They knew the like counts were hurting kids. They built the fix. They did not ship it.
This is the most complete sequence in the record — research, design, delay, dilution, and a public credit-claim taken for the fix that barely shipped. Each step is in a document. The documents are not in tension with each other. They are in sequence.
What follows is the documented order of events. We are not characterizing intent. The record shows a problem identified, a remedy researched, a remedy built, a delay, a weaker version released behind a setting most users never find, and that remedy’s existence later cited publicly. A reader can draw their own conclusion from that sequence. We are not going to draw it for them, because a plaintiff’s firm that tells a parent what to conclude about intent is a plaintiff’s firm that has overclaimed.
Quantified social comparison, and how many teenagers it reaches.
Visible like counts turn approval into a visible number. For a developing teenager, that converts every post into a comparison. The scale of the exposure is the part that is quantified.
A retained expert in this matter ties engagement quantification directly to harm, and puts a number on the affected population: approximately four to five million teens experience negative social comparison weekly. That figure is an expert estimate about a population, not a measurement of any individual child, and it should not be read as one.
Rebuttal trial report of Ravi Iyer, Ph.D., Dkt. 3168-10
The mechanism is not mysterious. A feed that displays counts rewards the behavior that generates counts. Adolescent psychologists describe appearance-focused, quantified feeds as a risk factor for disordered eating and self-directed harm. How much of that risk converts into injury for a particular teenager, and how much is attributable to other factors, is the contested part — and it is contested by serious researchers, not only by defense experts.
Project Daisy was researched, and it worked.
The remedy was not a suggestion from a critic. It was an internal product experiment run by the company, testing the removal of public like counts against harmful social comparison.
Project Daisy was an internal product experiment investigating the effects of engagement quantification on harmful social comparison, using A/B testing to measure whether removing public like counts reduced it. The premise — that the visible count was doing measurable psychological damage — was not contested inside the company. It was the basis of the experiment.
Rebuttal trial report of Ravi Iyer, Ph.D., Dkt. 3168-10
An internal well-being exhibit records that the team knew what it was testing: “We knew this since we first announced the test.”
Well-Being, JCCP trial exhibit 01495
Three years from first discussion to launch.
The remedy was discussed internally in December 2017. In 2019 the company’s head of Instagram made a public commitment about decisions that hurt the business. In May 2021, an opt-in version of the fix launched. A comparable feature discussed in 2017 shipped in January 2023 — more than five years later.
Hiding like counts first raised internally
The remedy enters the internal discussion record.
Alter trial report, Dkt. 3246-12
A public commitment about health over revenue
The head of Instagram states publicly that the company will make decisions that hurt the business if they are good for people’s well-being and health.
State AGs’ Fourth Supplemental Response, Dkt. 3024-1
An opt-in version ships — three and a half years later
Users can turn off like counts. Users must turn them off.
Alter trial report, Dkt. 3246-12
A related feature ships after five years
A comparable notification-control feature, also discussed in 2017, launches in 2023.
Alter trial report, Dkt. 3246-12
Three and a half years is a long time to ship a fix you already validated. It is also the kind of interval that a fact-finder hears, and the record does not offer a competing explanation for it in these documents.
The version that shipped was not the version under consideration.
This is the point most often missing. The delay is not the whole story. What eventually launched was put beside the version that had been considered for youth safety, in deposition.
Deposition questioning of a company executive on Project Daisy juxtaposed the version that launched against the version that was under consideration for youth safety. The questioning was about that difference.
Deposition of Vaishnavi Jayakumar, Dkt. 3246-16
The design choice that follows from this is the important one: an opt-in setting is not a fix for a default. If the harmful configuration is what a user gets without doing something, then the fix requires the user to do something — and most users do not. Teen accounts in particular still see engagement quantification by default.
The remedy was used by a fraction of a percent.
Internal funnel figures and an expert estimate give the same picture from two directions.
Tab enabled
An internal strategy document sets out the top of the funnel: the count that reached the stage where the control was available in-app.
IG Mental Well-Being: Multi-Half Strategy, Dkt. 3029-10
Tab displayed
The count that actually saw the control offered to them.
Dkt. 3029-10
Tab honors
The count that engaged the control. Roughly a third of those who saw it used it; a small fraction of those who could have reached it.
Dkt. 3029-10
Estimated use by teen users
An expert estimate of the share of teen users running the hide-likes setting across the deployed population — a range, and a wide one.
Rebuttal trial report of Ravi Iyer, Ph.D., Dkt. 3168-3
These are internal funnel counts from a specific program and period, and an expert estimate across a population. They are not measurements of the same thing and should not be added together or presented as a single rate. We show both because they point the same direction from independent sources.
The existence of the fix was then cited as evidence of good faith.
A voluntary feature that most users did not enable was used in public messaging to demonstrate that the company had acted. The harmful default was unchanged throughout.
An expert describes the company using its proposal to hide like counts — known internally as Project Daisy — to enhance the credibility of its public messaging.
Trial report of Adam L. Alter, Dkt. 3234-1
Whether the public use of that fact was accurate is a fair question, and it is a question the documents let a reader ask. What can be said from this record is that the remedy existed, it was opt-in, uptake was very low, and its existence was cited favorably. Those four statements are all sourced, and none of them require us to say anything about why.
What this brief does not show.
It does not show that a like count caused a specific child’s injury. Counting an approval is not itself a medical event. Individual causation runs through a medical chronology, a behavioral record, and a differential analysis of what else could explain the same outcome — and it has to survive a defense that will argue the prognosis was unchanged. This brief establishes the feature, the knowledge, the fix, the delay, and the uptake. Causation is a different question and belongs to a different proof.
It does not show intent. Sequence is not the same as motive, and we are not going to convert a documented timeline into an assertion about what anyone meant. We are showing the order. The order is in the documents.
The population figure is not an individual finding. Roughly four to five million teens experiencing negative social comparison weekly is an expert estimate about a group. It does not establish that any particular teenager in that group was harmed by any particular feature.
This is a summary of a public record for general information. It is not legal advice and it does not create an attorney-client relationship.
Defaults, and what a family can actually do.
The practical lesson from this record is narrower and more useful than the legal lesson: an opt-in remedy for a harmful default is not a remedy. The default is what almost every child experiences.
This is why the same features — visible counts, engagement-ranked feeds, appearance filtering — appear across multiple matters and multiple practice areas. Not because the injuries are the same, but because the design choices are shared, and the choices are made once, centrally, for everyone.
For a family, the actionable steps are concrete and do not depend on how any of the above is resolved: preserve the device and the platform data before anything changes, and preserve the medical chronology that establishes when the change began.