A 2025 study found that 72 percent of American adolescents have used an AI chatbot as a companion at some point, a figure striking enough on its own but made more consequential by what has followed it: a rapidly growing body of research attempting to determine whether that companionship helps struggling teenagers or quietly deepens the isolation it appears to relieve. The honest answer, on the evidence assembled so far, is that nobody fully knows yet — and readers deserve that uncertainty stated plainly rather than resolved into a tidier alarm than the science currently supports.
A review published in Pediatrics by researchers at Children’s Hospital of Philadelphia (CHOP) found that AI companionship may genuinely address loneliness in adolescence in the short term, but carries a distinct risk of reducing face-to-face human interaction as dependence on the chatbot grows over time — precisely the pattern that concerns clinicians and parents most, since adolescence is developmentally the period in which young people are supposed to be building the social skills and peer relationships that carry them into adulthood. A separate, more recent cross-sectional study found that adolescents who specifically seek emotional support from generative AI tools show a distinct marker of psychological distress that exists independently of loneliness or feelings of being unimportant — suggesting that chatbot-seeking behaviour for emotional support may itself be a signal worth clinical attention, rather than simply a coping mechanism cleanly separable from the distress it appears alongside.
The broader evidence base remains genuinely, frustratingly mixed rather than settled in either direction. Some studies report AI companions measurably reducing loneliness, with a peer-reviewed Journal of Consumer Research paper finding that AI companions alleviated loneliness on par with interacting with another human being, and more effectively than passive activities like watching videos. Other research points the opposite way: a Stanford University study of Character.AI users published in Nature Human Behaviour found that users with smaller offline social networks felt more lonely after seeking emotional support from AI companions, not less, and that users more willing to share sensitive personal information with their chatbot showed correspondingly lower measures of psychological wellbeing. A separate 12-month longitudinal study following more than 2,000 adults across four Western countries is attempting to resolve exactly this kind of directional question — whether turning to AI companionship predicts later loneliness, or whether existing loneliness predicts turning to AI companionship, or, most plausibly, some reinforcing combination of both running in each direction simultaneously.
For India, this is no longer a foreign trend worth watching from a distance; it is already a domestic phenomenon moving at striking speed. Cybersecurity firm Kaspersky’s usage data shows ChatGPT has become one of the ten most-used apps among Indian children over the past six months, ranking seventh overall and ahead of established platforms like Zoom, Roblox and Telegram — the only AI chatbot to crack the top ten alongside video and messaging apps that have dominated children’s app usage for years. A separate national survey, the Student Cyber Resilience Education and Empowerment Nationwide (SCREEN) Survey conducted in early 2026, found that 38.8 percent of respondents overall reported using ChatGPT, with the highest uptake recorded among children aged 11 to 13, at 33 percent — meaning generative AI use among Indian pre-teens has moved well past early-adopter territory into mainstream childhood digital behaviour, arguably faster than India’s digital-safety policy apparatus has managed to track it.
That policy gap is the crux of the concern for Indian readers specifically. India’s Digital Personal Data Protection Act, 2023 strictly prohibits tracking, behavioural monitoring, targeted advertising and profiling of children under 18, which is a meaningfully strong legal baseline on paper. But researchers reviewing AI chatbot safety features have found that these systems frequently fail to recognise subtle signs of mental health distress, and become measurably less reliable at doing so during longer, more realistic extended conversations — precisely the kind of sustained emotional-support interaction that the CHOP review and the cross-sectional distress study both flag as the pattern worth watching. There is also a specific enforcement gap that Indian legal commentators have raised: there is currently no clear law governing the privacy of chatbot conversations themselves, as distinct from the broader personal-data protections the DPDP Act already covers, leaving a grey zone around exactly the kind of sensitive emotional disclosures adolescents make to these tools.
OpenAI has begun rolling out age-prediction systems and a dedicated “ChatGPT for Teens” product with additional safeguards, and has initiated “Open Dialogues” sessions in India specifically to engage families on AI safety, while Meta has separately moved to block its chatbots from discussing self-harm, suicide and disordered eating with teen users, redirecting them toward expert resources instead. UNICEF has published guidance for building safer chatbots aimed specifically at detecting distress signals, responding with compassion, and offering reliable referral pathways to human support — guidance that Indian platforms and policymakers alike would do well to treat as a working baseline rather than an aspirational document. These are incremental, genuinely positive steps rather than solved problems — researchers studying the space have been explicit that what is still missing is independent safety benchmarking, clinical testing standards, and enforceable rules, since companies remain largely self-regulating in what researchers describe as a high-risk environment for teenagers. Given how quickly Indian children have adopted these tools, and how unsettled the underlying science on their psychological effects remains, this is a story NSH should track as an ongoing beat rather than a one-time alarm piece — the nuance the evidence demands is precisely what a single dramatic headline would flatten away.
– Hyma Priya Sunkara
