Most people do not need a warning label to understand that they have been scrolling for a while. The thumb knows. The battery knows. The clock on the lock screen definitely knows. What a warning label can do is something different: it can interrupt the activity with an official description of what that activity might mean.
That distinction sits at the center of new research from Stanford Medicine, where investigators tested 15 messages with 1,012 teenagers and young adults to see which kinds of social-media warnings people found most discouraging. The messages covered topics including overall mental-health harms, depression, anxiety, body image, sleep disruption, addiction and the idea that social media has not been proven safe for young people. Participants viewed the messages on a mock social-media page and rated how much each one discouraged them from wanting to use social media and how much it increased their awareness of possible harms.
The strongest responses went to warnings about broad mental-health harms and messages specifically naming depression and anxiety. The researchers found that nearly all of the health warnings were perceived as more effective than a simple screen-time-break message similar to prompts platforms sometimes provide voluntarily. Teenagers and young adults responded in broadly similar ways, despite the researchers' expectation that people growing up only a few years apart might react differently to digital-media language.
The finding sounds simple: warnings can matter. The more interesting result is that wording matters too.
A label is easy to think of as a container for facts. Put the relevant risk inside a box, display it prominently enough, and the user either pays attention or does not. But the Stanford experiment suggests the sentence itself is part of the intervention. “Take a break” is not psychologically identical to “this activity is associated with anxiety,” even when both appear at the same point in the same scrolling session. One describes a scheduling choice. The other gives the user a way to classify what may be happening to them.
That is where social-media warning labels move into familiar ProbleMattic territory. Language does not merely report behavior; it helps organize behavior. Once a repeated action is officially placed next to words such as anxiety, depression, addiction or mental-health harm, the action acquires a new interpretive frame. A person who yesterday might have said, “I wasted an hour scrolling,” may eventually have a different sentence available: “I am doing the thing the warning keeps telling me may be affecting my mental health.” The behavior can remain physically identical while its meaning changes.
For now, it is important to keep the research result in the right lane. The Stanford study did not establish that participants actually reduced their social-media use after seeing a warning. Its primary outcome was perceived message effectiveness: participants said the warnings would discourage them from wanting to use social media. The researchers themselves describe actual behavioral testing as the next step. But that does not make the current result trivial. Regulation often begins by deciding what language people should encounter before anyone knows exactly how that language will alter long-term behavior.
And that decision is already leaving the laboratory.
When Stanford published the findings Sept. 4 in JAMA Health Forum, the university noted that California, Colorado, Minnesota and New York had already enacted forms of social-media warning requirements. The details differ substantially. Minnesota's law, effective July 1, requires a conspicuous mental-health warning each time a user accesses a social-media platform and directs the state health commissioner to maintain evidence-based guidance for the label. New York's law, scheduled to take effect Jan. 1, 2027, goes much further into presentation: the required warning must occupy at least 25 percent of the screen on initial daily access and later expand to at least 75 percent after three hours of cumulative active use.
California has written the words themselves into law. Its warning says that the Surgeon General has warned that social media may have benefits for some young users while also being associated with significant mental-health harms and not having been proven safe for young users. The Stanford team tested that language alongside its other messages and found that it performed relatively well despite being longer and more legally constructed than the shorter experimental warnings.
This is where the policy problem becomes unusually linguistic. Legislatures are not simply deciding whether a box should appear. They are choosing nouns, verbs, levels of certainty, screen size, timing and repetition. “May contribute to” is different from “causes.” “Mental-health harms” is different from “depression and anxiety.” “Take a break” describes an action. “Not proven safe” describes a status. Each version tells the user not only that there is risk, but what kind of risk they are supposed to notice.
That makes warning-label policy a form of behavioral infrastructure. The wording will be inserted into the normal architecture of the product: between login and feed, between one hour and the next, between a person and the next piece of content. It may arrive every day, or after a usage threshold, or during late-night sessions. Repetition turns the warning from a one-time statement into part of the environment.
We already know how that can happen in other domains. A warning on a cigarette pack, a medication insert or a gambling interface does more than transfer information. Over time, it teaches a vocabulary for talking about the activity. People learn to describe a side effect, a risk category or a pattern of use through the terminology made available to them. Social media is different from a physical product because the warning itself can be programmed into the experience, timed to behavior and displayed at exactly the moment a regulator considers significant.
That difference could make the label either more useful or easier to ignore. A cigarette package cannot notice that somebody has smoked for three hours and enlarge its warning. An app can. But an app can also train users to dismiss pop-ups with extraordinary efficiency. Anyone who has reflexively clicked through a cookie banner, update notice or subscription prompt knows that repeated digital interruption can become invisible through familiarity. The eventual success of warning-label laws may depend not only on the message but on whether the message remains cognitively present after the hundredth exposure.
The Stanford researchers are already thinking in those terms. Lead author Anna Grummon said subtle changes to warnings can matter, and the team plans further work to determine whether the labels actually reduce social-media use. That next stage is critical because there are at least two plausible effects to measure. A warning may change what people do. It may also change how they understand what they are doing, even if they continue doing it.
The second effect should not be dismissed as merely rhetorical. Self-interpretation can alter future choices. If a user begins to recognize a late-night scrolling session as something associated with sleep disruption, that recognition may eventually affect when they stop. If a teenager encounters repeated language connecting social-media use with anxiety and then notices anxiety during or after a session, the official label has supplied a possible explanation. Whether that explanation is accurate in any individual case is a separate question; the point is that the label becomes available as a category for interpreting experience.
That power is also why the words deserve scrutiny. Behavioral labels can clarify, but they can also flatten. “Social media” covers very different activities: messaging a friend, watching short-form video, participating in a hobby community, following breaking news, posting art, arguing with strangers, comparing appearance, doomscrolling or simply wasting ten pleasant minutes. A warning designed for a platform has to speak broadly enough to apply across those contexts while being specific enough to mean something. The Stanford results suggest that more concrete mental-health language may resonate, but regulation still has to decide how much complexity can fit inside an interruption without turning the warning into legal wallpaper.
There is also a political and constitutional fight running alongside the behavioral one. Stanford noted that state warning-label laws are being challenged by the social-media industry. Colorado's youth-notification requirement, for example, became law in 2024 but has faced litigation over compelled speech. That means policymakers are simultaneously trying to answer three different questions: what message is scientifically supportable, what message is behaviorally effective, and what message can survive legal review.
New Jersey is circling the same issue. A bill introduced in the state Legislature would require a black-box warning using language similar to California's, while a January state youth-mental-health report separately recommended warning labels for social-media platforms. The proposal has not become New Jersey law, but it illustrates how quickly a Surgeon General's recommendation has turned into model legislative language moving from state to state.
That spread is culturally significant because official warnings have a way of becoming ordinary. The first time a social platform tells a young person that prolonged use may affect mental health, the message may feel unusual. A few years later, it may be as unsurprising as a nutrition label or prescription warning. The language moves from controversy into the background architecture of everyday life.
And once it is there, it can change the story people tell about their own behavior.
That may be the most interesting outcome to watch as these laws take effect. The obvious metric will be minutes: did users spend less time on the platform? Did sessions get shorter? Did late-night use decline? Those are important questions. But there is another measurement problem underneath them. Did users start noticing their own behavior differently? Did they become more likely to connect certain feelings with certain patterns of use? Did “I was scrolling” become “I was scrolling even though I knew what the warning said”?
Regulation sometimes works by physically restricting an action. Sometimes it works by making an action harder or more expensive. Warning labels operate in a stranger space. They insert language into the moment of choice and hope that the language changes what follows.
The Stanford study suggests that not all language has equal force. That means the coming debate over social-media warnings will not end with the question of whether platforms should display them. It will move, inevitably, to the exact sentence.
Because once the warning is part of the feed, the words are part of the product too.
SOURCE NOTES
• Stanford Medicine / Stanford Report — “Warning labels could curb young people’s social media use,” Sept. 4, 2026
• JAMA Health Forum — “Designing Social Media Health Warnings for Teenagers and Young Adults: A Randomized Clinical Trial,” Sept. 4, 2026
• Minnesota Statutes §325M.335 — Mental Health Warning Label
• New York General Business Law §1521 — Requirement for Warning Labels
• New York General Business Law §1522 — Display Requirements and Restrictions
• Colorado General Assembly — HB24-1136, Healthier Social Media Use by Youth
• New Jersey Legislature — A4013 (2026), proposed social-media warning requirements
• U.S. Department of Health and Human Services — Social Media and Youth Mental Health
Study findings attributed to Stanford Medicine / JAMA Health Forum materials cited in SOURCE NOTES. The study measured perceived message effectiveness, not proven reduction in use. New Jersey A4013 is proposed, not enacted. Cultural framing is RMN's.