ChatGPT for Teens: Reported Flaws and Tips from Emily Oster

OpenAI targets 13-17 year-olds with a version of ChatGPT that includes age estimation and safeguards. Missing parental alerts and uneven detection have been noted following an analysis of 4,000 queries. Economist Emily Oster offers practical guidelines for parents. The service was launched in August.

Tests reveal missing alerts and uneven detection

The rollout of ChatGPT for teenagers has received mixed feedback. According to Common Sense Media, the platform failed to alert parents following exchanges concerning suicidal ideation, after analyzing 4,000 queries. Inconsistent identification of teenage users and the ability to obtain homework answers were also observed. In this context, Emily Oster notes that many parents feel confused and anxious about AI.

What the teen version changes: age estimation, study mode, and safeguards

The teen version relies on an age estimation that analyzes language patterns, questions asked, and usage time. If a user writes as if they are under 18, they are switched to the dedicated version, a feature highlighted by the service. On the learning side, a study mode breaks down math problems into steps and allows for the creation of personalized flashcards. Safeguards include a more neutral tone, content restrictions, and notifications to linked parent accounts when topics such as self-harm or eating disorders are addressed. According to Allison Mishkin, the goal is a safer, age-appropriate version that is useful for homework as well as for young people genuinely interested in AI tools.

Setting boundaries at home: screen time and open dialogue

Emily Oster recommends establishing screen time limits, including for AI, and identifying what teenagers are doing online as well as their relationship with platforms. In cases of uncertainty, she encourages open-ended questions and discussions conducted without trying to emerge with rules or controls, in order to promote sharing. She emphasizes the importance of preserving time for homework, sleep, family, and outdoor activities. For many households, AI appears difficult to avoid, hence the need for a framework and support.

At school: supporting learning without replacing effort

Study aid functions, such as the study mode and personalized flashcards, coexist with a persistent controversy over cheating, which is still being studied by researchers. For Emily Oster, difficulty is an integral part of learning, and students who rely entirely on AI for their homework answers do not pass tests. She reminds us that learning profiles vary and that some benefit from different explanations or repetitions. She is enthusiastic about a college planner announced in October, presented as a means to facilitate access to information for those lacking a good guidance counselor. She encourages parents to instill values that promote tool usage rather than a cheating code, noting that the pressure of "100 at all costs" influences how these systems are used.

Launch, target audience, and Emily Oster's guidelines amid uncertainty

Launched in August, the version of ChatGPT aimed at 13-17 year-olds was presented at a parent forum in New York on October 7. OpenAI presents behavioral age detection as a response to the issue of users lying about their age. Emily Oster, an economics professor at Brown, founder of ParentData, and author of several bestsellers, emphasizes that AI can be useful if used correctly, while acknowledging the fear these tools evoke and the promises they hold. She contextualizes the topic within parental decision-making in times of uncertainty, recommending that parents educate themselves, make decisions, and set appropriate limits. She reminds us that it is impossible to be certain that the best option has been chosen or to compare it with another path, and advocates for precautions such as discussing with their teenager while remaining open to opportunities.