Reference
Social media harm, by the numbers
Forty figures documenting what these platforms earn, what they do to people, what their own researchers found, and what happens when you leave. Every number traces to a primary document, court record, regulatory finding, or peer-reviewed study — with the year and the source named beside it. Free to cite; attribution appreciated.
Section 01
The business model
What your attention is worth, and where the money comes from.
| Figure | What it measures | Source | Year |
|---|---|---|---|
| $200.97B | Meta's total annual revenue | Meta earnings reports | 2025 |
| 98% | Share of Meta's revenue that comes from advertising | Meta earnings reports | 2025 |
| $57.03 | Meta's average annual revenue per user worldwide (up from $49.63) | Meta earnings reports | 2025 |
| $250–350 | Meta's annual revenue per North American user — 5–6× the global average | Meta earnings reports | 2025 |
| 9% | Year-over-year rise in average price per ad on Meta | Meta earnings reports | 2025 |
| ~$16B | Revenue Meta internally projected from scam, fraud, and banned-goods ads — about 10% of 2024 sales | Reuters, internal documents | 2025 |
| 15M/day | Higher-risk scam advertisements Meta showed users daily | Reuters, internal documents | 2025 |
Section 02
Engineered outrage and addiction
The mechanics, measured — mostly by the companies themselves.
| Figure | What it measures | Source | Year |
|---|---|---|---|
| 5× | How much more Facebook's algorithm weighted an "angry" reaction than a like | The Facebook Files (WSJ) | 2021 |
| 689,003 | Users whose feeds Facebook secretly manipulated to test emotional contagion, without consent | PNAS 111(24) | 2012/2014 |
| 13% | Increase in sharing per additional moral-emotional word in a post (meta-analysis, N=4.8M posts) | PNAS Nexus 4(11) | 2025 |
| 64% | Share of extremist-group joins on Facebook driven by its own recommendation algorithm | Internal study, Facebook Papers | 2020 |
| 35 min | TikTok's internal estimate of time to "addiction" for a new user — about 260 videos | Unsealed TikTok documents | 2024 |
| 1.5 min | Actual daily reduction in teen usage from TikTok's 60-minute screen-time "limit" (108.5 → 107 min) | Unsealed TikTok documents | 2024 |
| 2.6 min | How quickly TikTok recommended suicide-related content to accounts registered as 13-year-olds | CCDH, "Deadly by Design" | 2022 |
| 93% | Share of videos about depression served to a WSJ test account after ~36 minutes of profiling | WSJ bot investigation | 2021 |
Section 03
Teen mental health
Including the findings Meta's own researchers produced and its executives read.
| Figure | What it measures | Source | Year |
|---|---|---|---|
| 1 in 3 | Teen girls for whom Instagram makes body image issues worse — Meta's own internal finding | Internal Meta slide, leaked | 2020 |
| 66% | Teen girls reporting negative social comparison on Instagram, in Meta's internal research | Internal Meta research | 2020 |
| 13% | British teens with suicidal thoughts who traced the desire to Instagram, per Meta's research (US: 6%) | Internal Meta research | 2021 |
| 51% | Instagram users reporting a bad or harmful experience in the prior seven days, per Meta's own surveys | Béjar, Senate testimony | 2023 |
| ~2% | Share of reported harmful content that was actually removed, per the same internal surveys | Béjar, Senate testimony | 2023 |
| ~145% | Rise in major depression among U.S. teen girls, 2010–2020 | Haidt; Twenge & Campbell | 2010–20 |
| 0 of 85 | Self-harm posts removed by Instagram in a month-long Danish study, against a claimed ~99% removal rate | Digitalt Ansvar | 2024 |
| 8 of 47 | Instagram teen-safety features that worked as advertised when independently tested | "Teen Accounts, Broken Promises" | 2025 |
Section 04
Bots, fake accounts, and manufactured opinion
How much of the "conversation" is real — and how little it takes to bend it.
| Figure | What it measures | Source | Year |
|---|---|---|---|
| 51% | Share of all web traffic that was automated — the first year bots passed humans | Imperva Bad Bot Report | 2024 |
| 3.5B | Fake accounts Facebook removed in a single year — about 43% of the world's population | Meta transparency data (VAB) | 2025 |
| 2.7× | Peak ratio of fake accounts banned to actual monthly active users | Meta transparency data (VAB) | 2019 |
| 2–4% | Share of a network that needs to be bots to flip its perceived majority opinion, in 2 of 3 simulations | European Journal of Information Systems | 2019 |
| 6% → 31% | Bots were 6% of accounts but spread 31% of tweets linking to low-credibility content | Nature Communications 9 | 2018 |
| 18M of 22M | Public comments to the FCC on net neutrality that were fake; 8.5M used real people's stolen identities | NY Attorney General | 2021 |
| 7,704 | Facebook accounts removed in the "Spamouflage" takedown — the largest known covert influence operation | Meta Adversarial Threat Report | 2023 |
| 81 of 81 | Countries surveyed that showed organized social media manipulation campaigns | Oxford Internet Institute | 2020 |
| 0.32% → 80% | 2,107 "supersharers" — a third of one percent of a voter panel — spread 80% of election fake news | Science 384 | 2024 |
| 25% → 97% | The most active quarter of U.S. adults on Twitter produced 97% of all tweets | Pew Research Center | 2021 |
Section 05
Fines, verdicts, and regulatory findings
What courts and regulators have concluded, with numbers attached.
| Figure | What it measures | Source | Year |
|---|---|---|---|
| $5B | FTC penalty against Facebook — largest ever against a tech company at the time | US Federal Trade Commission | 2019 |
| 87M | Users whose data Cambridge Analytica harvested without consent | Guardian / NYT investigation | 2018 |
| €1.2B | Largest GDPR fine ever issued, against Meta for EU–US data transfers | Irish Data Protection Commission | 2023 |
| $1.4B | Texas biometric-data settlement with Meta | Texas Attorney General | 2024 |
| >$9B | Cumulative fines and settlements paid by Meta across all jurisdictions | Compiled regulatory actions | 2018–24 |
| ~$942M | Meta's total liability in New Mexico's child-safety case ($375M verdict + $567M abatement) | N.M. First Judicial District | 2026 |
| 126M | Americans reached by Russian Internet Research Agency content on Facebook | Senate Intelligence Committee | 2015–17 |
Section 06
What happens when you leave
The other side of the ledger — all from randomized trials.
| Figure | What it measures | Source | Year |
|---|---|---|---|
| 25–40% | Well-being gain from 4 weeks off Facebook, as a share of the effect of clinical depression treatment | American Economic Review 110(3) | 2020 |
| ~1 hour | Time returned per day by deactivating Facebook — redistributed to socializing, exercise, and sleep | American Economic Review 110(3) | 2020 |
| 12% | Lasting voluntary reduction in Facebook use after a paid four-week deactivation ended | American Economic Review 110(3) | 2020 |
| 1 week | Time off social media that produced significant improvements in well-being, depression, and anxiety | Cyberpsychology (Univ. of Bath RCT) | 2022 |
| 30 min/day | Cap that significantly reduced loneliness and depression within three weeks | J. Social & Clinical Psychology (UPenn) | 2018 |
| g = 0.17–0.25 | Pooled effect size of reducing social media, across 42 randomized trials — small but consistent | Two meta-analyses (N=1,491; N=5,544) | 2024–25 |
| ~821 hours | Annual cost of average social media use — 34 full days, or 51 waking days | DataReportal / We Are Social | 2024 |
On using these numbers. Every figure above is linked to its primary source on the Sources page and discussed in context on the page named beside it. Two honest caveats worth carrying with the numbers: effect sizes in the quitting literature are small-to-moderate averages, not promises of transformation; and every bot and fake-account estimate is exactly that — an estimate, made harder to verify since Meta shut down its CrowdTangle research tool in 2024. Where evidence is contested, the case files say so.