Almost every American teenager is now online. Up to 95% of youth aged 13 to 17 use a social media platform, and more than a third say they use one "almost constantly". ¹ That near-universal exposure has turned a research question into a political one.

In spring 2023, US Surgeon General Vivek Murthy released an advisory stating there is growing evidence that social media is causing harm to young people's mental health. ² The same year, the American Psychological Association issued its own guidance, striking a more cautious note: social media is not inherently beneficial or harmful, and its effects depend on an adolescent's personal characteristics, the content they see, and their circumstances. ¹

Between those two poles sits a large and uneven body of research spanning the US, UK, Australia, the Netherlands and beyond. Some of it is primary government survey data; some is peer-reviewed longitudinal analysis; some is advocacy. This report follows the evidence question by question, keeping the grade of each finding visible throughout.

The most demanding test of harm is to follow the same young people over time and see whether more social media use predicts worse mental health later. When researchers do that carefully, the alarm often fades. An analysis of 3,205 Australian adolescents across two survey waves found little evidence of a longitudinal relationship between changes in messaging, posting or viewing behaviors and either psychological distress or mental wellbeing over twelve months. ¹ A UK study tracking 3,228 adolescents aged 10 to 15 found that, in adjusted analysis, more time on social media was not significantly associated with poorer mental health two years later. ¹

Yet the picture is genuinely divergent, not simply reassuring. A separate UK longitudinal study reported a trend linking increased social media time to diminished mental health, while explicitly cautioning that such an association does not necessarily imply cause and effect. ¹ And an analysis of the 2023 US Youth Risk Behavior Survey, covering 10,334 adolescents, found moderate social media use associated with 1.31 times greater odds of depressive symptoms, rising to 1.77 times for high use, a clear gradient. ¹ Roughly speaking, heavy users had about three-quarters higher odds of depressive symptoms than lighter users.

These four studies appear to conflict, but much of the gap is methodological rather than substantive. As the analyst's triangulation notes, the two null-finding studies rely on adjusted, over-time analyses, whereas the YRBS finding is cross-sectional, a single snapshot that cannot separate whether platforms depress teens or depressed teens use platforms more. ³ The studies also measure different things: "psychological distress," general "mental health," and "depressive symptoms" are not interchangeable constructs, and they span age bands from 10 to the late teens. One review of the wider literature captures the ambiguity numerically: of 25 studies examined by McCrory and colleagues in 2020, four suggested no relationship at all between social media use and young people's mental health, meaning most did find some association, but far from all. ¹

The headline dose-response figure most often cited alongside the Surgeon General is stronger and better corroborated: teens aged 12 to 15 who used social media more than three hours daily faced twice the risk of negative mental health outcomes, including depression and anxiety symptoms. ² Two independent origins, Yale Medicine and the Surgeon General's advisory itself, report this identical finding from a sample of 6,595. ² It remains, however, an observational association at a single point in developmental time.

Open: Whether the divergence between cross-sectional and adjusted longitudinal findings reflects reverse causation, differing outcome measures, or genuinely different effects at different ages remains unresolved within this evidence set.

For all the focus on harm, young people themselves report that social media does real good. In Pew Research Center's 2024 survey, 74% of teens said social media makes them feel more connected to their friends' lives. ¹ The Surgeon General's advisory, drawing on earlier survey data, reported majorities feeling more accepted (58%), supported through tough times (67%), given a place for creativity (71%), and more connected to friends (80%). ¹ The connection figures from the two sources agree once the different survey years are accounted for, converging on roughly three-quarters to four-fifths of teens. ²

For some young people the platforms function as a lifeline. The Digital Wellness Lab at Boston Children's Hospital reports that 68% of LGBTQ+ youth had access to affirming spaces online, 14 points higher than the next most cited source, school. ¹ The same population's relationship with these platforms is strikingly two-sided: 96% of LGBTQ+ youth described social media as positive for their mental health, and 88% described it as negative. ¹ Both figures can be true because the same person can experience both. It is worth noting these benefit figures rest on single origins, the Digital Wellness Lab citing Trevor Project data, without independent corroboration in this evidence set. ¹

Benefits are not evenly experienced, and they are eroding. Pew's tracking shows the share of teens saying social media has a mostly negative effect on people their age climbed from 32% in 2022 to 48% in 2024, a 16-point jump. ¹ Over the same window, the share feeling supported through tough times fell from 67% to 52%, a 15-point drop. ¹ The arithmetic on both shifts checks out against Pew's own baseline figures. ¹ A diary study of Dutch adolescents adds nuance from within individuals: time spent on Snapchat positively affected friendship closeness for 71.5% of adolescents, well-being for 41.4%, and self-esteem for 23.7%. ¹ In other words, the most social benefit is widely shared, but positive effects on self-esteem reach only about one in four.

Open: Whether the two-year decline in teens' positive perceptions reflects changing platforms, changing users, or heightened public alarm cannot be distinguished from the survey data alone.; The LGBTQ+ affirming-spaces and positive/negative sentiment figures lack independent corroboration beyond the Trevor Project data cited.

Correlation cannot settle the question; experiments and interventions come closer. Here the evidence is genuinely suggestive of causation, but sparse. A 14-day social media abstinence trial among 86 female participants found significant decreases in body image dissatisfaction and screentime for the abstinence group. ¹ A review cites Hunt and colleagues' 2018 randomized trial, in which limiting social media to roughly 30 minutes a day over three weeks reduced loneliness and depressive symptoms. ¹ The Surgeon General's advisory reports a natural experiment exploiting the staggered rollout of a platform across US colleges, spanning 359,827 observations, associated with a 9% increase in depression and a 12% increase in anxiety over baseline. ¹

The intervention evidence, however, does not run all one way. In the same abstinence trial, depression, anxiety, problematic social media use, fear of missing out, and loneliness all decreased over the intervention period in both the abstinence and control groups, with no significant difference between them. ¹ That pattern weakens the case that removing the platform, specifically, drove improvement. A meta-analysis pointing in a more constructive direction found that social-media-based interventions, that is, programs delivered through the platforms rather than restrictions on them, reduced mental health issues with an overall effect size of 0.32 across 17 randomized trials involving 5,624 participants, with separate effects on anxiety, depression and stress. ¹ This is a moderate effect, but it rests on a single meta-analysis with no competing synthesis in this evidence set.

Authoritative bodies are candid about why causal certainty remains elusive. The APA states that findings suggesting causal associations are rare because the data required for cause-and-effect conclusions are challenging to collect or held within tech companies, and that long-term multiyear longitudinal research is often unavailable, leaving associations with outcomes into adulthood largely unknown. ¹ One further causal-sounding claim, that habitual checkers exceeding 15 times a day showed altered developmental trajectories in the brain's reward regions, comes from a single non-peer-reviewed source of unclear affiliation and cannot be treated as established. ¹

Open: The natural-experiment estimate of a 9% rise in depression and 12% rise in anxiety, reported by the Surgeon General's advisory, has no independent replication here.; Whether the neuroimaging claim about habitual checkers reflects a real developmental effect awaits location of a peer-reviewed original study.

The strongest through-line in the evidence is that effects are not uniform. The APA's core position is that social media is not inherently beneficial or harmful, and its impact depends on adolescents' personal characteristics, the content they encounter, and their context. ¹ The data bear this out repeatedly. In the YRBS analysis, higher parental monitoring significantly buffered the negative association between social media use and depressive symptoms, with protective interaction terms of 0.80 and 0.87. ¹ The same analysis found that gender did not significantly moderate the association with depressive symptoms, its interaction terms non-significant, complicating any simple claim that girls are uniformly more vulnerable to depression from social media. ¹

Body image is where a gender difference does surface sharply, and where a heavily circulated statistic lives. The National Center for Health Research reports that leaked 2021 Instagram documents found about one in three teen girls felt worse about their bodies from using the app, compared with 14% of boys. ¹ That figure is graded as reported: it rests on a single advocacy source in this set, and it reflects self-reported perception during platform use rather than a clinical outcome. Its wide circulation, however, is itself part of the story.

Socioeconomic background shapes not just whether harm occurs but what young people do online in the first place. A Cambridge literature review reports that youth from higher socioeconomic backgrounds tend toward capital-enhancing digital use aimed at education and career, while those from lower backgrounds tend toward entertainment use. ¹ Aggregate effect estimates can therefore mask opposing realities. The Dutch diary study found overall within-person effects to be small, with well-being (β=-0.08), self-esteem (β=-0.09) and friendship-closeness (β=-0.06) coefficients all small in magnitude, even as a majority of the same adolescents experienced positive friendship effects. ¹ And in the UK mediation analysis, 68% of the unadjusted effect of social media on mental health ran through self-esteem, suggesting that pre-existing vulnerability, not the platform alone, carries much of the load, though this comes from an unadjusted analysis and cannot be equated with the same study's adjusted null finding. ¹

Open: No study in this set stratifies simultaneously by individual vulnerability, content type, and active-versus-passive usage, so the interaction of these dimensions remains unmeasured.; Whether socioeconomic usage patterns translate into different mental-health outcomes, rather than just different activities, is not directly tested.

The evidence supports four rival readings of what social media does to young people, and they are not equally well supported.

One explanation is that social media use causes dose-dependent harm: more exposure, worse mental health. The dose-response gradient in the YRBS survey, the doubled risk at three-plus hours daily, the UK trend, and the college rollout natural experiment showing rises in depression and anxiety all point this way. ⁴ But two large adjusted longitudinal studies directly contradict it, finding no significant association over time. ² This reading is plausible but not established, because its strongest causal plank, the natural experiment, is reported by the Surgeon General's advisory but has no independent replication in this evidence set.

A second reading holds that the observed associations are spurious, driven by reverse causation or shared third factors, for instance pre-existing distress driving both increased use and poor mental health. The adjusted null findings, the APA's warnings about missing causal data, and the abstinence trial in which both groups improved equally support this. ⁴ It is undercut by the experimental studies that do show change when exposure is manipulated. ³ Also plausible, and unresolved.

The best-supported reading is that effects are highly heterogeneous, varying by individual, content and usage. No confirmed claim contradicts it, and many support it: the APA's context-dependence position, the self-esteem mediation, the null gender effect for depression alongside a strong gender gap for body image, the parental-monitoring buffer, the small average but widely varying diary effects, and the socioeconomic split in usage. ⁴ This hypothesis absorbs the apparent contradictions between the others.

A fourth reading is that benefits outweigh or coexist with harms for most youth. High majorities report connection and support, most Dutch adolescents saw positive friendship effects, LGBTQ+ youth rely on affirming online spaces, and platform-delivered interventions reduce symptoms. ⁴ Against it: rising negative perceptions and a single advocacy source reporting that psychological distress among New Zealand 18-24-year-olds more than quadrupled over 14 years. ² This is plausible but sits in tension with the deteriorating sentiment trend, and the strongest contradicting harm claim is only weakly sourced. ¹

Open: A preregistered trial randomly assigning adolescents to graded exposure levels over a year, or genetically informed designs testing confounding, would discriminate between the harm and spuriousness readings but does not exist in this evidence set.

The evidence does not force a verdict of crisis, nor of exoneration. It forces a verdict of conditional, heterogeneous, and partly unproven effect. The single most authoritative statement in the record is the APA's: causal findings are rare because the necessary data are held within tech companies, and long-term effects into adulthood are largely unknown. ¹ Every stronger claim, in either direction, has to be read against that limit.

What can be stated with confidence is narrow but real. Nearly all teens use these platforms. ¹ Majorities derive genuine social benefit from them, and for LGBTQ+ youth the affirming function is documented and substantial. ³ At the same time, heavy use is associated with elevated depressive symptoms in cross-sectional national data, and controlled interventions that reduce or remove use can improve loneliness, depressive symptoms and body image. ³ These are not contradictory once heterogeneity is accepted: the same technology helps some youth and harms others, and context, including parental monitoring and content, shapes which. ²

Where the evidence genuinely stalemates is on population-level causation. Two well-designed longitudinal studies find nothing over time; a national survey finds a gradient; a natural experiment, reported by the Surgeon General's advisory, finds harm but has no independent replication in this evidence set. ⁴ That is not a resolved question dressed as an open one; it is an open question.

One speculative implication, labeled as such and not established by the claims: if the deterioration in teens' own assessments continues, with negative perceptions rising 16 points and felt support falling 15 points in just two years, the benefit-dominant reading may weaken over time even if the underlying platforms do not change. ¹ This is a projection from a two-point trend, not a finding, and the reasoning is offered transparently rather than as a conclusion the evidence compels.