SB 976 stands: personalized ranking is now its own regulated surface

A federal district court in San Jose has denied Meta, YouTube and TikTok the preliminary injunction they sought against California's SB 976, the Protecting Our Kids from Social Media Addiction Act, according to Techdirt ^{[1] [4] [5]}. The reasoning matters more than the outcome: Judge Edward Davila found the companies had not shown, at this stage of litigation, that their personalized feeds are "expressive," pointing to their reliance on predictive algorithms that use watch history and other data to suggest content the system anticipates will be engaging ^[6].

That draws a line through the middle of a system most teams treat as one thing. In Moody v. NetChoice, the Supreme Court majority treated content moderation as editorial discretion protected by the First Amendment, while a footnote declined to address the status of algorithmically recommended content ^[7]. Justice Barrett's concurrence raised, without answering, what happens when an algorithm simply presents to each user whatever it predicts the user will like, such as content similar to posts the user previously engaged with ^[8]. Techdirt reports that a handful of courts, this one included, have taken that opening as licence to hold that algorithmic recommendations get no First Amendment protection ^[9]. Blocking, removing and demoting sit on the protected side of that line; personalized ranking, for now, does not.

The obligations attached to the unprotected side are product requirements, not policy language. As Techdirt describes the statute, SB 976 defines "addictive feeds" as effectively any algorithmic recommendation system with some level of personalization ^[10], requires sites with such feeds to block them for children absent verifiable parental consent ^[11], and imposes a default limit of one hour per day on children's access ^[12]. The Attorney General decides by what means a site must determine whether a user is a child, which Techdirt argues amounts to a power to mandate intrusive age verification ^[13]. Read together, the consent gate and the default cap mean the compliant state for a user you have not verified is non-personalized and clocked ^[1].

That implies four artifacts a memo cannot produce: a non-personalized feed path that is the default rather than a fallback, a consent record you can produce on demand, an age signal built to a method you may not get to choose ^[13], and a per-day consumption counter wired to feed impressions rather than to session length. The litigation structure reinforces that this is per-architecture work. The court found the trade association NetChoice lacked standing to bring an as-applied challenge, on the basis of Moody's facial-versus-as-applied language, so Meta, YouTube and TikTok brought those challenges as themselves ^{[2] [3] [4]}. Davila's description of the disqualifying design, watch history plus engagement prediction, fits most production ranking stacks ^[6].

Three things to watch. First, what the Attorney General designates as an acceptable means of determining that a user is a child, since that determines whether the age gate is an inference or an identity check ^[13]. Second, whether the "not expressive" finding hardens past the preliminary posture; the phrase "at this stage of litigation" is doing real work in the order ^[6]. Third, the evidentiary fight on the merits: Techdirt calls the law's scientific justification junk, notes that research suggests labelling feeds addictive is itself harmful, and points to repeated unconstitutional rulings against regulating children's access to television and video games ^{[15] [16]}. The law was passed and signed by Governor Gavin Newsom, with an earlier version introduced by state Senator Nancy Skinner ^[14].