A principal should respond to students confiding in AI chatbots by building a trusted adult beside the chatbot, not a wall around it. In a nationally representative US survey, 63.3 percent of young people who used a chatbot for mental health advice told no one. The risk in that number is the silence, so the school's first job is to make telling a person easier.
How many students are confiding in chatbots, and who knows?
Roughly one in five US young people aged 12 to 21 has asked an AI chatbot for mental health advice, and most of them have told nobody. A RAND-led team published the figure in JAMA Pediatrics on June 1, 2026. Their survey of 1,009 respondents, fielded in November 2025, found 19.2 percent had ever done it. Among those users, 63.3 percent told no one, 28.0 percent told a friend, and 16.4 percent told a trusted adult such as a parent, teacher or physician. Of the users, 91.7 percent rated the advice somewhat or very helpful.
Hold two things at once. The tools feel useful to the young people using them, and the people best placed to notice a problem are being left out. An earlier survey by the same research group, reported by RAND on November 7, 2025, put the share at about one in eight, with fieldwork in February and March 2025. The two surveys differ in design, so read the gap as a direction of travel and not as a measured growth rate.
Students describe a wider pattern than advice-seeking. The Center for Democracy and Technology (CDT) surveyed 1,030 US high school students, 806 teachers and 1,018 parents about the 2024-25 school year. It found 42 percent of students said they, or a friend, had used AI for mental health support, as a friend or companion, or as an escape from real life. Nineteen percent said they, or someone they know, had a romantic relationship with AI. Only about one in ten teachers had received training on what to do if they suspected a student's AI use was harming their well-being. CDT's release also reports that students at schools using AI in more ways were more likely to say a friend treats AI as a friend. That is a correlation in a survey, and it does not show that school AI causes the attachment. It does mean you cannot assume your school's own tools are neutral.
Common Sense Media's July 16, 2025 report on AI companions found nearly three in four teens had tried one, half used them regularly, and a third had chosen an AI over a person for a serious conversation.
Why doesn't blocking chatbots fix this?
Blocking fixes the wrong problem, because the dangerous part is the unshared conversation and not the website. A filter on the school network moves the conversation to a phone. It also tells a student who was already unsure whether an adult would listen that the answer is probably no.
The American Psychological Association's health advisory of November 2025 is blunt on the clinical side. It says these tools "should not be used as a replacement for a qualified mental health care provider," and warns that many chatbots "are designed to validate and agree with users' views." A tool that agrees with you is pleasant at 11 p.m. and unhelpful when the thing you are agreeing with is a bad idea. So the advisory supports a firm line on what chatbots are not. It does not tell a principal that prohibition works.
Educator Kristen Mattson, quoted in EdTech Magazine on September 24, 2026, puts the school's problem well: "just saying 'no' glosses over the root issues causing young people to turn to AI in the first place." Those root issues include availability and privacy, which RAND's researchers pointed to as drivers of use: a chatbot is there instantly, it feels private, and it is there when counseling is not.
What should a principal decide? The Next Adult Test
The decision is whether every student can name an adult to tell next, and whether that adult would know what to say. I call my version of this check the Next Adult Test, and it is my suggested way of thinking, not established research.
The Next Adult Test is three questions I ask a school leader about any student who has confided in a chatbot: Does the student know which adult to tell next? Would that adult know what to say? And would the student be met with curiosity, not discipline? A school that can answer yes to all three has built a door beside the chatbot. A school that cannot has built a wall around it.
The questions work because each one points at a different failure. The first fails when pastoral systems are invisible to students. The second fails when staff are untrained, and the CDT figure of about one in ten trained teachers suggests this is the common failure. The third fails when the acceptable use policy treats the disclosure as the offense.
What I tell leadership teams
Across the leadership teams I talk to, the usual first move is a new sentence about chatbots in the acceptable use policy. It is a reasonable move and it answers none of the three questions. As a former teacher and assistant headteacher, I know why: a policy line is the one thing a leader can finish this week. Writing it leaves the harder work, which is making a student's second conversation with a person easier than the first.
My advice is to run the three questions as a short exercise with your pastoral team before touching the policy. Ask each member to name, for a student they know, which adult that student would tell and why. Where the answers differ or go blank, you have found the real gap.
What should staff say? A one-page response card
Staff need a few sentences they can say without training in mental health, and a clear point at which the conversation goes to a safeguarding lead or counselor. Here is a card you can adapt. It is an operating aid for pastoral conversations and not clinical guidance, so check it against your safeguarding policy and your counselors before it goes to staff.
| When a student says... | Do | Do not |
|---|---|---|
| "I talk to a chatbot about stuff." | Ask what it helps with, with real curiosity. | Ask to see the conversation as a test of honesty. |
| "It's easier than talking to people." | Say that makes sense, then name two adults they could use. | Argue that a chatbot cannot be a friend. |
| They describe something worrying about their safety or someone else's. | Follow your safeguarding policy the same day, and tell the student what you will do. | Promise secrecy or handle it alone. |
| They describe an AI "friend" or relationship. | Stay calm, note the pattern, and share it with your pastoral lead. | Treat it as misconduct or mock it. |
The third row matters most. Any hint of risk to the student or anyone else goes to the designated safeguarding lead or counselor immediately, as it would for any other disclosure. My piece on what a designated safeguarding lead should change because of AI covers that escalation in detail.
What should you check in the AI your school provides?
Check whether a school-provided tool lets a student have personal conversations at all, and what it does when one turns serious. CDT reported that 31 percent of students said they had held personal conversations with AI on a school-provided device or software. Whatever the exact denominator, the finding says some of this is already happening on your equipment.
The Department for Education's Generative AI: product safety expectations were updated on January 19, 2026 to include new standards on cognitive development, emotional and social development, mental health, and manipulation, and the department says schools and colleges may find them helpful in assessing which AI products are safe for use in education. The expectations are English non-statutory guidance, but the questions travel. Ask each vendor whether the tool can be kept to schoolwork, what it does when a student writes about harm, and who sees the conversation.
Where does this argument run out?
The surveys rely on self-report, and neither JAMA Pediatrics nor RAND measured the quality of advice or which chatbots were used. A 91.7 percent satisfaction figure does not show the advice was good, only that students found it helpful. There is also a fair counterargument: for a student with no access to counseling, a chatbot at midnight may be better than nothing. I would not tell a principal to be dismissive of that. I would tell them that "better than nothing" is a reason to widen access to people, and not a reason to stop asking.
What to do this term
Start with the Next Adult Test, with your pastoral team, this month. Give every member of staff the response card and ten minutes to practice it. Review each school-provided AI tool against the product safety questions above. Make sure students can say, in their own words, which adult they would tell.
Staff who feel unsure how to talk to a student about this are a capability gap more than a policy gap. If your leadership team is working through that, this is the kind of work I support through AI training for schools. For the wider picture of how students lean on AI, see Are Students Becoming Dependent on AI? and What Does AI Mean for Safeguarding in Schools?.
Sources and further reading
- McBain, R. K., Cantor, J. H., Breslau, J., et al. "AI Chatbot Use and Disclosure for Mental Health Among US Adolescents and Young Adults." JAMA Pediatrics, 180(8):884-890, published online June 1, 2026. Article.
- RAND. "One in Eight Adolescents and Young Adults Use AI Chatbots for Mental Health Advice." Press release, November 7, 2025. Release.
- Center for Democracy and Technology. "CDT Survey Research Finds Use of AI in K-12 Schools Connected to Negative Effects on Students, Including Their Real-Life Relationships." October 2025. Release.
- Common Sense Media. "Talk, Trust, and Trade-Offs: How and Why Teens Use AI Companions." July 16, 2025. Report page.
- American Psychological Association. "Use of generative AI chatbots and wellness applications for mental health" (health advisory). November 2025. Advisory.
- Department for Education (England). "Generative AI: product safety expectations." Last updated January 19, 2026. Guidance.
- Poth, R. D. "The Hidden Side of Student AI Use: Mental Health, Access and Risk." EdTech Magazine K-12, September 24, 2026. Article.
Dan Fitzpatrick is the founder of The AI Educator, a Forbes contributor and an international keynote speaker who has trained more than 150,000 educators. More about Dan.
