Social Media Warnings Outdo Platforms’ Own Screen-Time Nudges

Every message Stanford researchers tested to warn young people about the harms of social media, from a blunt line about depression to California’s carefully worded statutory notice, made participants say they’d be less inclined to keep scrolling than a message about seatbelts did. In a preregistered trial of 1,012 US teenagers and young adults, warnings on all six health topics the researchers designed, plus California’s official language and a milder screen time nudge, beat a neutral control message at discouraging intent to use social media and at raising awareness of its risks. What the study cannot yet say is whether any of that translates into someone actually closing the app.

The finding matters because warning requirements are no longer theoretical. California, Colorado, Minnesota and New York have all passed laws mandating some version of a social media warning since the Surgeon General called for one in 2024, and Texas has proposed similar legislation. California’s law, the only one to specify exact wording so far, takes effect January 1, 2027, and requires the notice to reappear the first time a user opens a covered platform each day, again after three hours of active use, and roughly hourly after that. That question of what a warning should even say has come up before on the softer end of the policy spectrum too. When ScholarPeer covered a survey of 29,169 European children on outright social media age bans, researchers argued that young people experienced the harsher tool as a loss of autonomy rather than protection, and pushed instead for safer platform design. A warning label sits well short of a ban, but it raises the same underlying question of which lever actually works.

A Weak Nudge Beats No Nudge, But a Real Warning Beats Both

Participants, about half of them 13 to 17 and half 18 to 29, saw all 15 candidate messages in random order on a mock Instagram page and rated each one on how much it discouraged them from wanting to use social media and how much it raised their awareness of the harms. Twelve of the messages were two versions each of six health warnings the researchers wrote themselves, covering addiction, body image, sleep, depression and anxiety, overall mental health, and the claim that social media has not been proven safe. A thirteenth reproduced California’s statutory language verbatim. The last two were a screen time reminder resembling one TikTok already uses voluntarily, and a neutral message about seatbelts that served as the baseline.

All eight warning types, health related or not, scored higher than the seatbelt message, with average differences of 0.32 to 0.62 on the study’s 1 to 5 rating and small to medium effect sizes. More tellingly, every health warning except the one about addiction beat the milder screen time reminder, by margins of 0.16 to 0.30. That distinction speaks to a live policy question: platforms have offered voluntary nudges as an alternative to mandated labels, and this data suggests those nudges land noticeably softer than a message that names a specific harm.

Feeling Warned Is Not the Same as Logging Off

Among the health topics, warnings about overall mental health harms and about depression and anxiety scored highest, both for discouraging use and for raising awareness, edging out the addiction and body image messages by small but statistically reliable margins. Age made no measurable difference in how participants responded, contrary to the researchers’ own expectation that teenagers and young adults might see these risks differently. Gender and self-reported social media use did interact with the results, but the pattern traced back to how harshly different groups rated the neutral control message rather than to how they rated the warnings themselves. Girls and heavier users simply found the seatbelt message less discouraging to begin with, which widened the apparent gap.

California’s own warning, longer and more clinically worded than anything the researchers wrote, performed on par with the strongest messages despite that added complexity. Study author Thomas Robinson, MD, said the language reflects trade-offs beyond persuasion. Robinson said in a statement: “We talked to the people who created the bill, and they told us the rationale behind it. It’s not the message we would have designed, but we think we understand why it performed well, and it was meant to stand up to legal scrutiny. All these bills are in litigation, being challenged by the social media industry.” Courts have previously struck down product warnings for using causal language regulators couldn’t fully substantiate, a constraint that shapes what a legally durable message can say even when a punchier one might test better.

The entire trial rests on a single proxy: whether a message felt discouraging the moment someone saw it, not whether they used social media any less afterward. Lead author Anna Grummon and her coauthors point to earlier research linking that kind of perceived effectiveness to real changes in smoking and drinking behavior, and argue social media warnings deserve the same scrutiny. But the relationship between how persuasive a message feels and how much it actually changes behavior has its own contested history. A widely cited 2018 meta-analysis by communication researcher Daniel O’Keefe compared perceived and measured effectiveness across 151 message pairs and found the two agreed on which message worked better only 58 percent of the time, barely above a coin flip. Nothing in the new trial resolves that debate, and its own authors acknowledge the gap: their single item scale has never been validated specifically for social media warnings, and every participant judged each message only once.

That leaves an odd asymmetry. Lawmakers already have a policy mechanism moving toward rollout, complete with prescribed trigger points for when a warning must reappear on screen, while the evidence for what that warning should say rests on how a message feels in a single sitting rather than on any measured change in how long someone actually spends on the app. Grummon put the gap plainly in a statement: “With substances like tobacco and alcohol, we find, on average, that well-designed warning labels change behavior. But social media is a totally different product from cigarettes or alcohol or sugary drinks, where you just put the warning on the product and it lives there. Social media laws are written so that these warnings are basically pop-ups. We need to find out more about how that changes behavior.” Until someone runs that test, the four states with warning laws already on the books are, in effect, running the experiment themselves.

FAQ