Case Facts
The order comes out of the multidistrict litigation captioned In re: Social Media Adolescent Addiction/Personal Injury Products Liability Litigation, pending in the U.S. District Court for the Northern District of California as Case No. 4:22-md-03047-YGR (MDL No. 3047). School district plaintiffs allege that Meta, Google, ByteDance, and Snapchat designed Facebook, Instagram, YouTube, and TikTok to foster compulsive and problematic use among students, forcing the districts to absorb the resulting disruption. At the motion-to-dismiss stage, the court had already allowed the districts to proceed on a “core theory of injury [which] focuses on the impact of compulsive use itself, irrespective of third-party content, defendants’ protected publishing activity and defendants’ protected first-party speech.” By summary judgment, the districts had submitted expert reports to support that theory, and the defendants moved under Federal Rule of Evidence 702 to exclude the opinions of six school-district-specific experts (the “SD Experts”) in their entirety. Judge Yvonne Gonzalez Rogers’s February 17, 2026 order, Document 2750, resolves only that motion - two related motions aimed at the plaintiffs’ general-causation experts remained pending - given its bearing on the first bellwether trial, involving the Breathitt school district, scheduled for June. The order followed oral argument held January 26, 2026.
The Setup
One of the six challenged experts, Brian Osborne, is not a data scientist or a mental-health professional but a career school administrator. He holds an Ed.D. and an M.A., both from Harvard, in administration, planning and social policy; has more than 25 years of experience in public education, including as a teacher and superintendent of schools in New York and New Jersey; and currently consults with educators across New Jersey and Pennsylvania, spending 40 to 50 days a year observing classroom instruction. His assignment was to offer what his report calls an “experience-based account of what is happening in schools,” drawn from that career and his ongoing classroom work. The report also cited academic literature to support that account. At his deposition, according to the defendants, Osborne admitted that some of the articles his report purported to rely on do not exist, and that other citations in the report were miscites the defendants attributed to artificial intelligence.
The Challenge
Defendants moved to exclude Osborne’s opinions in their entirety over the citation problem. Plaintiffs offered a narrower account: none of the underlying academic articles Osborne relied on were themselves false, they said - the citations had simply been formatted incorrectly because an AI citation tool had been used to generate them, and the formatting had since been corrected. The court accepted that account. “The Court declines to exclude Osborne’s opinions on this basis,” the order states. “Defendants may, however, raise the issue on cross-examination” - leaving the AI-citation problem to the jury’s assessment of weight and credibility rather than treating it as a threshold bar to admissibility.
The Ruling
Judge Gonzalez Rogers denied the Rule 702 motion as to all six SD Experts, not only Osborne. Applying Rule 702’s four-part standard - that expert testimony help the jury, rest on sufficient facts or data, follow from reliable principles and methods, and reliably apply those principles to the case - the court walked through and rejected each of the defendants’ separate challenges to Osborne, including that his opinions fail to isolate specific “actionable” platform features, that his reliance on hearsay-derived information from teachers and administrators made his methodology unreliable, that he was unqualified to address adolescent mental health or platform design, and that his description of public education as a “public right” was an impermissible legal conclusion. On each of those points, apart from the AI-citation issue, the court found the objections went to weight rather than admissibility.
But the ruling was not a clean pass for Osborne’s report. In a section addressed to his eventual trial testimony, the court flagged, as guidance for what would and would not be permitted, several categories of statements it said would be excluded if offered as written: his repeated description of education as a “public right,” which the court found “appears to conflict with controlling precedent” that education is not a fundamental right protected by the Constitution; passages the court said “attempt to advance a narrative rather than offer expertise to a jury,” including one asserting that social media “functions not to help educate all children but instead to profit from exploiting students’ attention, disrupting the operations and educational mission of school districts and their schools in the process”; portions where Osborne simply restates deposition transcripts or outside studies rather than offering his own analysis; and a passage asserting that students “fear being recorded, mocked, or turned into a meme” - a claim about student psychology the court found he is not qualified, as a non-psychologist, to make.
The Kicker
The single paragraph of an eighteen-page order devoted to Osborne’s AI-generated citations produced no exclusion at all. It was the parts of his report with nothing to do with artificial intelligence - his forays into legal conclusions, narrative argument, and the inner lives of students - that drew the court’s sharpest warnings about what could not reach a jury.
What Was Alleged vs. What the Court Found
→ Alleged: Osborne’s report cited academic articles that do not exist, plus additional miscites the defendants attributed to AI, admitted at his deposition.
→ Response: the underlying articles were real; only the citation formatting, produced by an AI citation tool, was wrong, and it was corrected before the motion was decided.
→ Finding: the AI-citation issue does not warrant exclusion; defendants may raise it on cross-examination at trial.
→ Separately flagged: portions of Osborne’s report unrelated to the AI issue - legal conclusions, narrative argument, restated hearsay, and opinions on student psychology outside his expertise - will be excluded if offered as written.
The Lesson
The order draws a line between a citation-formatting error traced to an AI tool and the kind of testimony a court will actually keep from a jury. Once the underlying sources checked out, the AI-generated formatting problem became material for cross-examination rather than grounds to strike an expert’s opinions - while the court reserved its real scrutiny for testimony that reached beyond Osborne’s stated expertise in school administration, regardless of how it was produced.
Takeaways
If you’re retaining experts:
Confirm how any citation-management or research tool was used well before deposition, and be ready to show the difference between a formatting error and a fabricated source - the two were not treated the same way in this ruling.
If you’re an expert:
The court’s sharpest guidance here concerned restating outside studies and deposition testimony as if they were the expert’s own findings, and venturing opinions outside a stated field - not the citation errors themselves.
If you’re opposing counsel:
A citation problem that turns out to be a formatting issue rather than fabrication may not be enough on its own to exclude an expert; this ruling turned as much on the report’s other vulnerabilities as on the AI issue.
