Zlonk! What a Los Angeles Jury Found That Fifty Years of Research Could Not – Part 3

This is Part 3 of a three-part series. Part 1 examined the five-century history of media moral panics and why the smartphone debate fits that template imperfectly. Part 2 traced the engineering logic behind algorithmic feeds back to the television scheduling techniques of the 1970s and 1980s. 

On 25 March 2026, a Los Angeles Superior Court jury found Meta and Google legally liable for the depression, body dysmorphia, and suicidal ideation suffered by a plaintiff who had begun using YouTube at age six and Instagram at age nine. The jury ordered the companies to pay six million dollars in compensatory and punitive damages. Both companies announced plans to appeal.

That verdict does not resolve the scientific debate this series has tracked across two posts (see part I and part II). The average effect of social media use on adolescent well-being remains small in the population data. The direction of causation remains genuinely unclear. What the verdict does is establish something the research literature cannot: that the engineering intent was deliberate, documented, and directed at children, and that a body of twelve ordinary citizens found the harm to one of them legally attributable to that intent. What the science rates as ambiguous, the law has begun to price in a specific case.

What the science actually shows

The empirical record on social media and adolescent mental health is a picture with a clear foreground and a contested background.

The foreground is the null or near-null finding on average effects. As mentioned in Part I, Amy Orben and Andrew Przybylski at Oxford, in their 2019 Nature Human Behaviour paper analysing approximately 355,000 participants across three large datasets, found that digital technology use explained roughly 0.4% of the variation in adolescent well-being. They described the effect as comparable to eating potatoes. Candice Odgers, in her April 2024 Nature review, wrote that hundreds of researchers have searched for the large effects suggested by Jonathan Haidt and found instead a mix of no, small, and mixed associations.

The contested background is more complicated, in part because of a consistent measurement problem. Screen time is a blunt instrument that aggregates passive scrolling, active messaging, video calling, and creative production into a single variable. If the harm is concentrated in specific features, for example, algorithmic short-video feeds served to young adolescent girls, a measure of total screen time will wash out that signal.

What the platforms knew internally

The gap between the precision of the science and the precision of the platforms’ internal knowledge is where the legal argument lives.

Internal Meta documents shown to the jury in the Los Angeles trial included a memo stating that if the company wanted to win big with teens, it needed to bring them in as tweens, and data showing that 11-year-olds were four times as likely to return to Instagram compared with competing apps, despite the platform’s own minimum age of 13. A New Mexico court that same week ordered Meta to pay 375 million dollars after finding the company misled users about platform safety.

The insider testimony predates the trial. In November 2017, Facebook’s founding president Sean Parker told Axios the platform was designed around a social-validation feedback loop that exploited a vulnerability in human psychology. Weeks later, Chamath Palihapitiya, Facebook’s former VP of user growth, told a Stanford Business School audience that the short-term, dopamine-driven feedback loops the company created are destroying how society works, adding that he does not let his own children use the platform. Tristan Harris, a former Google design ethicist, formalised the argument through the Center for Humane Technology and the 2020 documentary The Social Dilemma, contending that persuasive design deliberately exploits psychological vulnerabilities at scale.

None of this constitutes scientific proof of population-level harm. What it establishes, in a way that no predecessor media panic ever could, is that the engineering intent was documented, the awareness of potential harm was internal and contemporaneous, and the commercial decision was made to proceed. Fredric Wertham fabricated his evidence against comic books. The evidence against Meta and Google was written by Meta and Google.

The science-to-policy translation problem, again

The pattern by which ambiguous research becomes confident policy is not new to this debate. Part 2 of this series described the 1972 Surgeon General’s report on television violence, which spent three years and 23 commissioned studies reaching a context-dependent, heavily qualified conclusion, and was then translated by the press and politicians into a definitive causal link. The same translation problem applies today, in both directions.

The table above is not an argument that social media and television violence are equivalent concerns. It is an argument that the institutional process by which scientific ambiguity becomes policy certainty is consistent across both cases. The 1972 report did not prove that television caused violence. It proved that the relationship was real in some populations, mediated by many factors, and exceedingly difficult to isolate. Current social media research is in the same position. Policy is being made as though it is not.

What the series has argued

Three posts. One argument in three parts.

Part 1 established that the panic structure surrounding social media and children follows a template that is more than five centuries old. The magnitude claims will be overstated. The rewiring metaphor will not be supported by neuroscience. The moral entrepreneurs making the loudest arguments are not reliable narrators. All of that is consistent with the historical pattern.

Part 2 established that the engineering logic behind algorithmic feeds is directly inherited from broadcast television. Hotswitching, stripping, bridging, and hammocking are the ancestors of autoplay, push notifications, infinite scroll, and sponsored insertion. The intent to maximise retained attention for advertising revenue is identical across both eras. What changed is the precision of the instrument: from population-level scheduling to individual-level personalisation using tens of millions of data points per user.

Part 3 adds the element that neither the panic framework nor the historical comparison fully accounts for: the legal record. The research literature rates the causal claim as ambiguous. The internal documents rate it as settled. A jury in Los Angeles has now found, on the basis of those documents and the testimony of the platforms’ own former architects, that deliberate design caused documented harm to a specific child. That finding does not resolve the population-level scientific question. It answers a different question: whether the people building these systems knew what they were building and chose to build it anyway.

The verdict that fits neither headline

The historical comparison the series opened with cuts both ways. Five centuries of moral panics suggest the specific content fears will age badly, and they probably will here too. But every previous panic lacked something this one has: a contemporaneous internal record, written by the people building the system, documenting what they understood it to do to the people using it.

That’s not a scientific finding. It’s a different kind of evidence: the kind courts use.

Written by Caroline De Cock, LL.M., Head of Research