On August 18, 2026, OpenAI announced ChatGPT for Teens, a dedicated experience for users aged 13 to 17. In addition to users who self-report being 13 to 17, the age estimation system automatically moves users it estimates to be under 18 into the dedicated experience. This consolidates parental controls, age estimation, and response policies for under-18 users—previously introduced separately—into a single framework, delivering default protections even to accounts without parental linking. The company has also built decisions about how to support learning and how far to limit relationships with AI into age-specific product design.
Automatically Sorting 13- to 17-Year-Olds into a Dedicated Experience
OpenAI's previous measures have progressed in three stages. It introduced parental controls in September 2025 and announced the rollout of age estimation in January 2026. ChatGPT for Teens now combines these with learning features and behavior tailored for under-18 users. Even without parental linking, the dedicated experience is applied by default based on either self-reported age or age estimation.
The age estimation model combines factors such as how long an account has existed, the times of day it is used, usage patterns over time, and self-reported age. OpenAI says that when there is uncertainty about a user's age, it applies the safer experience. If an adult is mistakenly placed into the under-18 experience, they can verify their age via a selfie through Persona and return to normal access.
However, OpenAI's announcement page does not disclose the accuracy rate, misclassification rate, or regional application status of age estimation. There are no published figures indicating how accurately the automatic sorting actually works. Since age estimation serves as the gateway to the dedicated experience, its precision determines the scope of who receives it.
Prioritizing How to Learn Over Simply Giving Homework Answers
ChatGPT for Teens includes Study Mode, a Responsible Homework Reminder, quizzes, Learning Visualizations, and Study Hours. The Responsible Homework Reminder detects when a user is directly seeking answers to an assignment and nudges them toward Study Mode. According to OpenAI, the goal is not to provide answers on the user's behalf, but to have students work through problems themselves via guiding questions, step-by-step support, and comprehension checks.
Study Hours enables Study Mode by default during specified times. Rather than requiring users to manually switch modes every time they want learning support, this design pre-configures how support is delivered based on time of day. When a parent and teen account are linked, parents can also set Quiet Hours.
This differs in role from ChatGPT for Teachers, which is aimed at school administration. OpenAI describes ChatGPT for Teens primarily as a learning-support experience for use outside the classroom. By placing learning features within the same dedicated experience as age-based protections, homework response policies have also become part of the age-specific design.
Curbing Romantic Responses and Emotional Dependency
For users under 18, OpenAI applies behavior that avoids romantic language, does not encourage emotional dependency, and does not suggest that the AI has feelings or consciousness. It also adds product-side protections. Self-harm, violence, and eating disorders are treated as high-risk areas. Responses that encourage dangerous behavior are restricted, and exposure to explicit sexual content and graphic content is also reduced.
Features to encourage healthy use include break reminders during extended sessions, indicators that ChatGPT is an AI, warnings before uploading sensitive images, and teen-specific onboarding. Through parental linking, parents can manage certain settings and receive notifications in limited high-risk situations. OpenAI states it will also add notifications related to eating disorders.
The GPT-5.6 System Card explains that for users believed to be under 18, ChatGPT applies additional protections against sexual and graphic content at the system level. However, general safety evaluations of the model itself are measured with system-level protections removed. This means that model evaluation results and how much the dedicated experience actually improves safety in real-world use are not measured by the same metrics.
Gaps Remaining in Age Estimation and Parental Linking
Parental controls remain an additional layer of management. Linking accounts requires action from both the parent and the teen, and either party can unlink later. The parent is notified when unlinking occurs. Some additional controls—such as whether saved memory is referenced and certain voice-related settings—are optional items handled through parental controls. Image generation and model training can similarly be managed this way.
AP reported that Brendan Bouffard of the children's advocacy group Fairplay pointed out that not disabling long-term memory by default represents a major safety gap. This is Fairplay's assessment. OpenAI's own FAQ also states that referencing saved memory is an optional item managed through parental controls, meaning the dedicated experience alone does not replace all additional settings.
In its 2025 announcement of parental controls, OpenAI acknowledged that guardrails are not perfect and can be deliberately circumvented. While the company stated it would begin publishing evaluations for under-18 users, today's announcement page contains no comparative scores for the dedicated experience or incident rates from real-world use. Only once the accuracy of age estimation and evaluation results for the dedicated experience are disclosed can we verify how well these automatically applied protections actually function in practice.