One in three: what Meta’s own researchers wrote down

Somebody at Facebook wrote this on a slide in 2019, in the first person plural, as a finding:

We make body image issues worse for one in three teen girls.

Not “some research suggests”. Not “there may be an association”. We make. It is the most damaging sentence any technology company has written about itself this century, and the company wrote it internally, voluntarily, in the course of trying to find out what its product was doing.

The slide surfaced on 14 September 2021, when Georgia Wells, Jeff Horwitz and Deepa Seetharaman of The Wall Street Journal published the first of the Facebook Files. The underlying documents were later submitted to the US House Energy and Commerce Committee, which is why anyone can read them rather than take a reporter’s word for it.

What the documents say

The sentence above is a headline. Beneath it sits a March 2020 research finding stated more carefully: “32% of teen girls said that when they felt bad about their bodies, Instagram made them feel worse.”

The rest of the material is in the same register. Thirteen per cent of British and six per cent of American users who reported suicidal thoughts traced the desire to kill themselves to Instagram. More than 40 per cent of Instagram users who felt “unattractive” said the feeling began on the app. About a quarter of teens who felt “not good enough” attributed it to Instagram. Fourteen per cent of American teen boys said Instagram made them feel worse about themselves, and roughly 40 per cent of teen boys reported experiencing negative social comparison.

Two numbers set the scale. Twenty-two million American teenagers log onto Instagram daily. More than 40 per cent of the platform’s users are 22 or younger.

Meta’s answer deserves more than a sentence

Meta published a rebuttal, “What Our Research Really Says About Teen Well-Being and Instagram”, arguing that the Journal mischaracterised the work. This is a genuinely contested reading of the same documents, and the company’s case is stronger than the coverage of it has been.

Start with what the research is. It is small-sample internal product research, not epidemiology. It was not designed to establish population-level effects, was not powered to, and was never peer-reviewed. Read the way a headline reads it, a deck of qualitative user studies is being asked to carry a claim about tens of millions of adolescents.

Then look at the measure. Every one of the figures above is self-report attribution. Nobody measured a teenager’s body image, exposed them to Instagram, and measured it again. Researchers asked young people who were already struggling where they thought the feeling came from, and recorded the answer. People are famously unreliable narrators of their own causation. A teenager who feels unattractive and spends four hours a day on Instagram will name Instagram, because Instagram is what is in front of them. That is a real finding about how the experience is understood. It is not a measurement of what caused it.

Then read the two sentences against each other. The slide says one in three teen girls. The research says 32 per cent of teen girls said that when they felt bad about their bodies, Instagram made them feel worse. The second is a conditional statement about a specific moment. The first drops the condition, changes “made them feel worse” into “we make worse”, and turns a survey response into an admission of agency. Whoever wrote that slide was compressing for impact, which is what slides do — and the compression is precisely what made the sentence quotable.

Meta’s final point is the one worth sitting with. The company says the same research showed net-positive results on most measured dimensions. If that is right, then what the Journal published was the worst three lines from a body of work that mostly said something else, and a company that runs candid internal research on its own harms has been punished for running it. The incentive that creates is obvious and bad: the safest thing a platform can do is stop asking.

What the rebuttal does not do

It does not dispute that the documents exist. It does not dispute the wording. It does not argue that the researchers were wrong to look, or that the numbers were transcribed incorrectly. The disagreement is about interpretation and weight — about whether a set of small self-report studies supports the headline that was made from it.

So the defensible claim is narrower than the famous one, and it is still substantial. Facebook’s own researchers, using the company’s own users, found that a meaningful minority of teenage girls experienced Instagram as making body-image distress worse, and that a smaller but non-trivial group traced suicidal ideation to it. The company’s internal summary of that work was that it made things worse for one in three. Nobody outside Meta wrote that sentence. Meta’s argument is that its own researchers overstated it, which is an argument the company is entitled to make and which readers can weigh, because the documents are public.

What you cannot honestly do is either of the two things everyone does. You cannot cite “one in three teen girls” as an established scientific fact about adolescents. And you cannot wave the whole thing away as a media beat-up, because the company’s researchers were the ones in the room.

The first verdict, and what it is worth

The documents became litigation. On 24 October 2023 a bipartisan coalition sued Meta — 33 attorneys general in federal court in the Northern District of California, plus eight more in state courts, 41 in total — alleging violations of the Children’s Online Privacy Protection Act and, in California’s case, the state False Advertising Law and Unfair Competition Law. The complaints allege that Meta designed psychologically manipulative engagement features while misleading the public about safety. California’s Rob Bonta: “Meta has been harming our children and teens, cultivating addiction to boost corporate profits.”

Then, on 25 March 2026, the first such case reached a verdict. A jury in Los Angeles Superior Court found Meta and Google/YouTube negligent and awarded $6 million to a 20-year-old plaintiff from Chico, California — $3 million compensatory and $3 million punitive, with Meta assigned 70 per cent responsibility. Snapchat and TikTok settled before trial.

The rhetoric afterwards was about scale. Co-lead plaintiff counsel Joseph VanZardt: “Today’s verdict is a referendum — from a jury, to an entire industry — that accountability has arrived.” Lead trial lawyer Mark Lanier: “How do you make a child never put down the phone? That’s called the engineering of addiction.” A juror identified as Victoria: “We wanted them to feel it. We wanted them to realize this was unacceptable.”

Now the qualifications, which are not decoration. This is a first-instance jury verdict subject to appeal. It is not settled law. It establishes no precedent binding on any other court. It is twelve people in one Californian courtroom deciding one case about one plaintiff on the evidence they were shown, and the defendants have every ordinary avenue to challenge it. Six million dollars is, to Meta, a rounding error; the signal is the finding of negligence, not the sum.

Meta and Google argued at trial that teen mental health is multi-causal and cannot be attributed to a single app. That argument happens to be correct. It is also, in a negligence case, not a complete defence — the question is contribution, not sole cause. The jury apportioned it: 70 per cent to Meta. Whether an appellate court agrees is genuinely unknown, and anyone telling you otherwise in 2026 is selling something.

What survives all of it is the slide. Before any attorney general filed, before any jury was empanelled, a researcher inside the company wrote down what they had found and used the word we. Everything since has been an argument about how much that sentence is worth. It is not worth nothing.