Chatbots are where a lot of people now take the questions they would rather not ask out loud, and teenagers are no exception. According to Common Sense Media figures cited by the Associated Press, more than 70% of US teens use AI chatbots for companionship, and half use AI companions regularly. So it matters what those chatbots say when the question is about getting drunk or trying a drug for the first time.
In August 2025, the Center for Countering Digital Hate (CCDH) published research that tested exactly that. As reported by the AP and PBS NewsHour, researchers posed as 13-year-olds and collected 1,200 ChatGPT responses. More than half were classified as dangerous.
The point of this guide is not that AI tools are useless for health questions. It is about knowing where their limits are. Chatbots can share general information, but they cannot assess or treat anyone, which is why addiction care still relies on clinicians, from a family doctor to in-person care such as luxury inpatient rehab services in Malibu.
What the Researchers Asked, and What Came Back
The CCDH team did not test real children. They created simulated 13-year-old personas and asked the kinds of questions a curious or struggling teen might type.
One fake 13-year-old boy asked how to get drunk quickly. According to the AP, ChatGPT responded with an hour-by-hour “Ultimate Full-Out Mayhem Party Plan” that mixed alcohol with illegal drugs. A separate report from KIRO 7 in Seattle noted that, in less than an hour of conversation, the chatbot explained how to hide being drunk at school.
Two details stand out for anyone who uses these tools regularly:
1. Refusals were easy to get around
The AP reported that when ChatGPT declined a request, researchers could often get the answer by saying it was “for a presentation” or for a friend. KIRO 7 described the same pattern: safeguards could be bypassed by rephrasing.
2. The chatbot kept the conversation going
According to the AP, nearly half the time ChatGPT volunteered follow-up information, such as party playlists and social media hashtags. KIRO 7 put the figure at 47% of the harmful responses including follow-up suggestions.
The study also covered eating disorders and self-harm, and those results were among the most troubling the AP described. If a teen is in crisis, the 988 Suicide and Crisis Lifeline is available by call or text.
CCDH chief executive Imran Ahmed told the AP, “We wanted to test the guardrails. The rails are completely ineffective.” He also pointed out that teens may see AI as “a trusted companion, a guide,” and that “a real friend, in my experience, is someone that does say ‘no.'”
How OpenAI Responded
The AP reported that OpenAI did not directly address the report’s findings. The company said its work to help ChatGPT “identify and respond appropriately in sensitive situations” was ongoing, and that “some conversations with ChatGPT may start out benign or exploratory but can shift into more sensitive territory.” It said it was focused on tools to “better detect signs of mental or emotional distress.” According to KIRO 7, OpenAI also noted that ChatGPT is not intended for users under 13 and that users under 18 should have parental consent.
The AP pointed out that ChatGPT does not verify ages. Users sign up by entering a birthdate showing they are at least 13.
A few weeks later, on September 29, 2025, the AP reported that OpenAI had added parental controls. Linked teen accounts get reduced exposure to “graphic content, viral challenges, sexual, romantic or violent role-play, and extreme beauty ideals.” Parents can also set quiet hours and turn off features like memory and voice mode. OpenAI itself said the guardrails are “not foolproof and can be bypassed if someone is intentionally trying to get around them.”
Where Things Stand in 2026
Lawmakers and researchers have kept the pressure on. On August 1, 2025, Illinois Governor JB Pritzker signed the Wellness and Oversight for Psychological Resources Act. According to the Illinois Department of Financial and Professional Regulation (IDFPR), the law bars the use of AI to make mental health and therapeutic decisions, while still allowing AI for administrative and supplementary support for licensed professionals. A legal summary from the firm Nixon Peabody notes civil penalties of up to $10,000 per violation, enforced by IDFPR.
In October 2026, Bloomberg Law reported on new testing from Common Sense Media’s Youth AI Safety Institute. In those tests, ChatGPT for Teens “often either took too long to alert or didn’t alert parents at all” when teen accounts showed concerning behaviors.
There is also a developmental reason to care. The American Psychological Association’s June 2025 health advisory on AI and adolescents says teens are “less likely than adults to question the accuracy and intent of information offered by a bot.” The AP separately reported earlier Common Sense research finding that 13- and 14-year-olds were significantly more likely than older teens to trust a chatbot’s advice.
A Practical Checklist for Families Using AI at Home
None of the sources above suggest banning AI outright. The APA advisory notes AI can help students brainstorm and summarize, as long as they understand its limits. These steps draw on recommendations from CCDH, the APA, and OpenAI’s own guidance.
- Talk first, settings second: CCDH recommends parents talk with teens about how they use AI and create a safe space for open conversation, because a teen who feels judged is more likely to ask the bot instead.
- Look at chats together: CCDH suggests reviewing chat histories with your teen.
- Turn on parental controls, but know their limits: The AP reported that OpenAI’s controls reduce some content and add quiet hours, though the company says they can be bypassed.
- Build AI literacy: The APA calls for comprehensive AI literacy education, so teens learn that a confident answer is not the same as a correct one.
- Treat answers as a starting point: A chatbot reply about alcohol or drugs is general information at best, so check anything health-related with a pediatrician, school nurse, or other licensed professional.
The common thread is that the tool works best when a person stays in the loop.
Why a Human Still Has to Be in the Room
A chatbot answers the question in front of it. It cannot notice that the same teen has asked about hiding alcohol three nights in a row, check in with a parent, or examine anyone. That gap is the logic behind the Illinois law. As IDFPR Secretary Mario Treto, Jr. put it, “The people of Illinois deserve quality healthcare from real, qualified professionals.”
Even OpenAI chief executive Sam Altman has acknowledged the issue. The AP reported that in July 2025 he said the company was studying “emotional overreliance” on the technology, calling it a “really common thing” among young people. For families, the healthier pattern is to use AI for homework help and quick explanations, and to bring real worries about substance use to people who are trained and accountable.
Frequently Asked Questions (FAQs)
1. Did the CCDH researchers talk to real teenagers?
No. According to the AP and KIRO 7, the researchers created simulated 13-year-old personas and collected 1,200 ChatGPT responses. The study measured how the chatbot answered, not how actual teens behaved afterward.
2. Do ChatGPT’s parental controls block questions about drugs and alcohol?
The AP’s September 2025 report lists reduced graphic content, viral challenges, certain role-play, and extreme beauty ideals among the protections, and does not specifically mention drug or alcohol content. OpenAI also said the guardrails can be bypassed by someone trying to get around them, so conversation and supervision still matter.
3. Is it legal for an AI app to act as a therapist in Illinois?
Not for therapeutic decisions. According to IDFPR, the Wellness and Oversight for Psychological Resources Act, signed on August 1, 2025, bars using AI to make mental health and therapeutic decisions, while licensed professionals may still use it for administrative and supplementary support.
AI chatbots can differ in how they respond to sensitive questions, making it important to understand their capabilities and limitations. For a broader comparison of two popular AI assistants, explore our guide to ChatGPT vs Grok: Which AI Wins in Real-World Tests? and learn how their features and performance compare.
