Have you noticed headlines about Generative AI and teens just about everywhere lately? Despite the technology’s newness, the basic storyline feels… all too familiar. We’ve been here before. The social media era gave us a preview of what happens when powerful new technology reaches young people faster than our understanding of it does. In that era, research lagged behind deployment by years, and solutions like content moderation, screen time tools, and age verification that eventually emerged were generally bolted on after the fact, rarely shaped by expertise in adolescent development and well-being or by young digital natives themselves. As our colleagues in youth-led coalitions like Design It For Us argue powerfully, we are collectively still paying dearly for that. Litigation, legislative fights, belated industry reckonings, not to mention the unfolding impacts on all of us, young people in particular: these are the costs of building first and understanding later.

Compared to that earlier era, however, AI is moving even faster and reaching even deeper into the parts of adolescent development that matter most. How young people form relationships, regulate emotions, and figure out who they are. We don’t get a ‘do-over’ on social media, but we do get a choice about how the next chapter goes, and about the future we collectively bring into being.

The choice comes down to who is in the room.

Over the past year or two, I’ve been getting a lot of invitations. Convenings about AI and youth well-being. Roundtables about AI companions. Working groups about what we know, what we’re worried about, and what we should do. I suspect this is true for many of us working on youth mental health and well-being. And I imagine that, like me, folks are saying yes to many of these invitations, precisely because these conversations feel so urgent and the implications so profound.

But I kept noticing something in those rooms.

At these convenings, often focused broadly on topics like AI companions, social connection, or human flourishing writ large, something predictable kept happening. The conversation kept drifting back to young people. It always does. We can’t (and shouldn’t) talk about the societal or individual implications of AI, or technology more generally, for that matter, without talking about the generation growing up in its midst. So many of these rooms include amazing colleagues and experts — academic researchers focused on technology and media, data scientists, trust and safety leaders from frontier AI labs, policy and advocacy experts from a wide range of key areas, including children and technology. All genuinely and authentically grappling with the implications of AI for young people.

And yet, there are often no developmental scientists in the room. Often, few young people themselves are in expert positions. If any young researchers are present, they might be assigned to take notes.

Around the same time, colleagues working in our space — advocates, funders, policy leaders — kept asking us to point them to the best research on the developing adolescent brain and generative AI. What do we actually know? What should we be worried about? What does the evidence say?

Our team went looking. We scanned the NIH grants database and the published academic literature. We couldn’t find much. And it made us wonder — how will we do it differently this time around, when we are already grappling with decisions about how generative AI is designed, tested, regulated, and deployed into young people’s worlds, while the adolescent developmental science and developmental neuroscience seem largely missing?

That’s why Hopelab commissioned SPARK (Survey of Possibilities on AI Research and Knowledge).

We partnered with the UNC Winston Center on Technology and Brain Development and a brilliant team of adolescent developmental scientists led by Dr. Annie Maheux, Dr. Kaitlyn Burnell, and Dr. Anthony Vaccaro — and asked them to map the landscape: what research exists, what’s in the works, what’s missing, and what the field most urgently needs to study.

The SPARK team, led by Dr. Annie Maheux, Dr. Kaitlyn Burnell, and Dr. Anthony Vaccaro, surveyed 100 researchers and professionals, interviewed 15 experts across industry and academia, and systematically reviewed all peer-reviewed studies on adolescent AI use published between 2020 and 2025, 82 in total. This work has been submitted for publication at prominent peer-reviewed research journals and can be accessed in pre-print form below.

Here’s the finding that stays with me: there are zero published studies looking at AI’s effects on the developing adolescent brain. That’s not completely surprising — we know neuroscience studies, especially with teens, take time, meticulous planning, and significant funding — but it IS a signal. A signal about the existing mismatch between the developmental science and the urgency of the questions from the field about how AI is impacting teenagers’ development today.

This mismatch was borne out by the full research findings across all available published social science on adolescent AI use.

As the UNC team concluded: “Collectively, the literature reflects a field in its earliest stages. Despite the rapid emergence of evidence, studies are geographically concentrated, methodologically limited, and organized around a narrow set of questions that do not yet capture the full scope of how adolescents use AI and the potential implications for development.”

 

The developmental research that does exist in the academic literature skews heavily toward education and learning — how students use AI for homework, whether it helps or hurts their writing. Less than 15% of it addresses mental health. Fewer studies still look at AI companionship, social development, or what it means for a teenager to form an emotional attachment to something designed to be perpetually agreeable and endlessly available.

This gap matters: the adolescent brain is not a smaller version of an adult brain. Adolescence is a profoundly consequential neural remodeling phase, when new brain cells form and build connections at a remarkable rate, second only to early childhood. This is a window of development when direct experiences in our environments and relationships leave a lasting mark on how the brain gets wired. The systems that are most active during the adolescent years, between ages 10-25, are the ones that make peer approval feel so good (and rejection sting so much), that govern the still-developing ability to manage big emotions and read social and emotional cues like facial expressions or social rejection, and that push us out into the world to explore, take risks, practice trying and failing, contribute, and figure out who we are. Those are precisely the systems that AI is most likely to interact with. The work of this landscape review points to exactly this collision – the most critical questions for AI use and adolescent development are exactly those which we have the absolute least empirical understanding of.

Compounding the problem: the researchers trying to build this new science are working in isolation — they’re not using the same definitions or the same tools, so their findings can’t be compared or combined into a bigger picture. And the pace mismatch is relentless. As one researcher in the SPARK survey put it, the work is “already outdated and still under review.” Forget getting it to anyone who could act on it in time.

The SPARK report is a starting point. It maps who’s doing this work, what they’ve found so far, what’s missing, and what the field most urgently needs to study. It also points toward what a more coordinated, better-resourced, and more community-engaged research ecosystem could look like, and the critical role that “connective tissue” infrastructure at the intersection of research, youth expertise, and the tech sector might play in building it.

We’ve seen this story before, and it’s easy to accept the narrative that the future is on repeat, inevitable. But the truth is, there’s a lot that’s different. For one thing, we know a lot more than we did before the social media era, and we’re using those learnings to see what’s coming this time around and call it out clearly. Important work done by The Rithm Project has been tracking this dynamic, from courtroom verdicts to sycophantic design features in AI that echo social media’s infinite scroll — and asking what it means for the choices we make now. As they and other activists and advocates point out, there is another possibility, a future that young people deserve. And it starts with our ability to collectively imagine something different, to get this right. Not because the research alone will solve the new challenges facing us — it won’t. But building a movement that makes this technology work for young people, rather than the other way around, requires a coordinated collective effort from the start. That means developmental scientists, young people, innovators, advocates, and technologists collaborating to do work that actually connects to the decisions product teams and policymakers are making right now, not years from now.

We’ll be sharing more about what we’re learning and building in the months ahead. For now, we’re proud to share this work as a resource for the field, and we invite the researchers, advocates, young people, parents, educators, funders, and product teams who care about these questions to join this conversation, share your ideas for radically different connectivity across this ecosystem, one that puts young people, and the science behind them, in the center of a new story about the future that doesn’t just repeat the past.