According to OpenAI’s first report on teenagers’ usage patterns of ChatGPT, this group interacts with the chatbot for less than 15 minutes per day on average, and fewer than 2 percent of them use it for more than three consecutive hours. The release of these statistics comes amid growing concerns about the impact of AI chatbots on teenagers’ mental health and safety.
With the expanding use of artificial intelligence in education and daily activities, developers of this technology face an important challenge: how to provide the educational and practical capabilities of chatbots to teenagers while simultaneously reducing the potential risks of their use?
Special tools for teenagers’ use of ChatGPT
OpenAI began testing a special version of ChatGPT for teenage users last August. This version, aimed at increasing the safety of young users, offers additional protections against sensitive content and, alongside that, provides various educational features.
Among these features is a study mode that, instead of giving direct answers, provides step-by-step guidance for learning. Creating flashcards for review, designing quizzes, and helping plan college application requests are other educational uses of these tools.
Another feature considered for teenagers is sending reminders to take breaks during extended chatbot use. According to OpenAI’s report, about half of the conversations in which a break reminder was shown ended or were stopped within the next five minutes.
These statistics indicate that short reminders may help stop chatbot use in some situations; however, on their own they do not prove that this feature will reduce excessive use or improve teenagers’ mental health in the long term.
Parental controls and alerts about high-risk situations
OpenAI has also provided parental control tools that allow families to link a teenager’s account to their own account and manage certain settings related to using the service.
These mechanisms are designed to increase parental oversight and identify sensitive situations. Notable concerns include possible signs of self-harm, suicidal thoughts, and eating disorders-topics that are especially important in discussions of AI chatbot safety for underage users.
However, the presence of protective tools does not necessarily mean they function perfectly. The effectiveness of such features depends on how high-risk situations are detected, the timing of alerts, and the conditions under which account settings are active.
Common Sense Media’s criticism of protective measures’ performance
In contrast, Common Sense Media has criticized some aspects of these protective measures’ performance. After reviewing over four thousand test cases, the organization questioned the functioning of some protections and stated that not all features worked as expected in the tested conditions.
Based on the results presented in this evaluation, in some cases related to signs of suicide, self-harm, and eating disorders, the expected alert was not sent to parents. Also, some educational features may behave differently than expected in specific situations, including during exams.
These criticisms highlight one of the main difficulties in designing AI chatbots: the system must be able to distinguish between ordinary conversations, educational requests, and possible signs of crisis-something that is not simple in all situations.
OpenAI’s response to the criticisms
OpenAI has explained in response to these criticisms that some tests may not fully reflect the real-world performance of protective tools when these features are active.
The company also noted that after linking a teenager’s account to a parent’s account, it may take several hours for some alerts to become available. Such a delay can be important when evaluating the performance of monitoring systems, especially when the issue involves identifying sensitive situations.
According to Eitna, the statistics on teenagers’ duration of ChatGPT use, alongside disagreements about protective tools’ performance, indicate that assessing AI safety cannot be done solely by examining user usage. Evaluating the quality of responses, the performance of risk-detection systems, and the effectiveness of parental monitoring tools also play important roles in this assessment.
Ultimately, OpenAI’s challenge is not limited to developing educational capabilities; the company must demonstrate how reliable its protective tools are in sensitive situations. With the increasing presence of AI in teenagers’ lives, independent assessment of these tools’ performance and transparency about their limitations could become more important for families, schools, and policymakers.