Teenage girls describe negative mental-health effects from Instagram and TikTok at roughly twice the rate of teenage boys, which is the clearest demographic signal in the whole literature, and the gap is now consistent across academic syntheses, government surveys and independent polling. Social media anxiety is the cluster of worry, comparison and physiological arousal that builds in people who use platforms like Instagram, TikTok and Snapchat in ways that begin to feel out of their control, and 2025 made it impossible to discuss that cluster honestly without engaging the gendered split.
This piece walks through what the gap actually is, what the 2025 Open University of Catalonia (UOC) research and Pew Research Center’s 2025 teen survey each contribute, what plausibly drives the difference, and, most importantly, what it does and does not justify saying about adolescent girls. The framing matters: we want to take the data seriously without sliding into the kind of essentialism that treats girls as inherently fragile. The 2025 evidence does not support that reading. It supports something more useful, and more actionable.
Why this matters in 2025
For most of the last decade, the social-media-and-anxiety conversation was conducted in aggregate. Researchers and journalists pooled adolescents into a single category and asked whether average exposure produced average harm. The answer was usually “yes, a little”, small effect sizes, large debates, very little practical guidance. Then the 2025 work began routinely disaggregating by gender, and a sharper picture emerged. The aggregate effect was being driven, disproportionately, by what was happening to girls.
That shift changes what the Anxiety topic hub at this site can responsibly say about prevention. It changes what a parent should look for. It changes what platform regulators should target. And it changes the kind of advice a clinician working with a 14-year-old should be ready to offer. None of those changes require a panic. They require precision, which is what this article tries to provide.
A second reason 2025 matters: this was the year that government bodies, peer-reviewed syntheses and independent polling stopped contradicting each other on the gendered finding. In earlier years it was possible to read the US Surgeon General’s 2023 advisory as alarmist, the Pew polling as soft self-report, and the academic meta-analyses as too noisy to commit to. The 2025 papers closed that gap. Three independent evidentiary streams, qualitative platform research (UOC), nationally representative survey work (Pew), and pooled effect-size analysis (the Behavioral Sciences meta-analysis and the Journal of Adolescent Health review), now point the same direction and broadly agree on the magnitude. That convergence is rare in this field. It is the kind of result that ought to move advice, not just headlines.
The gap, sized
Pew Research Center’s 2025 survey is the cleanest single number. Asked whether social media has a positive, negative or no effect on their mental health, about 25% of US teen girls answered negatively, compared with around 14% of teen boys. The ratio is close to 1.8-to-1 and is more or less the same number Pew has been finding for several survey waves now, the gap is durable, not a one-off blip. [pew-2025-teens-social-media]
The academic literature gives a noisier picture in effect-size terms but the same direction of effect. The 2025 Behavioral Sciences meta-analysis of 24 primary studies finds that female-leaning effects are most pronounced for appearance-comparison and disordered-eating outcomes, somewhat smaller for generalised anxiety, and smallest for sleep complaints (where boys catch up). [mdpi-2025-meta-adolescents] The 2024 Journal of Adolescent Health systematic review on adolescents and anxiety likewise reports that the female-leaning gap is most visible in mid-adolescence, the window from roughly age 13 to 16, and starts to narrow into late teens. [tahir-2024-jah-syst-review] If you compare with our broader research roundup, the gendered finding is the single most consistent demographic split in the 2025 evidence base.
It is worth pausing on what a 1.8-to-1 ratio actually feels like in a classroom. If 30 teens sit in a Year 10 form room, the headline number predicts that roughly 4 of the boys and roughly 8 of the girls describe social media as actively bad for their mental health right now. Those are not the only ones with phones, and not the only ones being affected, they are the ones with the self-awareness and language to call the experience harmful. Underneath that number are larger groups on both sides whose harm is less articulate or less acknowledged, and a still-larger group whose use is genuinely unproblematic. The gendered split is a real signal, not an accusation against girls as a category.
What the UOC researchers found
The 2025 work coming out of the Open University of Catalonia is significant because it does not stop at the headline ratio. Its authors examine why girls’ platforms of choice produce worse outcomes, and they name specific design elements rather than gesturing at “social media” as a monolith.
Three findings carry the argument. First, the platforms with the strongest female-leaning harm signal, TikTok and Instagram, are the ones with the most appearance-centric content surfacing and the most visible engagement metrics (likes, followers, comment counts). Second, the algorithmic recommender on both platforms tends, in practice, to converge on appearance and lifestyle content for young female accounts even when those accounts have not asked for it. Third, the harm is concentrated not in users who scroll briefly but in users whose use becomes problematic, compulsive, comparison-driven, sleep-displacing.
That third point matters because it splits the population. Most teen girls use Instagram and TikTok without crossing into a clinically meaningful harm pattern. The harm is concentrated in a substantial minority. That subgroup, in the UOC framing, is who policy should target, not “all teen girls”, which is both wrong and patronising. Our piece on TikTok vs Instagram vs Snapchat goes deeper on how the platforms compare on these specific design dimensions.
A fourth UOC observation deserves attention because it is missing from most popular coverage. The researchers note that what girls see on these platforms is shaped by an interaction between what the recommender chooses and what previous engagement has reinforced. In other words, the feed a 14-year-old encounters today is not an exogenous shock; it is partly the product of micro-engagements she made at age 12, 13 and 14 that the system has been steadily learning from. That has two implications. First, the harm signal observed at age 15 reflects months of accumulated personalisation, not a fresh exposure. Second, interventions that simply restrict access for a week change today’s feed without resetting the recommender, which is why detoxes often produce smaller effects than people expect and why a meaningful intervention probably requires changing engagement patterns on return, not just taking time away.
What the Pew data adds
The 2025 Pew survey gives the gendered finding its representative-sample backbone. Where the UOC work is mechanistic and qualitative, Pew is the headcount. Three details from the 2025 wave deserve attention.
First, the 25%-vs-14% split is not just about extreme outcomes. Pew finds gendered gaps in milder negative experiences too, feeling worse about one’s life after scrolling, feeling tired after late-night use, feeling pressure to maintain a public online persona. Girls report all of these at higher rates than boys, with the multiplier ranging from roughly 1.3 to about 2 depending on the item.
Second, Pew finds that the gap is wider in older teens than younger ones. Among 13- to 14-year-olds the male-female difference is smaller; among 15- to 17-year-olds it opens up. That is consistent with the academic finding that mid- and late-adolescence is when appearance-based comparison becomes most psychologically loaded.
Third, and this is the under-reported piece, Pew documents that teens themselves are already aware of the gendered pattern. A majority of girls in the 2025 survey say they think social media is harder on girls than on boys, and a majority of boys agree. The harm is not invisible to the population experiencing it; what has been missing is structural acknowledgement.
The mechanisms most repeatedly named
Reading the UOC, Pew and academic syntheses together, four mechanisms come up again and again as the engine of the gendered split.
Appearance-based comparison. Both TikTok and Instagram are visual-first platforms, and the visual content disproportionately surfaced to female accounts is appearance-coded, face, body, clothing, lifestyle. Adolescent girls socialised since childhood to monitor their appearance enter these feeds with priming that boys do not have at the same intensity. The 2025 Behavioral Sciences meta-analysis finds that appearance-comparison mediates a substantial fraction of the female-leaning effect.
Public popularity metrics. Likes, followers, comment counts and view counts are public-facing on both platforms. Girls, again, because of how they have been socialised about social standing, engage with those metrics more anxiously than boys, and report more distress when the numbers underperform expectations.
Relational aggression amplified by group chats. Snapchat, group-message channels and quote-comment features on TikTok all turn what used to be private peer conflict into semi-public, durable record. Girls’ peer conflict has historically run through relational rather than physical channels, and platforms amplify that channel rather than dampen it.
Algorithmic compounding. A single appearance-coded interaction can shift the recommender for days. The 2024 Journal of Adolescent Health review notes that this compounding effect, engaging with one comparison post and then receiving more of the same, is structurally different from older media exposure patterns and is harder for individual users to break out of without active intervention.
None of these four mechanisms is essentially feminine. Each is a platform design choice or a socialisation pattern. Each is, in principle, modifiable. Our look at the signs and symptoms of a feed that has tipped over into harm covers what these mechanisms feel like from the inside.
Why this is not biology, and why the distinction matters
It is tempting to read the gendered split as a story about adolescent female biology, hormones, stress reactivity, identity formation differences. That reading is both unsupported by the 2025 evidence and editorially harmful, so it is worth saying clearly why.
Unsupported, because the pattern emerged historically. The roughly two-to-one ratio in teen girls’ negative mental-health reports is a 2010s-and-after phenomenon, tracking the rise of appearance-centric, algorithmically ranked platforms. If the cause were female biology, the gap should have been visible for decades; it was not. The 2025 UOC framing explicitly identifies the platform shift as the driver.
Harmful, because framing the gap as biological implies the harm is inevitable, that this is just how teen girls are, and the only available response is to be sad about it. That framing protects platforms from scrutiny and pathologises a normal developmental population. The 2025 evidence supports the opposite move: name the platform design choices that drive the gap, and ask whether they can be different. They can.
This is also why the right framing for parents is not “your daughter is fragile”. It is “your daughter is using a product designed to monetise the social-evaluative concerns of adolescence, and she would benefit from active partnership in pushing back against that design”.
Practical implications for parents and teens
The data should change what specific people do this week, not just what they think. A short list of what the 2025 evidence supports.
For parents of teen girls: keep phones out of the bedroom at night, this is one of the few interventions with consistent, replicable evidence. Engage the platforms with your daughter rather than only legislating around them; ask what her feed actually shows, sit and scroll with her once a week, treat the recommender as a force you are jointly studying rather than a black box you are afraid of. Watch for the problematic-use pattern, loss of control, distress when the phone is unavailable, late-night use displacing sleep, rather than minute counts.
For parents of teen boys: the lower headline number does not mean nothing is happening. Late-night use, gaming-adjacent content and harassment in group chats are the mechanisms most likely to be active. Boys often under-report; the absence of obvious appearance distress is not evidence of absence of harm.
For teens of any gender: structured pauses help more than total bans, which usually backfire. The 2025 detox literature finds modest but real benefits from one-week and longer breaks. Pair that with rebuilding offline friendship density, which is the most robust protective factor across the 2024–2025 syntheses. Our coverage of mothers’ anxiety shows that adults are not exempt from any of this, modelling matters.
For clinicians and teachers: screen for the problematic-use pattern, not for total time. Ask about appearance comparison, public-metric distress and night-time use specifically. These are the three places the gendered split most reliably lives, and they are answerable in a brief assessment. The 2025 evidence is also clear that sleep is the single most leveraged behavioural variable in this picture, devices out of the bedroom, charging in the kitchen overnight, with no exceptions, produces measurable changes inside two weeks for adolescents who can sustain the rule.
For policy-minded readers: the gendered finding is the strongest available case for design regulation rather than blanket age restriction. The mechanisms identified by UOC, visible like and follower counts, appearance-coded recommender outputs, semi-public group-chat aggression, are all design choices, not laws of nature. [uoc-tiktok-girls-2025] Several of them have been altered by individual platforms in narrow trials (Instagram’s hidden-likes experiment is the most cited example). That those design choices have not been more widely changed despite a decade of evidence is itself a finding about where the gendered harm is coming from, and where the leverage to reduce it sits.
A schematic of the mechanisms named in this article. Not measured data.
What we still do not know
Three open questions remain even after the 2025 syntheses. First, the experience of non-binary and gender-nonconforming teens is under-sampled in every dataset reviewed here; the binary split this article describes is a real pattern in the available data but is probably a simplification. Second, the cross-cultural generalisability of the gendered gap is uncertain, most of the 2025 work is US, UK and Spanish. Third, the durability question: will the gap close as platforms redesign under regulatory pressure, or persist because the underlying socialisation patterns persist? The honest answer is that we will only know in five years.
What we do know, and what the evidence supports as carefully as anything in this field, is that the gendered split is real, mechanistic, design-driven and modifiable. None of those four words requires panic. All four require precision.
When to seek help
Speak to a doctor if a teenager you know has been withdrawing, sleeping badly, or has stopped doing things they used to enjoy for more than a couple of weeks. The gap described here is a population finding and says nothing about any individual, which is worth holding onto in both directions: it is not a reason to watch a daughter more closely than a son, and it is not a reason to discount a boy who says he is struggling.
Go sooner if there is anything about appearance, eating or self-harm in what they are saying, and take examples rather than impressions. Our guide to teenage social anxiety covers the adjacent pattern, and the free anxiety screener (GAD-7) is one a teenager can do themselves in two minutes without an account.
Go urgently if they have talked about harming themselves. Contact your local emergency services or a crisis helpline.