AI-driven social media monitoring tool aims to identify teen mental health decline

As concerns grow about the effects of social media on adolescent mental health, researchers at the Children’s Healthcare of Atlanta Pediatric Technology Center at Georgia Institute of Technology are exploring whether smartphones themselves could become tools for early intervention. Investigators are evaluating an opt-in artificial intelligence (AI)–driven social media surveillance platform designed to identify signs of worsening mental health in teenagers before a crisis occurs. The project will focus on adolescents aged 13 to 17 years who have previously presented to emergency departments for severe mental health concerns and are now receiving outpatient follow-up care.

The research is funded by the Children’s Healthcare of Atlanta Pediatric Technology Center at Georgia Tech, a collaboration between Children’s Healthcare of Atlanta clinical experts and Georgia Tech scientists and engineers focused on developing technological solutions to problems affecting the health and care of children. The initiative combines clinical expertise in pediatric behavioral health with advanced computational analysis of online behavior. John Constantino, MD, chief of behavioral and mental health at Children’s Healthcare of Atlanta, is collaborating with Munmun De Choudhury, PhD, professor of interactive computing at Georgia Tech, whose prior research has examined how linguistic patterns, online interactions, and digital behaviors may predict adverse mental health outcomes. In this Q&A, Constantino and De Choudhury discuss the clinical rationale behind the project, how the tool is being developed and validated, and the ethical considerations surrounding real-time monitoring of adolescent social media activity for suicide prevention and mental health support.

Contemporary Pediatrics: This study introduces an AI-driven, opt-in social media monitoring tool for adolescents. What clinical gaps in current pediatric mental health screening and monitoring does this approach aim to address?

Constantino: This approach aims to address the absence of validated safeguards on social media activity. Current parental controls are relatively untested and unproven in relation to protecting children from serious adverse consequences of negative social media experiences.

De Choudhury: One major gap in pediatric mental health care is that clinicians often rely on intermittent, retrospective assessments that occur only during clinic or emergency visits. Yet many indicators of escalating distress, suicidality, or social isolation unfold in real time in adolescents’ everyday digital environments. Our work aims to bridge this gap by developing an opt-in, patient-centered system that can identify patterns of acute risk from social media activity between clinical encounters. The broader goal is not to replace clinical judgment, but to provide pediatric providers and caregivers with earlier, contextually grounded signals that may enable more timely support and intervention.

Contemporary Pediatrics: Can you walk pediatricians through how the tool stratifies risk into different alert levels, and how these alerts might integrate with existing clinical workflows or caregiver communication?

Constantino: Risk stratification is highly aligned with clinical decision-making about information-sharing in adolescent medicine practice. As is standard for patient care in adolescent psychiatry, participating youth would be informed that surveillance information would not be shared with parents or authorities unless it reflects a potential imminent risk of harm. The first level of alert (risk that falls below the threshold for concern for imminent risk) is to the youth themselves, encouraging perspective-taking, pausing if necessary, and providing compensatory messaging to reframe social media experience and healthy responses to it. If youth manifest signs of risk of harm to self or others, parents are notified and provided information on immediate approaches to ensuring appropriate evaluation and support. If the youth communicates an overt risk of harm to self or others, and parents cannot be immediately reached or are unable to respond, an emergency responder alert would be issued by the AI agent.

De Choudhury: The system is designed to stratify risk based on the severity, persistence, and trajectory of signals detected in social media content and related behavioral markers. For example, transient expressions of distress may generate low-level monitoring flags that engage in compensatory messaging with a teen, whereas sustained indicators of suicidality or self-harm escalation may trigger higher-priority alerts. Importantly, the intent is to align these alerts with existing pediatric behavioral health workflows—such as triage protocols or caregiver outreach—rather than creating parallel systems. We are also working closely with clinicians and families to ensure that alerts are interpretable and actionable, and to minimize unnecessary burden or alarm.

support.

Contemporary Pediatrics: Your study focuses on adolescents with prior emergency department visits for severe mental health concerns. How do you see this tool being applied in broader pediatric populations, including primary care settings?

Constantino: The current study features a high enough “yield” from retrospective social media data collection to “train” the AI agent on predictors of serious risk and to refine the model for continuous surveillance and prediction. Instituting a safeguard that is relevant to all pediatric smartphone/media users would promote a “safety first” culture surrounding social media activity, coupled with an opportunity for all youth to balance deleterious social media exposures with positive, compelling messaging, the goal of which is to buffer the potential toxicity of social media and continuously present approaches to resilience.

De Choudhury: Adolescents with prior emergency mental health visits represent a particularly high-risk population where the need for proactive monitoring is urgent. However, the longer-term vision is broader. Many young people first present signs of distress in primary care settings, where providers often face limited time and limited visibility into patients’ lived experiences outside the clinic. An approach like this could eventually support earlier identification of risk trajectories before crises emerge, particularly for youth who may not otherwise engage with specialty mental health care. At the same time, any broader deployment would require careful validation across diverse populations and settings to ensure equity, reliability, and clinical usefulness.

support.

Contemporary Pediatrics: Given the use of social media and electronic health record data, what safeguards are in place to address privacy, consent, and ethical considerations—especially for minors?

Constantino: These issues are taken very seriously in the approach to participants, and the very highest levels of data security have been incorporated in the study design and methods. There are numerous protections from potential breaches, and the investigational review boards of the respective institutions of Georgia Tech and Children’s Healthcare of Atlanta have determined that the potential benefits of knowledge to be gained in relation to the serious risks incurred by social media experience for many youth outweigh the potential risks of privacy breaches, given the safeguards. Youth are approached from a position of collaborative exploration of how to ensure safety. The study is completely voluntary, and participating adolescents are typically engaged in the spirit of recognizing the risks of social media for themselves and their peers, and with the objective of making social media activity safe.

De Choudhury: Privacy, consent, and ethical governance are central to this work. The system we envision is fully opt-in, with informed consent and assent procedures designed specifically for adolescents and caregivers. We will collect only data that participants explicitly choose to share, and all data will be stored within secure, HIPAA-aligned research infrastructures with rigorous de-identification and access controls. Importantly, we view this as a patient-centered partnership. We are also working closely with clinicians and youth stakeholders to ensure that the technology respects autonomy, minimizes potential harms, and remains transparent about how risk assessments are generated and used.

support.

Contemporary Pediatrics: If validated, how might this real-time surveillance model change the way pediatricians approach early identification and intervention for suicide risk and other acute mental health crises?

Constantino: There will be an opportunity for participating youth and their parents to share with pediatricians the tiered support alerts generated by the AI surveillance tool. In tier 1, compensatory messaging is specifically keyed to the elements of social identity or mood deterioration that are detected by the AI agent and therefore reflect targets for intervention that are of potential clinical relevance to pediatricians and therapists supporting a participating youth.

De Choudhury: If validated, this model could help shift pediatric mental health care from a largely reactive framework to a more proactive and continuous model of support. Instead of identifying risk only after a crisis has escalated to an emergency visit, clinicians may be able to recognize concerning trajectories earlier and intervene in more timely, preventive ways. More broadly, it reflects a growing recognition that adolescents’ digital environments are an important part of their social and emotional lives, and that responsibly integrating those signals into care—when done ethically and collaboratively—could strengthen how we support youth mental health.