It starts with something normal, innocuous: a meal prepping video. The next time you’re absent-mindedly scrolling, your feed shows more cooking content. Maybe you pause on one about low-calorie swaps, just a half-second longer than the others. The algorithm notes the pause. Now there are more of those. You click on a video about cutting carbs. Then macros. Then a thread about net calories. You didn’t search for any of it. You didn’t decide this is who you are. But the feed has decided something about you, and it is walking you somewhere, until everything you see pushes in one direction: thin. Thin. Thin. That is what detractors say major social media platforms are doing, pointing to an algorithm that does not care whether it is pushing body dysmorphia, outrage, or extremism, only that it keeps pushing. For teenagers especially, that push carries a cost: plaintiffs across the country have catalogued its destinations, from disordered eating and self-harm to radicalization. Nobody wakes up an extremist. Nobody chooses a rabbit hole. They follow one, step by careful step, each nudge calibrated by software that allegedly knows where the path leads. The recommendation algorithm does not aim to inform or connect. Instead, it maximizes the time you spend on the platform, because time on platform is what platforms sell to advertisers. And the data is unambiguous: emotionally intense, extreme content keeps people watching longer than moderate content. So the algorithm amplifies. Click on something slightly right of center and the feed migrates right; slightly left, and it goes left, further and further in whichever direction you’ve been nudged. A passing curiosity about a fringe idea is followed by a community built around it. You did not choose that destination. You were walked there, one small engagement at a time. Thirty-six percent of U.S. teens report being on at least one social media platform “almost constantly.”1 The platforms, the evidence now shows, were designed to produce exactly that result. In re Social Media Adolescent Addiction/Personal Injury Products Liability Litigation,2 the mechanism is identified as Intermittent Variable Rewards (IVR), the same dopamine loop that makes slot machines addictive. A child experiencing depression who watches one video about suicide may find herself receiving more suicide-related videos. Instagram, according to internal documents produced in the MDL record, may hold likes on a post until it can deliver them at maximum emotional impact. Complaints against TikTok in the same MDL allege the same IVR architecture: unpredictable content delivery timed to override the user’s intention to stop. The result is the pattern one mass shooter described in his own manifesto, “I’m a literal addict to my phone. I can’t stop consuming.”3 Ohio’s legislature has noticed. In 1985, Tipper Gore co-founded the Parents Music Resource Center and pressed Congress to require warning labels on music albums with explicit lyrics. The effort produced the now-familiar “Parental Advisory” sticker, adopted voluntarily by the recording industry under congressional pressure. The music industry’s objections were familiar: warning labels are government-compelled speech, vague, and chilling to expression. Ohio’s HB 808, introduced in April 2026, draws on the same playbook. The bill would require pop-up addiction warnings on apps featuring “addictive” design elements (autoplay, push notifications, algorithmic amplification) with $5,000-per-violation fines. The platforms’ First Amendment objection is the same one the record labels made in 1985: you cannot force us to say things we don’t want to say. Whether they can or not depends on what kind of speech the warning label is. The First Amendment bars government-compelled speech, but courts have long recognized an exception for required factual disclosures about a business’s own product. That is why nutrition labels and cigarette tar disclosures survive First Amendment challenges.4 The platforms will argue that labeling their features “addictive” is not a neutral factual disclosure but a contested conclusion the government is forcing them to endorse. If a court agrees, HB 808 faces a much higher constitutional bar, and one Colorado’s nearly identical law has already failed to clear. NetChoice (whose members include Meta, Google, TikTok, and X) has already obtained a preliminary injunction against Colorado’s law on exactly that theory. HB 808’s fate may depend less on the legislature than on the courts, and the courts have been busy. Plaintiffs across the country are suing the platforms not for what their users posted, but for how the platforms were built. Plaintiffs advance two product liability theories: defective design (an algorithm engineered to maximize engagement at the cost of user safety is a flawed product) and failure to warn (the platforms had internal data documenting the harm and disclosed none of it to users or parents). The platforms have one primary defense. Section 230 of the Communications Decency Act, 47 U.S.C. § 230, enacted in 1996, gives platforms a powerful shield: they cannot be treated as the publisher or speaker of content their users post. For decades, platforms have argued that this immunity covers not just what users say, but how the algorithm decides what you see next. Two cases illustrate how that argument is beginning to crack. In 2015, Dylann Roof walked into Mother Emanuel AME Church in Charleston and murdered nine people, including Reverend Clementa Pinckney. Rev. Pinckney’s daughter later sued Facebook, alleging that the algorithm had spent years feeding Roof content that “nurtured, encouraged, and ultimately served to solidify and affirm” his racist views, recommending extremist groups to him and rewarding divisive posts with likes and shares. The Fourth Circuit affirmed the dismissal of the case. Facebook’s decisions about how to display content are traditional editorial functions, the court held, and Section 230 protects them.5 One judge dissented. When Facebook’s own “Groups You Should Join” feature recommended that Roof join an extremist group, the dissenter wrote, Facebook was not publishing a user’s content. It was speaking for itself. That distinction, the dissenter argued, Section 230 was never meant to erase. Seven years after the Roof shooting, in Buffalo, a teenager drove two hundred miles to a grocery store and killed ten people. He had spent years consuming content the platforms served him, and he knew what it had done to him. His manifesto said so plainly: “[W]hy do I always have trouble putting my phone down at night?”6 New York’s Appellate Division affirmed the dismissal on Section 230 grounds. Two dissenters disagreed, and their reasoning points toward where the law may be headed. The plaintiffs were not challenging anything a user had posted. They were challenging autoplay, infinite scroll, push notifications sent at midnight, and design features built to prevent users from leaving. Those are engineering decisions, not editorial ones. A company is liable for a defective product regardless of what its customers do with it, and Section 230 was written to protect publishers, not product designers. The central question in every algorithm case is deceptively simple: when a platform’s algorithm selects content and delivers it to you, is the platform publishing someone else’s speech, or speaking for itself? Section 230 immunizes the former, not the latter. Nylah Anderson was ten years old when TikTok’s For You Page pushed the “Blackout Challenge” to her feed; content she never searched for, and that TikTok had been warned was killing children. She died. The Third Circuit held Section 230 offered no protection: when the algorithm selects and delivers content to a specific user, the platform is the actor, not a neutral conduit.7 In Myanmar, Facebook’s engagement-maximizing algorithm amplified anti-Rohingya hate speech with such efficiency that, allegedly, it contributed to the genocide. The Ninth Circuit affirmed dismissal, bound by its own prior precedent classifying algorithmic curation as a protected editorial function.8 But two judges, in a concurrence, refused to endorse the majority’s reasoning. Judge Berzon wrote that algorithmic recommendations are the platform’s own speech, not republished content, and Section 230 was never meant to cover them. Judge R. Nelson agreed that courts have stretched “publisher” so far beyond its ordinary meaning that it no longer resembles what Congress intended in 1996. Both called for the full Ninth Circuit to reconsider. The conflict between the Third and Ninth Circuits will have to be resolved. Tipper Gore wanted a sticker on a Prince album. Ohio’s legislature wants a pop-up on TikTok. The underlying anxiety is the same: allegations that a communications medium can harm its users; specifically minors. What is different this time is that the medium’s designers left a paper trail. Internal research, congressional testimony, and documents produced in MDL 3047 have given plaintiffs a detailed record of what the platforms knew about the amplification spiral and when they knew it. Whether the algorithm is a neutral conduit or a defective product is the question courts across the country are being asked to answer. The platforms say they are publishers protected by the First Amendment and Section 230. Plaintiffs say they are manufacturers who built something they knew was harmful and sold it to children anyway. Ohio sits at an intersection of every competing legal theory this litigation has raised. The next opinion that moves the line could come from Ohio.
Jenna Schneider is Special Counsel at Watstein Terepka LLP, where she represents technology, fintech, and consumer-facing companies. A former broadcast journalist turned litigator, she holds a CIPP/US certification and has more than a decade of experience litigating cases involving data practices, website tracking technologies, and consumer protection statutes. She is the founder of MediaMind Midwest, a nonprofit that teaches high school students to assess credibility and bias in today’s media landscape, and is admitted in Ohio, Georgia, and Texas.