Google, Meta, And TikTok File Three Separate Lawsuits Against California’s Ridiculous ‘Addictive Feeds’ Law

California legislators, led mainly by two of the most anti-science, anti-kid legislators in the state — Josh Lowenthal and Buffy Wicks — make sure that every year they experiment on our children with unproven, untested, scientifically illiterate laws that seek to “protect the children.” And, every year Gavin Newsom eagerly signs them into law forcing the state to waste a bunch of money until the courts strike them down. Of course, Wicks and Lowenthal will insist that they’re protecting the children from the scourge of the internet, but they keep buying into a debunked moral panic and seem positively allergic to what the science actually shows.

The latest bill that California passed (authored by Lowenthal with most of the rest of California’s legislators signing on as co-authors, including Wicks) is AB 1709 which declares (with zero evidence) that certain features (autoplay! recommendation feeds!) are “addictive” and requiring social media companies to (1) age verify everyone in order to make sure that (2) children never experience those dastardly features. Lowenthal and Wicks will not stop until they make sure that everyone’s data is at risk. For the children! As EFF noted earlier this year, the bill is an absolute disaster, defining “addictive features” so broadly that basically any recommendation algorithm counts. This is despite evidence showing that removing algorithmic recommendations can mean people see a lot more problematic content, especially for users who aren’t seeking such content, because those algorithms are a big part of how the bad stuff gets filtered out in the first place. Continuing to insist that algorithmic feeds are inherently bad for you is basically the equivalent of being an anti-vaxxer, but for the internet.

Even if you don’t personally like algorithmic feeds or autoplay, it’s not the government’s job to tell websites how to present content.

The law is set to go into effect on January 1st, meaning that if it’s to be blocked, lawsuits needed to be filed… about now. Unfortunately some similar laws were challenged in the Eleventh Circuit by a three judge panel who basically ignored half a dozen precedents to get to a ruling driven by extremely motivated reasoning. It would be nice if we could get a circuit split from the Ninth Circuit.

The three major social media operators impacted by this law — Google, Meta, and TikTok — each filed individual lawsuits seeking to have it blocked in federal court. If you’re wondering why they all filed their own lawsuits, rather than having a trade association like NetChoice or CCIA filing it for them, you can blame the Ninth Circuit, which last year, in a ruling on SB 976 (California’s earlier ‘addictive feeds’ law), held that trade groups lack associational standing to challenge feed restrictions like these, because whether a feed is “expressive” supposedly requires platform-by-platform proof (despite courts allowing such challenges in plenty of other cases).

Also, because every news article I’ve looked at refuses to link you to the actual lawsuits (journalistic malpractice) you can find them here: Meta’s was filed in the Northern District of California (San Francisco/Oakland), Google’s was also filed in the Northern District (San Jose), and TikTok’s was filed in the Central District of California (which, somewhat confusingly, covers Los Angeles, even though most people think of LA as “Southern California”). All three complaints are embedded below as well (no, I don’t understand why other news orgs don’t do this).

The three cases (all filed by different, but very competent, big-name law firms) take slightly different approaches and emphasize slightly different things, but all basically make a combination of First Amendment and Section 230 arguments. All three present themselves as “as applied” challenges, because the Supreme Court’s ruling in Moody turned a facial challenge, one that says “this whole law is just blatantly unconstitutional,” into a brutally steep climb (the Ninth Circuit tossed most of NetChoice’s facial challenge to SB 976 on exactly those grounds). So each complaint zeroes in on how specific requirements, like the feed and autoplay rules, violate the First Amendment, Section 230, or both.

While it’s annoying that there need to be multiple different lawsuits over this one bad law, it’s at least intellectually fascinating to see the different approaches each suit takes, what they emphasize, and what they play down. They all basically make the argument that curating content is protected First Amendment activity that the state doesn’t get to regulate. They point out that recommendations are generally a good thing (they surface better content for users), and nobody thinks they’re sinister when a bookstore or a newspaper does it. Here’s the Google/YouTube version of the argument:

AB 1709 regulates YouTube as a “publisher.” A law “treat[s]” a website “as” a “publisher” when it regulates the choices that make up publication—what third-party content is presented, to whom, in what order, when, and how. Barnes, 570 F.3d at 1105. A duty directed at those “traditional editorial functions” necessarily depends on the entity’s “status or conduct as a ‘publisher’” and thus implicates Section 230. Id. at 1102. AB 1709 seeks to restrict (1) which videos YouTube may show to its users (the personalization ban) and (2) when and how such videos may be shown to its users (the autoplay ban). Both bans flunk the “simple” “test” for whether a statute violates Section 230: They “prohibit publishing conduct” on their face. In re Apple Inc. App Store Simulated Casino-Style Games Litig. (In re Apple App Store II), 806 F. Supp. 3d 1015, 1030 (N.D. Cal. 2025).

The personalization ban imposes liability when YouTube chooses which third-party videos each user sees. AB 1709 prohibits YouTube from “recommend[ing], select[ing], or prioritiz[ing]” videos based on any “information provided by the user, or otherwise associated with the user or the user’s device.” Cal. Bus. & Prof. Code §§ 22682(b), 22683(a)(1). Whether YouTube is liable or not, then, turns on what content it chooses for its users, and how it chooses that content.

Making that choice is what a publisher does. See Doe 1, 174 F.4th at 1167 (“[M]atching users with content is publishing conduct.”); accord Force v. Facebook, Inc., 934 F.3d 53, 67 (2d Cir. 2019) (websites perform “the functions of publishers” when they “present users with targeted content,” even if via “algorithms”). But that is the very thing AB 1709 forbids.

And just to cut off the stupid argument that keeps coming up from woefully under-informed people, no, claiming to be a “publisher” doesn’t mean they’re giving up Section 230 protections; it’s the opposite. The very point of Section 230 is to encourage platforms to act as the publisher while shielding them from liability for third-party content. The platforms have always argued they were publishers. Some clueless politicians and media folks insist Section 230 means “you can’t act as the publisher,” but that’s reading the law exactly backwards. The reason all three companies highlight that these are actions of publishers in their complaints is because that’s how you get 230 protections, since it covers publisher activity for third-party content.

Anyway, YouTube also points out that, contrary to the claims from California that recommendations are some nefarious attempt to addict kids, they’re actually a useful feature of any content provider. They also note (again, contrary to popular belief) that YouTube’s own research found that tuning the algorithm toward diverse, novel content, rather than driving people down a rabbit hole of “more of the same,” boosts long-term satisfaction over short-term habitual use:

For example, YouTube strives to promote long-term satisfaction over short-term engagement. And YouTube has determined that values like discovery and novelty advance that goal. Therefore, YouTube has designed its algorithm to recommend to users content that is different from what they have viewed before so they can discover new content and new areas of interest they did not even know they had. That requires YouTube employees to make subjective decisions such as whether topics like physics and math are sufficiently different so that they count as diverse, to handpick training examples that teach the algorithm what qualifies as diverse, and to monitor the system to ensure it is correctly balancing new versus familiar content. In that way, YouTube’s algorithm acts as a tool for “implement[ing] human beings’ inherently expressive choice” to promote novelty and diversity. Moody, 603 U.S. at 746 (Barrett, J., concurring). Of course, YouTube cannot determine what will be novel and diverse to a user without considering what she has viewed before—precisely what AB 1709 prohibits.

YouTube also turns the Ninth Circuit’s SB 976 ruling (the same one that knocked out trade-group standing) back on the state:

The Ninth Circuit has also observed that “the more an algorithm implements human editorial directions, the more likely it is to be expressive.” Bonta, 152 F.4th at 1014. YouTube’s algorithm implements a multitude of human editorial judgments at every stage of the recommendation process. Humans at YouTube decide what editorial goals should be built into the system, select and design signals to be used as inputs, establish principles for how to translate those signals into recommendations, set metrics to measure success, and continually monitor, test, and refine the algorithm to ensure outputs reflect YouTube’s goals and values.

Oh yeah, YouTube also has this fun quotable line saying that California is basically telling a library to mix up the books on its shelf to make it harder for kids to read:

AB 1709’s legislative findings make clear that the State seeks to reduce the amount of time minors spend accessing speech on social-media platforms by forcing platforms to organize or display that speech in a less interesting way. The law’s objective is thus directly “[]related to the suppression of free speech”—just like a law requiring libraries to mix books up on shelves so minors cannot find ones that interest them and spend less time reading.

Meta similarly points out that recommending content people might like is kind of classic protected First Amendment activity and generally a good thing:

The State’s editorial standards would also interfere with Meta’s ability to recommend to users the content that Meta believes those users will want to see. For example, if a teen has previously demonstrated an interest in science experiments or content about local scholarships, the Act would preclude Meta from recommending additional content about science experiments or local scholarships to that user unless they specifically enter keyword searches for “science experiment” or “local scholarships” every single time they log in. See § 22682(b)(2) (exempting certain feeds, but only if they are recommending content based on “search terms that are not persistently associated with the user or user’s device”). The same is true of content about California trade schools, colleges and universities, military recruitment, local volunteer opportunities, the San Francisco Chronicle, and California public officials. And if a teen has already seen content from the San Francisco Chronicle or a public official—such as a breaking news story or a sound clip from a recent debate—the Act may preclude Meta from using that user-specific information to avoid showing the identical content to the same teen again. See § 22682(b)(1) (exempting certain feeds, but only if they are recommending content that “does not concern the user’s previous interactions with media generated or shared by others”).

TikTok’s version takes the argument even further to point out that banning recommended feeds actually can do real harm to the very kids California pretends to be protecting because it would disable safety features such as figuring out which videos not to show users if it looks like they may be unhealthy for them, demonstrating (again) how little the authors of this bill understand how anything works:

… the Act is also overinclusive (and counterproductive). By banning any consideration of a user’s past activity when recommending content, the Act undermines the State’s interest in minors’ wellbeing by making it more difficult for TikTok to avoid showing teenage users content that may be harmful to them or to show them content that may be beneficial for them. TikTok’s dispersion rules, which prevent a vulnerable teenager’s feed from filling with one video after another on the same difficult subject, depend on the user’s “previous interactions with media generated or shared by others”—which the Act forbids TikTok to consider. § 22682(b)(1). TikTok therefore cannot personalize feeds to reduce the amount of potentially harmful content a minor user consumes.

Moreover, the Act is overinclusive in barring personalization for every teenager under sixteen, every interest, and every topic, regardless of how much time the teenager spends on the platform, what her parents want, or whether she has experienced any harm at all. For example, it reaches TikTok’s STEM feed—a feed the user affirmatively selects, confined to science, technology, engineering, and mathematics, and organized by relevance to the user rather than by chronology. That feed bears scant resemblance to the “addictive” feeds the Legislature appears to target; the Act sweeps it in only because it is prioritized “based, in whole or in part, on information provided by the user.”

The arguments on the “autoplay” ban are actually pretty different between the three companies. TikTok argues that because autoplay is not well-defined in the law, they’re not even sure TikTok qualifies, since users have to swipe to see the next video. Does that count? ¯\_(ツ)_/¯ The law is unclear (great drafting as always).

The Act bans “autoplay” without defining it—leaving platforms to guess whether California regulators will apply the statute to platforms’ choices with respect to how they present their content. The lack of any definition in the statute renders this provision unconstitutionally vague. It will chill speech, as platforms are pressured to err on the side of caution and not provide any expressive product that could be deemed “autoplay.”

Meta makes the most complete argument challenging the autoplay rule:

The autoplay ban requires Meta to force users to take affirmative action before Meta can display fully protected audiovisual speech, and therefore burdens Meta’s First Amendment right to engage in the free and unhampered distribution of speech. See id. at 167 (“requiring a permit as a prior condition on the exercise of the right to speak imposes an objective burden on some speech”); see also, e.g., Doe v. Harris, 772 F.3d 563, 573 (9th Cir. 2014) (“a law requiring addressees of ‘communist political propaganda’ to request in writing that the mailing be delivered violate[s] . . . First Amendment rights because the law impose[s] an impermissible ‘affirmative obligation’ and [i]s ‘almost certain to have a deterrent effect.’” (quoting Lamont v. Postmaster Gen. of U.S., 381 U.S. 301, 307 (1965))).

The autoplay ban is therefore subject to First Amendment scrutiny because it imposes an “impermissible affirmative obligation,” Harris, 772 F.3d at 573, that “would tend to restrict freedom to distribute information and thereby freedom of expression,” Teixeira, 873 F.3d at 688 (cleaned up) (quoting Talley v. California, 362 U.S. 60, 64 (1960)); see also, e.g., In re Soc. Media Adolescent Addiction/Pers. Inj. Prods. Liab. Litig., 702 F. Supp. 3d 809, 831 & n.16 (N.D. Cal. 2023) (restrictions on Meta’s “autoplay” feature “would inherently limit what [Meta is] able to publish”).

The autoplay ban independently implicates the First Amendment because it restricts Meta’s First Amendment editorial and curatorial rights. Meta’s feeds are shaped by Meta’s “expressive choices” in deciding how to display particular user-generated content and thus receive First Amendment protection. Moody, 603 U.S. at 716 (a social media service’s “choices about what third-party speech to display and how to display it” are protected by the First Amendment (emphasis added)). That protection has long extended beyond a publisher’s “choice of material” to include decisions such as the “size” and “layout” of the material in a publication. Tornillo, 418 U.S. at 258. Whether to present audiovisual expression continuously or one item at a time is precisely such an editorial decision. Courts have accordingly applied Moody to claims targeting autoplay and similar features, recognizing that “[e]very method of displaying content involves editorial judgments regarding which content to display and where on the platforms.” Patterson v. Meta Platforms, Inc., 239 N.Y.S.3d 726, 733 (App. Div. 2025)

They then argue that the law would need to pass strict scrutiny… and it can’t:

The autoplay ban is seriously overinclusive. “[I]mposing small burdens on vast quantities of speech for no appreciable benefit is not consistent with the First Amendment.” Griffin II, 831 F. Supp. 3d at 833. The autoplay ban applies categorically to every user-generated video and animation presented to all minors, regardless of the amount of time the minor has been online, the underlying content, the duration, the context, the manner of use, the maturity of the user, and the user’s or their parent’s own preferences. The prohibition therefore reaches educational demonstrations, public service announcements, breaking news clips, artistic works, information about scholarships and public programs, civic speech, religious expression, health-related information, and supportive content, without any showing that allowing such expression to autoplay causes the harms asserted by the State. And it sweeps in services—like Facebook, Instagram, and Threads—that voluntarily engage in substantial content moderation efforts and provide supervisory tools for parents and guardians, and that will impose time limits on minors’ daily use by default.

The autoplay ban is also seriously underinclusive. It allows minors to receive the same audiovisual expression after selecting a play button and permits comparable automatic presentation of audiovisual expression on exempt services. It also permits minors to spend unlimited time viewing the same content in other manners, including on other non-covered services and by their own clicks. Those exceptions undermine any contention that prohibiting autoplay directly and materially advances any legitimate state interest. If the same expression may be viewed indefinitely after one additional click or may autoplay on an exempt service, the Act’s chosen distinction bears no adequate fit to its asserted objectives.

Numerous less restrictive alternatives are available. Meta already provides (or will soon provide) parental supervision tools, default two-hour time limits, nighttime restrictions, reminders, content controls, and settings through which parents and users may control how content is presented. And Meta, in the coming months, will provide an optional tool for parents and minors to restrict or turn off the automatic playing of video content.

Google/YouTube take a different approach, skipping the First Amendment argument on autoplay entirely and leaning just on Section 230’s preemption clause, which bars states from imposing liability inconsistent with 230, including liability for how a site curates third-party content.

AB 1709’s autoplay ban thus targets not only a decision related to publication but the very act of publication itself. Publication does not occur just once, when a video is uploaded to YouTube for the first time—instead, a video is published each and every time it is sent to a user. See Barnes, 570 F.3d at 1104 (“[e]very repetition … is a publication in itself”). But the decision whether “to publish” or to “postpone” third-party content is a decision Section 230 protects. Id. at 1102 (citation omitted). In this respect, too, AB 1709 attempts to “prohibit publishing conduct” on its face. In re Apple App Store II, 806 F. Supp. 3d at 1030; see also Doe 1, 174 F.4th at 1167 (holding that the decision to publish content to a user is protected “even when [a] user has not requested the content”).

Even if the ban were to be viewed only as a rule about how to publish content (as opposed to a rule about whether to publish content at all), it still would fail under Section 230. Since the advent of the written word, publishers have had to decide not only what follows a given work, but whether it should follow immediately, later, or only upon request. See Turner Broad. Sys., Inc. v. FCC, 512 U.S. 622, 636-37 (1994) (how to sequence others’ video programming implicates “editorial discretion”). That is the decision YouTube makes with autoplay. It leverages its Recommender System to keep showing related videos that are relevant, trustworthy, and valuable to the user, with only a short pause in between. Because making that decision is “what a publisher does,” Barnes, 570 F.3d at 1102, Section 230 protects that conduct.

They also point out that, you know, TVs autoplay:

To provide a seamless video-viewing experience akin to consecutive programming of traditional television, radio, and many other video- and music-streaming services, YouTube provides an autoplay feature. When autoplay is on and a user finishes watching a video, a related video will automatically begin playing shortly thereafter—just like the next show on television or the next episode on Netflix or Hulu starts playing without the viewer having to pick up the remote, or the next song on a Spotify playlist starts without the listener having to press play. When autoplay is off and a user finishes watching a video, YouTube does not automatically play another video. YouTube may display recommendations to the user, but the user must affirmatively select another video to play.

While that may feel a bit silly, calling out Netflix, Hulu, and Spotify sets up a nice underinclusiveness argument: the law goes after YouTube’s autoplay while leaving those services’ identical feature alone.

A few other highlights from the various complaints. I keep pointing out how anti-science the legislators behind this law are, and Google suggests that as well, pointing to an earlier lawsuit on a different law, showing that California was unable to produce literally any expert evidence to support its claims of harm:

Further, under strict scrutiny, the State must “show a direct causal link” between personalized feeds and “harm to minors.” Id. at 799. And intermediate scrutiny requires the State to “demonstrate that the recited harms are real … and that the regulation will in fact alleviate [those] harms in a direct and material way.” Doe v. Harris, 772 F.3d 563, 577 (9th Cir. 2014) (citation omitted). The State fails either standard because it cannot show any link between personalization and harm to minors. In defending SB 976, the State produced over 500 pages of expert materials and yet was unable to identify a single peer-reviewed or published study establishing that personalization causes or is correlated with negative health outcomes in minors. Indeed, the State’s only study that specifically considered YouTube found no link between personalization on YouTube and harm to minors. The State instead relied on studies purporting to show that social-media use in general, not personalization in particular, is associated with harm to minors. AB 1709’s legislative findings likewise refer vaguely to evidence relating to harms associated with social media generally, not personalized feeds specifically. But AB 1709 bans personalization, not access to social media. So the State must establish that personalization in particular harms minors. It cannot do so.

They also note that in the SB 976 litigation, the state’s own expert admitted if social media addiction actually exists, it affects a very small percentage of users:

The law also bans access to personalized recommendations for all under-16 users— even though the State’s experts have admitted that only a small number suffer from purported “[s]ocial media addiction.” Declaration of Dr. Jenny S. Radesky

That admission does double duty: it shows how much of this is just a moral panic, and it shows how massively overinclusive the law is, which matters a lot for the strict scrutiny analysis.

TikTok points out that the law gives really broad powers to the AG to just decide for themselves what is a bad feature:

The Attorney General also has “complete discretion,” Kolender, 461 U.S. at 358, to “defin[e] in regulations” the meaning of the term “addictive feature,” Cal. Bus. & Prof. Code § 22682(a)(3). The only potential limit on the Attorney General’s discretion to decide what counts as an “addictive feature” is the Act’s open-ended description of “psychologically exploitative features intended to maximize engagement that foreseeably lead to compulsive use.” Id. § 22682(a). The same official who decides what features might fall within that capacious description is also charged to enforce the Act with penalties of up to $50,000 per affected minor….

That arrangement presents the very danger the vagueness doctrine exists to prevent. The “more important aspect of the vagueness doctrine” is perhaps “ ‘not actual notice, but the other principal element of the doctrine—the requirement that a legislature establish minimal guidelines to govern law enforcement.’ ” Kolender, 461 U.S. at 358 (quoting Smith v. Goguen, 415 U.S. 566, 574 (1974)). Such discretion is even more likely to violate the Constitution when a government official may wield their authority to suppress or chill speech

Finally, I’ll close this post out where TikTok opens its brief: highlighting how history is chock full of governments completely losing their shit over some new form of tech or media.

The State doubtless “possesses legitimate power to protect children from harm.” Brown, 564 U.S. at 794. But the interests the State invokes to justify its Act are insufficient to survive the heightened First Amendment scrutiny applicable to the Act’s speech-restrictive provisions. Beginning with “dime novels,” “[r]adio dramas,” “movies,” “comic books,” “television,” “music lyrics,” or “video games,” states—and California in particular—have tried time and again to regulate new forms of expression based on concerns about harm to minors, and courts have time and again struck those regulations down because they conflict with the First Amendment.

Moral panics all around. Eventually we will look back at California’s obsession with stifling the internet the same way we now mock moral panics over comic books and dime novels.