Invest1 distinct publisher3 min readPublished
An internal document put teen adoption of Instagram's Take a Break at 1.8%. Mosseri's blog post advertised more than 90% retention. Those two numbers are the states' case now.
The Investor · Invest desk
Compiled by The InvestorSomething wrong?How this is made
A design case asks a factfinder to decide that ranking software was built badly. That is a fight between expert witnesses and it can be argued to a draw. A disclosure case asks something smaller: the company had a number, and it published a different one. The two Take a Break figures do not contradict each other as arithmetic. They contradict each other as communication, because the retention rate describes only the teenagers already inside the 1.8% [1][2].
Multiply the two and you get the number neither side put on a slide. About 1.6% of teenagers, roughly one in sixty, both switched the feature on and kept it on [1]. Meta released the percentage with the flattering denominator and not the one that showed how few teenagers the pop-up ever reached [3]. Prosecutors are using exactly that gap to argue Meta advertised favourable safety metrics while withholding adoption [4]. Mosseri's own assessment on the stand was that the pop-ups "helped, but not nearly as much as we hoped" [8].
The harder document is not about a feature at all. Internal communications put to him by Jason Slothouber, a senior prosecutor from the Colorado Attorney General's Office, record that Meta's lawyers advised product designers to limit what data Mosseri would see, to check his litigation exposure [9]. Mosseri said he was not aware of the discussion, and that it made sense for the people assembling the document to want the information in it to be accurate [10]. Slothouber's follow-up was whether someone who works with the numbers every day understands better than a lawyer which numbers should be shown to him [11]. A disclosure claim needs a curated decision-maker. That exchange is the states' attempt to show the curation had counsel behind it.
The money frames how much that theory is worth. A New Mexico judge this month ordered Meta to pay $567 million into a youth mental health fund after the state argued public nuisance [19]. The states in Oakland have signalled they could seek close to $200 billion [17], about 350 times the New Mexico order [2]. Earlier this year a Los Angeles jury found Meta and Google negligent and awarded $6 million to one woman who said she became addicted to Instagram and YouTube as a child [20]. Some legal specialists reach for the tobacco cases of the 1990s as the comparison [21]. The documents doing the work this week suggest the states have drawn the same lesson about paper.
Mosseri has run Instagram since 2018 [22]. The trial began last week and is expected to run about six weeks [15], with a further 25 states queued behind it and separate state court actions including one underway in Tennessee [18].
Ranked by verification strength, evidence, and original report placement.
An internal Meta document admitted into evidence showed that at one point 1.8% of teenagers had Instagram's Take a Break feature switched on, per NPR.
Around the time Take a Break launched, Adam Mosseri wrote in a blog post that more than 90% of the people who turned it on kept it running.
Meta published the 90% retention figure and not the 1.8% adoption figure.
Prosecutors are using the gap between the two figures to argue that Meta emphasized favorable safety metrics without disclosing how limited actual adoption was.
Take a Break sends a pop-up once someone has been in the app for a set stretch of time, suggesting they stop scrolling; until late 2024 it was optional, so a teenager had to go into settings and turn it on.
Meta made Take a Break the default on teen accounts in late 2024, the same year it introduced teen accounts, and Mosseri testified that the question of early uptake was "entirely moot" by then.
Follow any of these and your For You feed starts watching them — no settings page required.
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Courtroom exhibits and sworn testimony, but one publisher relaying two others
The core facts are strong in kind - a document admitted into evidence, direct quotations from sworn testimony by Mosseri and Fogu, and named prosecutors and judge. They are weak in provenance for this cluster: a single crypto/finance outlet relaying Associated Press trial reporting plus an NPR-sourced figure, with no transcript, filing or Meta statement supplied, and no description of the 1.8% document's population or measurement window.
Opt-in safety nudge reached about one teen in fifty before it was defaulted on
Adoption here is adoption of the safety feature itself, and it is quantified rather than inferred: 1.8% of teenagers had Take a Break enabled at one point, with more than 90% retention inside that group, implying roughly 1.6% net. The late-2024 default flip presumably raised reach, but the cluster supplies no post-default figure, so the only measured adoption level is the pre-default one.
Published safety metric overstated real coverage; the damages headline is a demand, not an award
Two overstatements point the same way. Meta's public communication surfaced a 90%-plus retention rate while withholding the 1.8% adoption base, and Mosseri's own testimony concedes the pop-ups helped 'not nearly as much as we hoped' - the published metric implied far broader protection than the evidence supports. Separately, the cluster's most quotable figure, close to $200 billion, is a signalled demand roughly 350 times the largest actual order described, and liability has not been decided. Positive gap, though the underlying courtroom facts themselves are not inflated.
Every voice in the story is a litigant, a witness, or a traffic-seeking relay
Incentives are unusually dense and all documented in the cluster. Meta had a reputational incentive to publish the flattering retention figure and, per the internal communications, a legal incentive to restrict what harm data its product head saw. Mosseri testifies as a central defence witness whose own exposure was the subject of counsel's advice. The Colorado prosecutor is building toward a demand approaching $200 billion across 29 states plus later trials. The reporting outlet is a crypto/finance publisher relaying wire copy, citing its own prior coverage and closing with a newsletter pitch.
Facts credible and quotable; single-publisher chain and missing context cap it
Named courtroom actors, verbatim testimony and admitted exhibits make the specific claims credible, and the internal arithmetic is checkable. But the cluster has one publisher relaying two others, no primary document or Meta response, no scope for the 1.8% measurement, and no post-default adoption data - so the interpretive step from 'low opt-in uptake' to 'Meta misled about safety' remains a litigated contention rather than a settled finding.
product
Instagram's break reminder stayed opt-in because a default cost metrics, jury told1 distinct publisher
leadership
A safety engineer says he briefed Zuckerberg 100 times. The escalation trail is now evidence.1 distinct publisher
product
Meta's under-13 data practices go to a jury: 29 AGs, COPPA, and a porous age gate1 distinct publisher
invest
Meta's Pay Structure, Not Its Policy Page, Is What the States Put on the Stand1 distinct publisher
Distinct publishers with included, body-backed reporting in this cluster.
1 article · August 25, 2026