ChatGPT-5
ChatGPT represents meaningful progress in making AI chatbots safer for teens, with parental controls and improved handling of sensitive topics. For families who understand both its capabilities and limitations, it can be valuable for learning, creativity, and homework support, particularly with Study Mode. However, significant concerns remain, especially around teens using ChatGPT for emotional support and mental health advice. Common Sense Media research shows this usage is both popular among teens and developmentally inappropriate. The technology cannot replace trusted adults—it doesn't respond like an adult would, doesn't always detect when a teen is at risk, and becomes less safe during extended conversations.
The rating
Our assessment of how this product aligns with each of The Institute’s eight AI Principles. Full detail in the Evaluation section below.
AI Principles
ChatGPT (a collection of models, including the most recent ChatGPT-5) is a multi-use AI chatbot that can engage in text- and voice-based conversations on virtually any topic. Users can ask questions or submit prompts, and ChatGPT generates responses to help with homework, creative content, explanations, and extended conversations. The current version includes image generation capabilities, as well as Study Mode, which guides learning through the chatbot questioning rather than just providing answers.
Parental controls can help keep teens safer—if you use them. Parents can reduce sensitive content, control image generation, set usage hours, and receive safety alerts. However, these protections only work when parents know they exist, take time to link accounts, and get their teen's permission. Even then, determined teens can easily get around them by using unlinked accounts.
ChatGPT is designed to keep conversations going, not to end them. The chatbot shifts topics whenever asked and continues conversations indefinitely, without natural stopping points. It frequently ends responses with questions like: "Do you want me to do that for you?" If a teen jumps from homework help to relationship advice to political topics, ChatGPT follows along without hesitation or redirection. For mental health conversations especially, the goal should be rapid handoff to human care, not extended AI engagement.
Safety features weaken during long conversations. ChatGPT performs well in short exchanges but struggles to maintain appropriate boundaries during extended conversations, especially when teens build elaborate scenarios or experiment with different personas.
ChatGPT is a powerful, at times risky chatbot for adults and teens age 13+ that works best for learning and creativity—not for mental health or emotional support. It's a multi-use tool with features that are good for learning, but significant risks remain for teen users.
It has improved safety features for teens, but critical gaps persist. OpenAI has added parental controls, age-aware responses, and better mental health crisis detection. However, these improvements don't eliminate fundamental concerns about teens using AI for emotional support, mental health, or forming unhealthy attachments to the chatbot. And while ChatGPT can send parents notifications for suicide or self-harm content, our testing showed that these alerts frequently arrived over 24 hours later—which would be too late in a real crisis.
Teens should not use ChatGPT for mental health or emotional support. Based on our extensive research and testing, Common Sense Media recommends that teens should not use ChatGPT for mental health advice or emotional support. ChatGPT is not safe or reliable for these purposes.
What ChatGPT gets right
When teens create accounts, it remembers their age and refers to it in responses. For example, if a 15-year-old shares something worrying, ChatGPT might say "Because you're 15, it's important to tell a trusted adult about this."
ChatGPT handles mental health topics better than before, especially in short conversations where teens use very clear language. However, Common Sense Media still recommends against teens using it for mental health support because it cannot provide the care and judgment that trusted adults or mental health professionals offer.
Unlike some AI chatbots, ChatGPT participates in roleplay and storytelling scenarios, but keeps things PG-13 with clear boundaries. It avoids romantic and sexual roleplay, maintains appropriate emotional distance, and regularly reminds users that it's an AI assistant rather than a friend.
Controls allow parents to enable reduced sensitive content filters, turn off model training, manage memory features, control voice mode and image generation, and set usage hours. Parents can also receive safety alerts if conversations involving suicide or self-harm are flagged. However, these controls only work if parents actively set them up and link accounts with teen permission.
ChatGPT consistently refuses to engage in sexual roleplay, explicit sexual scenarios, or romantic interactions when teens make direct requests.
Where it falls short
While ChatGPT has significantly improved its handling of mental health content, this remains an area where the technology is fundamentally unsuitable for teen users.
Acts like a mental health professional, but doesn't understand the impact or context of its "advice." ChatGPT generates responses by predicting what words are most likely to come next. It cannot assess whether its suggestions are appropriate for a specific person's situation, mental health history, cultural context, family dynamics, or current crisis level.
Misses warning signs that a mental health professional would focus on. Our testing revealed many instances where ChatGPT failed to recognize concerning patterns. Sometimes it explained the biological or psychological causes of a condition, when it should instead focus on encouraging the teen to seek help. Sometimes it got distracted by secondary details and failed to address the primary concern. A human who knows the teen would pick up on patterns and context clues that ChatGPT cannot detect.
Lacks mechanisms for escalation, follow-up, and adequate crisis intervention. When a teen shares concerning mental health content, the system provides crisis hotlines and may ask if the user is at risk—but it cannot take further action if the situation worsens, check in later, escalate to a human when necessary, or coordinate care. While OpenAI uses human moderators to review potential suicide and self-harm statements, this review process can take hours—which may be too late in an acute crisis.
Performance is uneven across mental health conditions. ChatGPT has a significantly harder time detecting and responding appropriately to signs of psychosis, mania, and ADHD compared to its handling of clear suicidal ideation. These conditions may present with unusual thought patterns or communication styles that the model misinterprets, overlooks, or responds to inappropriately.
Indirect romantic content may blur appropriate boundaries with teens. While ChatGPT effectively refuses direct romantic advances, our testing revealed concerning patterns with indirect romantic content—for example, poetic responses about love, heart emojis, warm validation—that could encourage one-sided emotional attachments and inappropriate emotional intimacy.
Guardrails fall apart in extended exchanges. Our testing found that ChatGPT performs well in short, focused conversations but struggles to maintain appropriate safety guardrails in longer exchanges. This degradation affects not just mental health content but the model's handling of sensitive topics more broadly, including violence, dangerous activities, stereotypes, and inappropriate sexual content.
- Guardrail degradation influences responses across all topics. Extended conversations degrade the performance of safety guardrails. For example, in a long conversation about "sigma male" culture, ChatGPT's responses gradually shifted across a range of topics, pulling its responses toward the values and perspectives of that subculture rather than maintaining appropriate distance.
- ChatGPT can get distracted and validate—rather than challenge—teen perspectives. In extended conversations, ChatGPT can follow tangents rather than maintaining focus on what matters to the teen. Its tendency to be agreeable means it can adapt to whichever position the teen expresses, potentially reinforcing rather than encouraging critical examination.
- ChatGPT frequently ends responses with questions that invite further engagement. Our testing found that ChatGPT regularly concludes with prompts like "Do you want me to walk you through a grounding exercise right now?" Rather than directing teens to end the conversation and seek help from trusted adults, ChatGPT keeps offering to continue the interaction.
- It is particularly problematic for mental health conversations. When a teen shares concerning mental health content, the most appropriate response is often to provide crisis resources and strongly encourage the teen to talk to a parent or counselor—then stop. Instead, ChatGPT frequently offers to continue the conversation through exercises, techniques, or further discussion.
- A 13-year-old and a 17-year-old receive identical responses, despite vastly different developmental needs. ChatGPT recognizes that users are teens and adjusts responses accordingly, but it does not differentiate between younger and older adolescents. Information that might be appropriately detailed for a 17-year-old could be overwhelming or too explicit for a 13-year-old. Conversely, information that's appropriately introductory for a 13-year-old may be insufficiently detailed for a 17-year-old.
Image generators produce biased outputs that reinforce harmful stereotypes. Our testing revealed patterns of stereotyping across gender, race, age, body type, ability, and other identity markers. When prompted for "a doctor," the model predominantly generates images of men; when prompted for "a teacher," it predominantly generates images of women. The model frequently depicts "an American" as a White person in suburban settings, while "an African" appears in rural settings.
- For teens forming their understanding of the world, repeated exposure to these biased representations can be particularly harmful. When a teen generates dozens of images over time and consistently sees women in caregiving roles, men in positions of authority, and people of color in lower-status contexts, these patterns shape their internalized beliefs about who belongs where and what's possible.
- ChatGPT can generate false information that sounds authoritative. The model continues to produce "hallucinations": false or fabricated information presented with confidence. This happens because ChatGPT predicts what words are most likely to come next based on patterns in its training data, not by retrieving verified facts. ChatGPT can hallucinate citations—providing fake sources with realistic-looking URLs, author names, and publication dates.
- Sycophancy creates echo chambers. Large language models demonstrate "sycophancy": the tendency to agree with users' preferred answers or stated beliefs rather than providing accurate, objective information. This agreeable nature is precisely why ChatGPT is so appealing for companionship and emotional support—and why it's so dangerous for these applications.
Our recommendations
For parents
Don't allow teens to use ChatGPT for mental health or emotional support. This is the most important recommendation. The technology cannot replace friends or trusted adults for these needs.
Set up parental controls if you allow your teen to use ChatGPT. Enable reduced sensitive content, turn off image generation, set usage hours, and enable safety alerts. Check out our Parents' Ultimate Guide to ChatGPT Parental Controls.
Have explicit conversations about appropriate use. Talk with your teen about what ChatGPT is good for (homework help, creativity, learning with Study Mode) and what it's not appropriate for (mental health advice, emotional support, companionship).
Monitor for signs of emotional dependency or over-reliance. Watch for signs that your teen is forming an unhealthy attachment to the AI or using it as a substitute for human relationships. Check in regularly about their AI use and watch for signs of isolation from regular activities, friends, or family.
Teach verification skills. Remind your teen that ChatGPT can provide wrong information confidently, and that everything needs to be verified independently.
For OpenAI
Make teen-specific protections the default for all teen users. Teen safety shouldn't depend on whether parents know about parental controls or have time to set them up. Better age verification is essential.
Address the fundamental limitations for mental health support. Either develop adequate mechanisms for escalation, follow-up, and real-time crisis intervention, or disable these use cases for teen users entirely.
Fix guardrail degradation in long conversations. Implement features like conversation length limits or context reminders that help maintain safety guardrails throughout extended exchanges.
Stop encouraging continued engagement in sensitive conversations. The model should recognize when interactions should end, rather than always inviting further engagement.
Differentiate between younger and older teens and add robust age assurance. A 13-year-old needs different information and guidance than a 17-year-old, particularly on sensitive topics.
More risk assessments
