Evidence quality 4.88/5
Eight-dimension review score against the quality rubric . Each dimension scored 1–5.
- D1 Source grounding
- 5/5
- D2 Source authority
- 5/5
- D3 Arithmetic
- 5/5
- D4 Uncertainty
- 5/5
- D5 Scope
- 5/5
- D6 Prose
- 5/5
- D7 Perception honesty
- 4/5
- D8 Caveat completeness
- 5/5
How the risk varies
The headline figure averages across very different situations. Here’s how the probability varies by scenario or context:
—
No reliable per-user probability can be assigned. Correlational studies find small negative associations with well-being (r ≈ 0.035-0.15), but these are measured on continuous scales, not as binary harm incidence. For the typical teen user at moderate usage levels, the population-level effect size is too small and too heterogeneous to convert into a meaningful individual risk number.
—
Heavy use is associated with larger negative effects in most studies, but the dose-response curve is inconsistent across datasets and may reflect reverse causation (teens who are already struggling use social media more). No study has produced a reliable incidence rate for clinical-level harm among heavy users specifically.
—
The largest documented subgroup effects in the literature. Facebook/Instagram internal research (2021) and several external studies find that girls with pre-existing internalising symptoms report the most negative experiences. Effect sizes are larger than the population average but still measured on well-being scales rather than as binary harm rates. This is the subgroup where the case for a clinically meaningful effect is strongest and where a per-user probability would be most useful — but the evidence does not yet support one.
—
Most studies find smaller or null associations between social media use and well-being outcomes in boys compared to girls. Some studies find positive associations with social connection. The gender difference is one of the more consistent findings in the literature, though it may partly reflect differences in platform use patterns (visual comparison-based content vs. gaming/interest-based content) rather than a biological sex difference.
—
A distinct and newer exposure from feed-based social media. Common Sense Media (2025) found 72% of US teens have used an AI companion and about half use them regularly, but adoption rate is not a harm rate. The hypothesised mechanism — a private, always-available, sycophantic chat that can isolate a teen and fail to interrupt an escalating crisis — is drawn from individual cases (the Sewell Setzer III suicide and Garcia v. Character.AI; the OpenAI/Adam Raine case) and litigation now settling (the Google/Character.AI settlement in principle, January 2026), not from any cohort study with an incidence rate. No reliable per-user probability exists for crisis or self-harm outcomes tied to companion-bot use; the evidence base is adoption surveys and case reports, not population data.
Bar length and shade rank these scenarios against each other, not against other risks. The exact odds are shown beside each.
Decisions this risk informs
Choices that turn on this risk — how people weigh the trade-off.
Child social media account access
% who regret this choice
Action dominates
Action regret 5.4× higher
Pick challenger
This entry carries the no_reliable_estimate flag, and the reason is worth spelling
out: the research literature on social media and teen mental health has produced
thousands of studies, multiple government advisories, and at least one bestselling
book, but it has not produced a defensible per-user probability of “serious
psychological harm.” The best-powered correlational analyses — Orben and Przybylski’s
2019 specification-curve study across 350,000+ adolescents, and the meta-analytic
evidence synthesised by Odgers and Jensen’s 2020 Annual Research Review — find
associations in the range of r = 0.035 to r = 0.15. Those numbers are not zero. They are also not large enough
to convert into a meaningful statement like “1 in X teens will be seriously harmed.”
The outcome measures are continuous well-being scales, not binary harm thresholds,
and the gap between “scored slightly lower on a life-satisfaction questionnaire” and
“developed a clinical anxiety disorder because of Instagram” is one the data cannot
bridge.
The perception gap here runs in an unusual direction. The US Surgeon General’s 2023 Advisory — the most authoritative single document on the topic — explicitly states that “we do not yet have enough evidence to determine that social media is sufficiently safe for children and adolescents.” That framing, combined with the Facebook/Instagram internal research leak and Haidt’s “The Anxious Generation,” has moved public intuition toward treating social media as a clear and present danger to most teenagers. The correlational literature supports a weaker claim: there is a real but small population-level association, with larger effects concentrated in specific subgroups (girls aged 13-15, teens with pre-existing internalising symptoms, heavy users above 3-4 hours daily). Whether those subgroup effects constitute a causal mental health crisis or reflect reverse causation and confounding is the central open question, and the honest answer in 2026 is that the field has not resolved it.
Where the number would matter most — for the parent of a 14-year-old girl who already has anxiety and spends four hours a day on Instagram — is precisely where the evidence is most suggestive and least precise. The Facebook internal research found that 32% of teen girls who already felt bad about their bodies said Instagram made it worse, but “felt worse” on an internal survey is not the same as “developed a diagnosable condition,” and the survey was not population-representative. Until the field agrees on a clinical outcome definition and measures it prospectively in a large representative cohort, this entry will remain flagged as having no reliable estimate. That is not a statement that social media is safe. It is a statement that the evidence, as it stands, does not support a specific number.
Related tidbits
Social media harm to teens: no reliable causal estimate despite years of study. School bullying over grades 6-12: 65%. The evidence-free moral panic gets congressional hearings. The measured epidemic gets posters in the hallway.
54% of parents who let their child open social media accounts wish they hadn't. Among parents who delayed or blocked access, the estimated regret rate is about 10%.
Claim ledger
Every number below is what each source reported, with the verbatim quote we relied on and how we arrived at our figure. Click any link to verify directly.
-
[1] Office of the U.S. Surgeon General (2023) — Social Media and Youth Mental Health: The U.S. Surgeon General's Advisory
Social Media and Youth Mental Health: The U.S. Surgeon General's Advisory- Statistic
The advisory states there is 'not enough evidence to determine that social media is sufficiently safe for children and adolescents' and calls for urgent action, while noting the evidence base is insufficient to establish causation or quantify per-user risk- Excerpt
“"We do not yet have enough evidence to determine that social media is sufficiently safe for children and adolescents." ”
- Source data from
- 2023-05-23
- Accessed
- 2026-04-18 · archived copy
- Calculation
- The Surgeon General's Advisory is the highest-profile US government statement on teen social media risk. It is cited here as an authoritative framing document, not as a source of a specific probability. The advisory explicitly avoids stating a per-user risk figure and notes the evidence base is insufficient to determine whether social media is safe. It summarises the correlational literature and calls for more research, platform transparency, and legislative action. No native or normalized probability is derived from this source because the entry is flagged no_reliable_estimate.
-
[2] Nature Human Behaviour (Orben & Przybylski 2019) — The association between adolescent well-being and digital technology use
The association between adolescent well-being and digital technology useSee all 2 Likelier entries citing this source →
- Statistic
Across three large-scale datasets (total N > 350,000), the association between digital technology use and adolescent well-being was negative but small: r = −0.035 for social media use specifically, explaining about 0.4% of the variance in well-being — less than the negative association with wearing glasses or eating potatoes- Excerpt
“"The negative effect of technology use on adolescent well-being is small — explaining at most 0.4% of the variation in well-being." ”
- Source data from
- 2019-01-14
- Accessed
- 2026-04-18 · archived copy
- Calculation
- Orben & Przybylski 2019 is the most-cited methodological counterweight to the crisis narrative. The specification-curve analysis across all defensible analytic choices produces effect sizes that are statistically significant (large N) but tiny in practical terms. r = −0.035 for social media use means the association explains roughly 0.12% of the variance in well-being. This is not zero, but it is far too small to convert into a per-user probability of "serious psychological harm" — the outcome variable is a continuous well-being scale, not a binary harm threshold. This source is the primary basis for the no_reliable_estimate designation: the effect is real but not meaningfully translatable into individual risk.
- Independence
- Fully independent of the Surgeon General's Advisory and the Facebook internal research. Uses UK Understanding Society, YRBS, and MCS datasets.
-
[3] American Psychological Association (2023) — Health Advisory on Social Media Use in Adolescence
Health Advisory on Social Media Use in AdolescenceSee all 2 Likelier entries citing this source →
- Statistic
The APA identifies social media as neither uniformly beneficial nor uniformly harmful, and notes that effects depend on the content, the adolescent's individual characteristics, and the amount of time spent; it provides no per-user probability of harm- Excerpt
“"Science demonstrates that social media use, in and of itself, is not sufficient to be helpful or harmful to young people." ”
- Source data from
- 2023-05-09
- Accessed
- 2026-04-18 · archived copy
- Calculation
- The APA health advisory is included as a second authoritative institutional statement. Like the Surgeon General's Advisory, it summarises the correlational literature without producing a probability figure. Its key contribution to this entry is the explicit statement that social media use "in and of itself" is insufficient to cause harm — effects depend on content type, individual vulnerability, and usage patterns. This supports the no_reliable_estimate framing: there is no single population-level probability that applies to "a teenager using social media."
-
[4] Wall Street Journal (Wells, Horwitz, Seetharaman 2021) — Facebook Knows Instagram Is Toxic for Teen Girls, Company Documents Show
Facebook Knows Instagram Is Toxic for Teen Girls, Company Documents Show- Statistic
Internal Facebook research found that 32% of teen girls surveyed said Instagram made them feel worse about their body when they already felt bad about it; the research also found Instagram was cited as a contributor to anxiety and depression in a subset of teen users- Excerpt
“"Thirty-two percent of teen girls said that when they felt bad about their bodies, Instagram made them feel worse." ”
- Source data from
- 2021-09-14
- Accessed
- 2026-04-18 · archived copy
- Calculation
- The WSJ Facebook Files series leaked internal Instagram research that became the public catalyst for legislative and regulatory action. The "32% of teen girls" figure is widely cited but refers to a specific question — "when you already feel bad about your body, does Instagram make it worse?" — not to a base rate of developing a clinical condition. The leaked research used convenience-sampled internal surveys, not population-representative methods, and "felt worse" is a subjective self-report, not a clinical diagnosis of serious psychological harm. Included because it drove the public perception shift, not because it anchors a probability.
-
[5] Journal of Child Psychology and Psychiatry (Odgers & Jensen 2020) — Annual Research Review: Adolescent mental health in the digital age: facts, fears, and future directions
Annual Research Review: Adolescent mental health in the digital age: facts, fears, and future directions- Statistic
Synthesising six prior reviews/meta-analyses plus large-scale preregistered cohort and momentary-assessment studies, the review finds the link between digital technology use and adolescent mental health to be a mix of conflicting small positive, negative, and null associations (e.g. Huang 2017 meta-analysis r = −0.07 overall, depression r ≈ −0.11; McCrae 2017 r = 0.13), with no support for causal claims at the population level- Excerpt
“"At present, narrative reviews and meta-analytic work do not support causal claims, or even strong and consistent correlational patterns, linking adolescents' digital technology usage with mental health problems." ”
- Source data from
- 2020-01-17
- Accessed
- 2026-04-18 · archived copy
- Calculation
- Odgers & Jensen 2020 (J Child Psychol Psychiatry Annual Research Review) synthesises three evidence streams: narrative reviews and meta-analyses (2014-2019), large-scale preregistered cohort studies, and intensive longitudinal / ecological-momentary-assessment studies. Across them the reported associations are small and inconsistent in direction — the Huang (2017) meta-analysis (N = 19,652) found a mean SNS-well-being correlation of r = −0.07 (depression r ≈ −0.11), and the McCrae (2017) meta-analysis found r = 0.13 for depressive symptoms. The review's key contribution to this entry is its conclusion that the evidence base does not support causal claims, "or even strong and consistent correlational patterns," and does not warrant the widespread panic — while the outcome measures remain continuous well-being and symptom scales, not binary harm thresholds. (Earlier drafts of this entry mis-cited this review under a Clinical Psychological Science DOI that in fact resolves to Vuorre & Przybylski 2023, "Global Well-Being and Mental Health in the Internet Age" — a separate paper; the identifier was corrected to the real Odgers & Jensen 2020 DOI 10.1111/jcpp.13190 in 2026.)
- Independence
- Independent review drawing on a broader evidence base than Orben & Przybylski 2019, including six prior systematic reviews/meta-analyses, large-scale preregistered cohort studies (e.g. Understanding Society, N = 12,672), and diary/EMA designs.
-
[6] Common Sense Media (2025) — Talk, Trust, and Trade-Offs: How and Why Teens Use AI Companions
Talk, Trust, and Trade-Offs: How and Why Teens Use AI Companions- Statistic
In a nationally representative survey of 1,060 US teens aged 13-17 (April-May 2025), nearly three in four teens (72%) had used an AI companion at least once and roughly half were regular users; the figure is an adoption rate, not a rate of resulting psychological harm- Excerpt
“"Our new research reveals that nearly three in four teens have used AI companions, and half use them regularly." ”
- Source data from
- 2025-07-16
- Accessed
- 2026-06-14 · archived copy
- Calculation
- Added 2026-06-14 to document AI companion chatbots as a distinct emerging teen-harm vector separate from feed-based social media. Common Sense Media's nationally representative survey establishes that the exposure is now mainstream (72% of teens have tried one; ~half are regular users). It is cited strictly as an adoption-prevalence figure: the survey measures how many teens use companion bots, not how many are harmed by them. No native or normalized probability is derived because the entry remains no_reliable_estimate — surveys report adoption, not crisis incidence.
-
[7] UNICEF (2026) — 'Deepfake abuse is abuse' — Statement by UNICEF on AI-generated sexualised images of children
'Deepfake abuse is abuse' — Statement by UNICEF on AI-generated sexualised images of children- Statistic
In a UNICEF, ECPAT and INTERPOL study across 11 countries (Disrupting Harm Phase 2, ~1,000 internet-using children aged 12-17 surveyed per country), at least 1.2 million children disclosed having had their images manipulated into sexually explicit deepfakes in the past year; in some countries this represents 1 in 25 children- Excerpt
“"In a UNICEF, ECPAT and INTERPOL study across 11 countries, at least 1.2 million children disclosed having had their images manipulated into sexually explicit deepfakes in the past year. In some countries, this represents 1 in 25 children — the equivalent of one child in a typical classroom." ”
- Source data from
- 2026-02-04
- Accessed
- 2026-06-14 · archived copy
- Calculation
- Added 2026-06-14 to ground the AI-synthetic-media caveat. The original change brief attributed this 1.2-million figure to Thorn; that attribution could not be verified. The figure is actually from the UNICEF/ECPAT/INTERPOL Disrupting Harm Phase 2 study, cited here from the verbatim UNICEF statement. It is a cross-national exposure-disclosure rate (children reporting their images were manipulated into sexual deepfakes), not a per-user probability of psychological harm, so it does not change the no_reliable_estimate status. It documents that generative-AI synthetic media is a victimisation channel distinct from, and sometimes embedded within, social-media platforms.