Several US states have proposed or adopted policies requiring health warnings on social media platforms, and the evidence base for what such a warning should say has been thin. This within-participants randomised study put 1,012 US teenagers and young adults aged 13-29 in front of a mock social media page and showed them 15 messages in random order: 13 health warnings covering different harms, a screen-time break warning of the kind some platforms already display voluntarily, and a neutral control.
All health warnings and the screen-time warning were rated as more effective at discouraging social media use than the control (average differential effects from 0.32 to 0.62 on a 1-5 scale). Every health warning except the one about addiction was rated more effective than the screen-time break warning (differences 0.16 to 0.30). The same pattern held for perceived awareness of harms.
What is being measured matters. Perceived message effectiveness is what participants think a message would do to them, rated immediately after seeing it on a mock page - not observed change in social media use, not mental health outcomes. It is a standard early-stage method for selecting warning content, and it predicts real-world effect imperfectly at best.
The policy-relevant finding is comparative and reasonably robust to that limitation: the voluntary screen-time nudges platforms have adopted were rated as the weakest option tested. If a jurisdiction is going to mandate a warning, naming a specific harm appears to land harder than telling someone they have been scrolling for a while. For clinicians advising families, the transferable point is the same - a specific consequence is more persuasive to an adolescent than a time limit.
- This is a message-testing study, not evidence that warnings change behaviour or mental health
- Naming a specific harm was rated more effective than a generic screen-time reminder
- The addiction framing was the one health warning that did not outperform the screen-time nudge
- No regulatory action has been taken here - these are proposed and state-level policies
- When advising families, name the specific consequence rather than setting only a time limit
The statistics, in plain English
An average differential effect of 0.32 to 0.62 on a five-point scale is a modest shift in rating, and the confidence intervals are narrow because each of 1,012 participants rated all 15 messages - a within-participants design that is efficient but measures relative preference rather than absolute impact. The addiction warning's interval (-0.004 to 0.14) includes zero, so it was indistinguishable from the screen-time message. Perceived effectiveness correlates with actual behaviour change only moderately in the warning-label literature, so treat the ranking as a guide to what to test next.
Read the rest in the app
You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.
Scan to keep reading on your phone. No account needed to start.
Tomorrow morning, before your first patient
One edition a day for psychiatry, written by the desk, every claim tied to its paper. Six minutes.
Get the app — free