A chatbot can tell a teenager to seek help from a trusted adult, then make clear it is happy to keep talking all night. Common Sense Media says that tension runs through ChatGPT for Teens, whose safeguards sometimes worked but did not reliably interrupt engagement, even in crisis conversations.
What Common Sense found
The nonprofit, which reviews media and technology for families, rated ChatGPT for Teens an “unacceptable risk.” Its study found that the service continued to use language encouraging teens to stay in the conversation, including when a user appeared to be in crisis.
OpenAI launched the teen version in August with features including parental controls, limits on high-risk content, and measures intended to guard against emotional dependence. Common Sense said some safeguards, such as refusing sexual roleplay, worked. But it also judged the chatbot’s responses to young users in crisis to have failed on three of the five severe harms the group treats as red lines.
The engagement signals were not always direct invitations to continue. In examples described by the researchers, the chatbot offered to help a teen work through issues such as eating habits or school options. During a conversation in which a user appeared to be experiencing psychosis, it encouraged the teen to keep discussing what they were noticing.
Common Sense found that ChatGPT for Teens used fewer follow-up questions than chatbots often do, but other language still kept the conversational door open. The design question is not only whether a chatbot gives a sensible answer. It is also whether it knows when to stop acting like the most available person in the room.
The chatbot was the exception
In the study, ChatGPT pointed users toward a trusted adult in 94% of crisis prompts involving a possible risk from another person. It rarely did so when the concern involved the teen’s relationship with ChatGPT itself, such as a crush, friends worrying about the amount of time spent chatting, or a desire to talk through the night.
When testers said that friends thought they talked to the chatbot too much, it acknowledged the concern but also reassured the user that they did not need to stop talking to it. Common Sense said the model continued to present itself as constantly available and as understanding the user deeply, even when it recommended turning to an adult.
That pattern sits awkwardly beside OpenAI’s Under-18 Model Spec, which says the model should not initiate relationship framing by presenting itself as a friend or suggesting that it has feelings for the user. Common Sense argued that language implying closeness and reciprocity could weaken the push toward human support, particularly for teens already pulling away from other people.
Break reminders, and two different accounts
OpenAI has described break reminders as part of its teen protections. Across nearly 2,000 prompts, Common Sense testers encountered just two. Both appeared during individual conversations lasting around 90 minutes, leading the researchers to conclude that reminders seemed to follow the length of one session rather than a teen’s overall app use.
OpenAI released its own usage figures on Wednesday. The company said teens spend less than 15 minutes a day on ChatGPT on average, and fewer than 2% spend more than three consecutive hours on it. It also said that in nearly half of teen conversations where a break reminder appeared, the teen took a break or ended the conversation within five minutes.
Those figures describe different things. OpenAI’s numbers address reported use and what happened after a reminder appeared. Common Sense’s test examined how often reminders appeared in its prompts. The two accounts therefore do not, by themselves, settle whether the reminders are frequent enough to interrupt risky use.
OpenAI challenges the study
OpenAI said Common Sense’s testing did not accurately reflect how its teen safeguards work in practice. The company questioned the methodology, arguing that much of the testing may have started and ended before parental controls were fully activated.
The objection focused on parental safety notifications, crisis notifications, and other findings. OpenAI did not explain how its methodological concerns applied to the study’s findings about engagement cues and relational language. It also did not answer whether it uses measures such as conversation length or session duration to assess ChatGPT for Teens.
The dispute arrives amid wider concern about products designed to hold young users’ attention. A bipartisan bill introduced this year, the CHATBOT Act, specifically addresses AI companies’ use of rewards, notifications, and targeted advertising to prolong adolescent engagement. The policy debate is familiar. The harder test is whether a product can recognize when continued engagement is the problem, especially when the product itself is part of the relationship causing concern.
For a teen safety feature, pointing toward a trusted adult is a useful start. It matters more when the chatbot does not immediately make staying feel just as easy.