Social Media Companies Have Internal Research Showing Their Products Harm Teens

This is no longer an allegation. It is a documented fact. Leaked internal research from Meta, published through whistleblower disclosures and subsequent congressional testimony, showed that the company's own researchers found Instagram worsened body image for a substantial portion of teenage girls who used it. The research was conducted, the findings were reviewed by executives, and the product continued with no material changes for years afterward. The gap between what these companies knew and what they disclosed publicly is the story.

The Internal Findings

Meta's internal research, portions of which became public through the Wall Street Journal's 2021 reporting, showed that 32 percent of teenage girls surveyed said Instagram made them feel worse about their bodies when they already felt bad about them. A separate internal slide deck found that among teens who reported suicidal thoughts, 13 percent of British users and 6 percent of American users traced the ideation to Instagram. These numbers came from Meta's own researchers, using Meta's own data, conducted under Meta's own methodological oversight. The company's public position during this period was that the research on social media and teen mental health was mixed and inconclusive. The internal documents suggest that position was not consistent with what the research team was telling leadership.

What Happened in Congress

The 2023 and 2024 Senate hearings on social media and teen safety produced some of the most striking moments in recent congressional history not because of what senators said, but because of what executives did not deny. Under oath and under sustained questioning, representatives from Meta, TikTok, Snap, and YouTube did not dispute that their platforms use engagement-maximizing algorithms that treat time-on-platform as the primary optimization target. They did not credibly dispute that these algorithms serve content based on emotional response rather than stated preference. What they disputed was the causal chain — whether algorithmic amplification of body image content, self-harm content, and eating disorder communities directly produced harm in users. That is a genuinely difficult causal question. Randomized controlled trials assigning teenagers to social media conditions are ethically and practically impossible at scale. The research that exists is correlational, and correlation in a domain as complex as adolescent mental health is always open to alternative interpretation.

The Algorithm Is the Product

Here is the thing that internal disclosures have made clearer than academic research alone could: the recommendation algorithm is not a neutral feature that surfaces content users want. It is a system explicitly tuned to maximize engagement, and engagement is highest with content that produces strong emotional arousal — which tends to mean content that triggers anxiety, comparison, outrage, or desire. Teenagers did not choose to spend hours in rabbit holes of idealized body images or self-harm communities. The algorithm routed them there because that routing produced more time-on-platform. A research team at New York University's Center for Social Media and Politics studied algorithmic amplification across platforms and found that recommendation systems routinely surfaced extremist and emotionally destabilizing content to users who had not searched for it, within sessions that began with entirely benign search terms. The same mechanics apply to body image content, eating disorder communities, and pro-self-harm content — these are not niche interests users arrive at independently. Many are algorithmically recommended destinations.

Why Disclosure Matters

The significance of the internal research is not only that harm occurred. Harm from consumer products occurs. The significance is what it reveals about the relationship between knowledge and action inside these companies. When a pharmaceutical company discovers safety signals in internal research and withholds them from regulators, that is fraud. The legal framework for social media companies is different — Section 230 provides broad immunity for platform content — but the moral logic is the same. The policy conversation has shifted accordingly. Proposals now circulating in state legislatures and in Congress include algorithmic audit requirements, age verification mandates, and restrictions on recommendation systems for users under 16. None of these are without tradeoffs. All of them are more targeted than the internal research initially prompted anyone to imagine would be necessary.