“Already outdated and still under review”: mapping the landscape of research on adolescent development and AI chatbot use

* The study involved a survey of $n=141$ researchers and professionals and semi-structured interviews with $n=15$.
* AI use cases rated for study importance ranged in mean scores from 2.61 to 3.85 on a 4-point scale.
* Priority AI use cases included those related to immediate harms (e.g., providing dangerous health information, cyberbullying) and AI literacy.
* The study identified five key research themes: studying everything but triaging for severe cases, AI literacy as a foundational skill, AI as a social/relational actor, information/agency in the AI era, and the risk of overreliance.
* Participants recommended methodological pluralism and triangulation across different research methods.
* Proposed guardrails include enforceable youth-centered regulation, developmentally informed design, and multidimensional AI literacy.
* A major barrier identified was the rapid pace of AI change relative to academic timelines and limited industry transparency.
* One participant noted that responsibility for safe AI should not rest primarily on adolescents or families.
* Research priorities included understanding how AI use impacts adolescent social, emotional, cognitive development, and executive functioning.

The pattern emerging from expert consensus is a fundamental tension between the urgent need to address acute, life-threatening harms posed by AI (e.g., self-harm promotion) and the necessity of studying more subtle, long-term developmental shifts resulting from social interaction and cognitive scaffolding provided by AI. The prioritization logic defaults toward immediate risk mitigation, which often overlooks complex but slower-developing psychosocial impacts on identity and relational skills unless they are framed as an acute emergency. This suggests a systemic bias where public attention—and potentially research funding—is channeled toward sensationalized safety issues rather than the slow-burn effects of altering normative social cognition or agency within adolescence.
The emphasis on AI as a "social and relational actor" highlights that current development science frameworks may be insufficient to capture the unique vulnerabilities introduced when technology simulates intimacy. The argument for multimodal methods suggests that no single disciplinary lens can capture this complexity; true progress requires bridging the divide between empirical developmental science, technological design, and social contexts through shared methodologies. The explicit recognition that AI biases reflect human-created data positions the issue not merely as a technical flaw but as an amplification of existing structural inequalities against marginalized youth. The challenge for the field is to move beyond simply cataloging use cases toward developing integrated frameworks that simultaneously account for immediate safety needs and the nuanced evolution of selfhood in an increasingly mediated social landscape.
What future research agenda is being implicitly challenged by the call for coordination? It challenges the traditional siloed approach, demanding that developmental science actively partner with technology designers and policymakers not just to study AI effects, but to co-design protections. If the field fails to achieve this coordination, the default trajectory risks allowing technological innovation, driven by commercial incentives and rapid deployment, to proceed independently of developmental necessity, potentially exacerbating existing inequalities rather than mitigating them. The most significant unanswered question is how to operationalize "developmentally informed design" when the mechanisms of adolescent development itself are still being dynamically mapped.