case is strong: democracy requires shared facts and exposure to opposing views, and algorithmic sorting undermines both. The empirical evidence is suggestive but hard to isolate from broader trends in political culture.

Evidence for

Cass Sunstein’s theoretical argument is rigorous: deliberative democracy depends on “unchosen encounters” with differing perspectives, and algorithmic personalization eliminates them. Fukuyama argues the internet is the root cause of political dysfunction, not a symptom. The speed of social media rewards reactive outrage over deliberative reasoning. Misinformation spreads faster than corrections.

Evidence against

Political polarization was increasing before social media (it tracks to the 1990s in the US). Polarization has increased most among older Americans who use social media less. Some evidence suggests people encounter more diverse viewpoints online than in their physical communities. The 2020 Facebook election experiment found no measurable effect on political attitudes from algorithmic changes.

How to think about this

Social media may degrade the quality of political discourse (more outrage, shorter attention spans, less nuance) without necessarily increasing ideological polarization per se. The distinction matters: people can become more angry and tribal without actually moving further apart on policy positions. The harm may be to the tone and quality of democratic life rather than to the distribution of political opinions.

Key sources

The internet enables genuinely new forms of human coordination

Strong evidence
+

Wikipedia, open source software, and new forms of collective action are real and well-documented achievements. The evidence that the internet enabled coordination that was previously impossible is among the strongest on any claim in this review.

Evidence for

Wikipedia is the largest encyclopedia in human history, maintained by volunteers, and consistently rivals traditional encyclopedias in accuracy. Open source software powers the majority of the world’s servers, phones, and infrastructure. The internet enabled coordination at scales and speeds impossible before — from disaster relief to political movements. Benkler’s theoretical framework (commons-based peer production) has strong empirical support. Shirky’s analysis of reduced coordination costs explains why these forms of collaboration emerged.

Evidence against

Many early hopes for internet-enabled coordination have not materialized or have been co-opted. Eghbal’s research shows open source communities face serious sustainability problems — most projects depend on a few overworked maintainers. Platform-mediated coordination (Uber, Airbnb) often recreates old power imbalances in new forms. Wikipedia’s editor base has been declining for years.

How to think about this

The internet’s coordination achievements are real but fragile. They depend on specific social conditions (intrinsic motivation, shared norms, low barriers to entry) that platforms can undermine. The question going forward is whether the coordination benefits survive the centralizing pressures of platform capitalism, or whether they were a temporary product of the internet’s early, more open era.

Key sources

Social isolation is increasing

Strong evidence
+

This is well-documented by federal time-use data, large-scale surveys, and a U.S. Surgeon General’s advisory. About half of American adults report being lonely, and time spent socializing in person has declined sharply, especially among young people. The trend predates the pandemic.

Evidence for

The U.S. Surgeon General declared loneliness an epidemic in 2023, citing that approximately one in two adults reported experiencing loneliness. In-person socializing among 15–25 year olds dropped from 61 minutes/day in 2003 to 39 minutes/day in 2019. The share of high school seniors gathering with friends “almost every day” fell from 44% to 32% between 2010 and 2022. Isolation levels have been rising since 2012. The health impact is severe — loneliness is associated with mortality effects comparable to smoking 15 cigarettes a day.

Evidence against

Some researchers argue loneliness is subjective and hard to measure consistently across time. The definition of “social isolation” varies across studies. Online interaction may partially substitute for in-person contact for some people, meaning isolation metrics that only count face-to-face time may overstate the problem. COVID-19 created a step change that may be reversing.

How to think about this

The trend is real and predates the pandemic — COVID amplified it but didn’t create it. The causal role of social media specifically is less clear than the trend itself. The decline in in-person socializing appears driven by a combination of economic factors (expensive social venues, longer commutes), urban design (car-dependent suburbs), and digital substitution. The strongest evidence is for the trend existing; the attribution to any single cause is weaker.

Key sources

People feel they waste time on social media

Strong evidence
+

Large majorities of users report regretting their social media use during or after sessions. This is one of the most consistently replicated findings in social media research and is supported by both survey data and behavioral studies using real-time tracking.

Evidence for

A Carnegie Mellon study analyzing 34,000 smartphone screenshots found users regretted at least some of their social media use in 60% of sessions and regretted all of it in nearly 40% of sessions. 45% of people regret using social media as much as they did when younger. The content most regretted is algorithmically recommended posts and comments — not content from friends. Pew Research found 42% of parents with teens cite time wasting as a concern. The regret is highest for non-intentional browsing, suggesting the issue is not social media per se but the algorithmic feed pulling people into unplanned use.

Evidence against

Stated regret may not match revealed preference — people continue using platforms they claim to regret, which could indicate the regret is overstated or that the benefits (connection, entertainment, information) outweigh the costs in practice. Some researchers argue “regret” conflates genuine harm with the kind of mild guilt people feel about any leisure activity (TV, snacking). People also report regretting time spent on many things that aren’t harmful.

How to think about this

The consistency of the finding across different methodologies (surveys, real-time tracking, behavioral analysis) makes it hard to dismiss. The most interesting nuance is that regret concentrates on algorithmic recommendations, not on intentional use like messaging friends. This suggests the problem isn’t social media as a category but the specific design pattern of infinite, algorithmically curated feeds that hijack intended use.

Key sources

Most people have tried and failed to quit social media

Strong evidence
+

A substantial share of users have attempted to reduce or quit social media and found themselves unable to. The pattern — repeated failed attempts to cut back on a behavior the person themselves considers undesirable — resembles clinical descriptions of compulsive behavior.

Evidence for

41% of users say they’ve tried to cut back on social media but were unable to. 34% have tried to take a break but couldn’t sustain it. Among millennials, 59% were unable to cut back as much as they wanted. Only 11% of users who tried quitting cold turkey lasted more than two weeks. Among teens, 61% who tried taking a break returned within a day. 52% of 18–29 year olds have attempted social media breaks, indicating the desire to reduce use is widespread even among the most engaged demographic.

Evidence against

Failed quit attempts don’t necessarily indicate addiction or compulsion — people also fail to stick to diets, exercise routines, and reading goals. Social media has real utility (staying in touch with friends, professional networking, news), so returning to it may reflect rational choice rather than compulsion. 34% of Gen Z report having quit social media entirely at some point, suggesting quitting is achievable. The framing of “addiction” is contested by researchers who argue it pathologizes normal behavior.

How to think about this

The most honest framing may be that social media platforms are deliberately designed to maximize engagement through variable reward schedules, infinite scroll, and notification systems — features borrowed from slot machine design. Whether this constitutes “addiction” in a clinical sense is debatable, but the pattern of use-against-stated-preference is real and widespread. The fact that the majority of users want to use these products less than they do is itself significant, regardless of the label.

Key sources