~70 primary sources
Every claim in your roadmap, traced back to a primary source. This is the evidence that filtering-and-monitoring quietly backfires with teenagers — and the specific, sex- and identity-skewed online risks the plan is built to weight. Where a figure comes from an advocacy group, a brand, a single case, or a country that isn't the US, we say so plainly rather than round it up.
The plan front-loads filters and monitoring for little kids, then steps them down on a published schedule. That order isn't a compromise — it's what the evidence points to. Restriction is good at buying time and bad at building judgment, and it stops working right when the stakes climb.
Filters barely move the needle on whether a teen actually encounters explicit material — almost everything that predicts it is something a filter can't touch.
Under 0.5% of the variance in exposure is explained by internet filtering.1
Preregistered UK study of ~1,004 families; the <0.5% figure is verbatim from the OII. Filters still cut accidental exposure for young kids — that's why the plan uses them there.
Across the whole literature, strict rules and active coaching do two different jobs: restricting cuts how much time kids spend, while talking with them and co-using is what actually reduces harm.
52 studies, N = 74,159: restriction reduces screen time; active mediation and co-use reduce media-related harm.2
A meta-analysis pooling 122 correlations; international.
The same strict rule can protect a younger child and backfire on an older teen. The crossover isn't a matter of opinion — it lands at a measurable age.
Internet-specific rules predicted less problematic use under age ~12.3, and more over age ~15.7.3
Dutch cohort. The outcome is problematic social-media use (Social Media Disorder Scale); the exact thresholds are ages 12.31 (protective) and 15.70 (counterproductive).
Monitoring assumes kids will come to you when something goes wrong. Most say they would — then, in the moment, don't. That gap is the argument for building trust, not surveillance.
40% of minors said they'd tell a caregiver if an adult sent them a nude — but among those it happened to, only ~10% did.4
US minors; the 40%/10% are from Thorn's summary page (its underlying 2021 survey puts the caregiver-specific "actually told" figure nearer 6%).
Kids experience heavy parental-monitoring apps as a breach, not a safeguard — and they say so, loudly, in their own words.
76% one-star: in 736 children's reviews of 37 parental-control apps, most panned them as privacy-invasive and trust-damaging.5
US; 736 child-written app-store reviews. (This finding is often miscited to the 2020 "Circle of Trust" paper — the primary is this 2018 study.)
These are population-level skews, not destiny — plenty of individual kids don't fit them. They tell you which risks to weight for a given kid, not what your kid will become. Where a threat hits both sexes, it's noted.
Left to run, recommendation feeds don't stay neutral for a boy — they steer hard toward toxic "manosphere" content, fast, on accounts that did nothing to ask for it.
Fresh teen-boy accounts hit toxic content within 23 minutes; by the end of the test, 76% of TikTok and 78% of YouTube Shorts recommendations were toxic.6
A controlled experiment on blank accounts registered as 16- and 18-year-old boys — it shows what the algorithm pushes, not how many boys end up there.
The drift compounds over days: the more a teen-coded account engages, the more misogynistic the feed becomes.
Misogynistic recommendations rose from 13% to 56% over five days.7
Research accounts, UK schools context.
This isn't a fringe corner of the internet for young men — watching "masculinity influencers" is now close to the norm.
63% of young men (16–25) regularly watch masculinity influencers.8
Survey of 3,000+ young men across the UK, US and Australia.
Andrew Tate is a useful barometer for how far this reaches into ordinary boyhood — near-universal awareness, and a meaningful minority who like what they see.
84% of British boys 13–15 had heard of Andrew Tate; 23% held a positive view.9
UK; boys aged 13–15.
And the attitude splits sharply by sex — the same figure young men admire, young women mostly don't.
Among 16–25s, 32% of young men held a positive view of Tate versus 9% of young women.10
Treat as UK data despite the /us path in the URL.
The coercive "send more or I'll share these" pattern that historically targeted girls has been eclipsed by a scripted, money-driven version aimed squarely at boys.
Today's dominant sextortion pattern is financial, fast, and aimed at boys — the mirror image of the relational coercion that has long targeted girls.
90% of detected financial-sextortion victims are boys aged 14–17.11
US detected-victim data; the money motive makes it scripted and rapid rather than relational.
The volume is not a trickle — reports have exploded in three years.
NCMEC financial-sextortion reports rose from 10,731 (2022) to over 50,000 (2025) — about 137 a day.12
US.
It's an organized operation with a known playbook, not scattered opportunists — which is why it moves at scale.
Traced to Nigerian "Yahoo Boys" targeting boys mainly via Instagram, amid a reported ~1,000% rise in sextortion over 18 months.13
The NCRI report page was unreachable for us. The ~1,000%-over-18-months figure is NCRI's own (covering North America and Australia) — not an FBI statistic; it and the "Yahoo Boys" attribution were corroborated through secondary outlets before publishing.
The reason this one gets its own emergency script in the plan: the time between contact and crisis can be hours, and it has been fatal.
At least 36 American teen boys have died by suicide after being sextorted since 2021.14
US; "at least 36" is verbatim.
When kids do report it, it's usually them or a parent coming forward — which is exactly why the relationship, not the monitoring, is the safety system.
94% of public reports where the relationship is known came from the victim or a parent.15
This 94% is specifically among public reports where the reporter's relationship is known — most reports overall actually come from platforms.
There's now a legal lever for getting intimate images down quickly — useful for both boys' sextortion and girls' image-based abuse.
Covered platforms must remove non-consensual intimate images within 48 hours of a valid request.16
Signed May 19, 2025; the platform-compliance deadline is around May 2026.
First exposure now typically lands in elementary or middle school, and more often than not it's stumbled into rather than sought — which is the case for age-appropriate filters early.
Average first exposure is age 12; 15% by age 10 or younger; 58% encountered it by accident.17
US.
The likeliest doorway isn't a porn site — it's mainstream social media, and a lot of what kids find there is violent.
The most common place UK children see pornography is X/Twitter (45%), ahead of dedicated porn sites (35%); 58% of those exposed had seen strangulation content.18
UK; figures are from the August 2025 report.
Gaming isn't a threat in itself, but it's where boys spend disproportionate time — so several risks concentrate there.
Boys game more, and far more intensely, than girls — which is why the harms below land on them harder.
97% of US teen boys play video games versus ~75% of girls; 61% play daily versus 22%.19
US, n=1,453.
In-game voice and chat is a routine harassment channel for boys, sometimes escalating to threats.
48% of US teen boys who game have been called offensive names while playing (vs 32% of girls); 15% were physically threatened (vs 9%).20
US.
Harassment in multiplayer games is close to universal for young players, and the identity-based kind is rising.
75% of players aged 10–17 were harassed in online games (up from 67%); identity-based harassment rose to 37% (from 29%).21
US; 2023 data.
Games are also a documented recruiting surface for extremist ideology aimed at young players.
~9% of teen gamers were exposed to white-supremacist ideology in games, and the ADL logged its first apparent extremist recruitment attempt in five years.22
US; the ~9% figure is consistent across reporting but is drawn from the full report — treat as medium-confidence.
The clinical end of "too much gaming" also skews male.
Internet gaming disorder was nearly twice as prevalent in boys as girls: 6.7% vs 3.6%.23
A single-province Chinese cohort of impoverished rural adolescents (Hunan); the boy-vs-girl split is the point, but treat the absolute rates as local, not international.
Loot boxes are the gambling-shaped mechanic inside kids' games, and boys buy in at triple the rate.
Boys are 3× more likely than girls to pay to open loot boxes — 37% vs 11% (ages 11–16).24
UK; ages 11–16.
And that mechanic is consistently linked to problem gambling — every study that has looked found the association.
Every study in the review that examined the link (12 of 16) found a positive association between loot-box spending and problem gambling; in one sample, 45.9% of buyers scored as problem gamblers.25
Scoping review; international.
The most extreme end of online predation now includes organized violent networks that coerce kids into self-harm — the risk behind the plan's most serious warnings.
The FBI's "764" alert documents violent online networks targeting 10–17-year-olds via social media, gaming platforms, and messaging apps.26
US; the PSA names "gaming platforms" (our shorthand "in-game chat" is a paraphrase).
Cold solicitations and the "let's move to a private app" move are common enough that every kid needs a scripted response ready — not just boys, but the volume is high across the board.
40% of minors (9–17) have received an online cold solicitation for a nude; 65% have been asked to move to a private conversation on another platform.27
US; ages 9–17.
AI companions are already mainstream among teens — and independent review found them unsafe for minors.
72% of US teens have used an AI companion; Common Sense Media assessed them as not safe for anyone under 18.28
US national survey.
Kid-heavy platforms also draw ordinary scams and malware — Roblox is a standout target.
Kaspersky detected about 1.6 million attacks disguised as Roblox-related files in 2024.29
Vendor telemetry (Kaspersky); the 1,612,921 count is verbatim.
Again: these are skews, not fixed outcomes. The point is to know which risks to weight — for many girls the pattern is body-image pressure, coercion, and relational harm rather than the boy-skewed threats above.
The teen mental-health decline is real and heavily gendered — girls report distress at roughly twice the rate of boys.
52.6% of US high-school girls reported persistent sadness or hopelessness versus 27.7% of boys; poor mental health in the past 30 days, 38.8% vs 18.8%.30
Nationally representative US government survey.
The gap holds at the most serious end, too.
In 2023, 12.6% of US high-school girls attempted suicide in the past year versus 6.4% of boys; 27.1% seriously considered it versus 14.1%.31
Primary CDC data, US.
Compulsive social-media use skews female — and it's climbing.
13% of adolescent girls show "problematic" social-media use versus 9% of boys; the overall rate rose to 11% (2022) from 7% (2018).32
~280,000 youth across Europe, Central Asia and Canada — not the US.
The dose-response link between heavy use and depression is stronger for girls.
At 5+ hours/day, girls' depressive-symptom scores ran ~50% higher versus ~35% for boys.33
Peer-reviewed UK cohort, n=10,904.
For an account that looks like a struggling girl, the feed doesn't just fail to help — it actively escalates toward the worst possible content, within minutes.
A new teen account was served suicide content in 2.6 minutes and eating-disorder content in 8 minutes; a girl-coded "lose weight" account got 12× more self-harm recommendations.34
An advocacy group's controlled bot experiment — it shows what the feed serves a signaling account, not how common this is across all users.
Signal an interest in mental-health content and, after a few hours, roughly half of what you're shown is potentially harmful.
After 5–6 hours, ~1 in 2 videos shown to mental-health-interested accounts were potentially harmful — about 10× the baseline.35
Research accounts; this study wasn't itself broken out by gender.
The stakes of that pipeline are not hypothetical — a coroner formally tied it to a child's death.
A UK coroner ruled online self-harm content contributed "more than minimally" to 14-year-old Molly Russell's suicide; she had engaged with ~2,100 of ~16,300 saved posts on such themes in her final six months.36
A single UK inquest (n=1) — illustrative of the mechanism, not a prevalence figure.
The platform's own researchers found it makes body image worse for a large share of teen girls.
Instagram made body-image issues worse for 1 in 3 teen girls; about 48% engaged in appearance-based social comparison.37
Leaked internal Meta research, accessed via an archive of the slides; the 1-in-3 figure (corroborated by WSJ reporting and testimony) refers specifically to teen girls who already felt bad about their bodies, not all teen girls — Meta disputes the broader reading.
Filtering and editing one's own face is now routine for girls well before they're teenagers.
80% of girls have used a filter or photo-editing app to change their looks by age 13; 67% try to change or hide a body part before posting.38
Brand-commissioned, n=503, Canada — treat as medium-confidence.
Cyberbullying tilts female among older teens, and the flavor girls get is disproportionately about their appearance.
54% of US girls 15–17 have experienced cyberbullying (vs 44% of boys); appearance-based harassment, 17% vs 11%.39
US, teens aged 15–17.
Girls are also far more likely to be harassed because they're girls, and to be sent unsolicited sexual images.
Gender-based harassment hits girls at 14% vs 6% for boys; 22% of 15–17-year-olds have been sent explicit images they didn't ask for.40
US; "gender-based harassment" is broader than sexual harassment specifically.
Unwanted sexual solicitation online is overwhelmingly something girls experience.
US girls are roughly 3–4× more likely to be sexually solicited online — 34.6% of women versus 9.2% of men reported it before age 18.41
Peer-reviewed US national survey of young adults recalling childhood, n=2,639.
The same holds for being deliberately groomed by an adult online.
Grooming by an adult before age 18: 8.4% of women versus 2.4% of men.42
Same peer-reviewed US study.
And image-based sexual abuse in childhood lands mostly on girls.
Image-based sexual abuse before age 18: 16.3% of women versus 5.4% of men; online child sexual abuse overall, 23.3% vs 7.6%.43
Same peer-reviewed US study.
The newer, AI-generated form of this abuse is almost entirely aimed at girls.
Of AI-generated child sexual abuse imagery the IWF assessed where sex was recorded, 98% depicted girls.44
UK; verbatim from the IWF.
So is "self-generated" abuse imagery, especially among the youngest children.
Girls were 99% of "self-generated" abuse reports among 7–10-year-olds in 2023; 11–13-year-old girls appeared in 51% of all actioned reports.45
UK.
And the broader deepfake-porn ecosystem — the tooling and demand behind it — targets women and girls almost exclusively.
99% of the people targeted in pornographic deepfakes are women and girls.46
Secondary / fact-check: we reached this via a fact-check organisation; the underlying "State of Deepfakes" report (Security Hero) is commercial and not peer-reviewed, and it describes the adult-dominated ecosystem. (Sensity's often-quoted 2019 "96%" measures a different thing — the share of deepfakes that are pornographic, not the share depicting women.)
One honest nuance: while the generated imagery is overwhelmingly female, being personally targeted by a deepfake nude is not cleanly a girls-only problem.
About 1 in 17 (6%) of young people 13–20 say someone made a deepfake nude of them — reaching ~10% among younger teen boys as well as young-adult women.47
US survey, n=1,200, ages 13–20. This is the one place the image-abuse picture is not female-skewed at the victim level, even though the imagery itself is 98–99% female.
Read this section carefully. Being LGBTQ+ is not the risk. The risk is how others target these kids — and, cut the wrong way, the loss of a community that the same research shows is often a lifeline. That's why the protective findings come first: for many of these kids, blunt bans and monitoring remove a support system before they remove any danger.
For many LGBTQ+ young people, online space is where they feel safest and can be themselves — often more so than anywhere offline.
44% felt very safe in online spaces versus 9% in person; trans and nonbinary youth disclosed their identity online at twice the in-person rate (80% vs 40%).48
US, ages 15–24, n=1,200+. Blunt monitoring or bans risk cutting exactly this lifeline.
For a majority, affirming community online is easier to find than at home or school — often the first place they meet peers and get accurate health information.
68% of LGBTQ+ young people access LGBTQ-affirming spaces online — a higher share than at home or school.49
US national survey.
And the absence of any safe online space tracks with worse mental health — which is the case for safer spaces, not fewer of them.
Youth who felt safe nowhere online (~7%) reported higher past-year suicidal ideation (47% vs 38%) and depression (63% vs 52%) than peers with at least one safe online space.50
US; an association, not proof of causation — but the direction argues for safer access, not less.
These kids know what's at stake, and they experience heavy-handed content restrictions as a threat to their support system.
76% were seriously concerned about government restrictions on LGBTQ+-affirming online content.51
US; blocking felt by these youth as removing a support system rather than a danger.
With that lifeline established: LGBTQ+ minors are also targeted more, and more coercively, online.
Roughly 2–3× more likely than non-LGBTQ+ peers to have experienced unwanted or risky online interactions.52
US minors; they were also ~10 points more likely to handle it alone rather than tell an adult.
Sextortion hits them at double the rate, and the threat of being outed makes it uniquely coercive.
36% of LGBTQ+ teens 13–17 had experienced sextortion versus 18% of non-LGBTQ+ teens; their extorters were likelier to carry out threats (26% vs 13%).53
US survey, ages 13–20; the LGBTQ+ subgroup figures rest on small bases (n≈77–78, which Thorn flags as under 100). Threats to distribute imagery or personal information ran ~12 points higher — a distinct lever against a closeted teen.
The underlying mental-health disparity is large, and largest for trans and nonbinary youth.
39% of LGBTQ+ young people seriously considered suicide in the past year — 46% of trans/nonbinary versus 30% of cisgender LGBQ peers; 12% attempted.54
US, n=18,663, ages 13–24.
Body dissatisfaction is nearly universal in this group, and it's tightly linked to suicide risk.
87% of LGBTQ youth reported body dissatisfaction (up to ~93% among trans youth); dissatisfied youth had about double the odds of a past-year suicide attempt (15% vs 7%).55
US 2022 National Survey, n=33,993.
A lot of the hostility is manufactured: coordinated "groomer" disinformation surged after ownership changes at one major platform.
1.7M+ tweets since 2022 paired LGBTQ+ terms with slurs like "groomer" or "predator" — up 119% after Musk's takeover.56
An advocacy group's primary analysis of X/Twitter.
That online narrative has an offline tail — the "groomer" conspiracy shows up repeatedly at real-world anti-LGBTQ+ incidents.
Of 350+ anti-LGBTQ+ incidents logged over ~10 months, the "groomer" trope was cited in 191 — the most common.57
Secondary reporting: the ADL and GLAAD primary reports were unreachable for us, so this cites ABC News relaying them. Treat as medium-confidence.
Where these kids are heading as adults is worse still — useful as trajectory context, not a statistic about minors.
~7 in 10 LGB adults have been harassed online (versus ~4 in 10 straight adults); 51% faced severe abuse versus 23%.58
These are LGB adults (18+), not minors — included only as trajectory context.
A few honest ground rules, so you can weigh each number for yourself rather than take it on faith.