Your students already voted

There's a meeting happening on campuses across the country right now, and it goes something like this: a leader brings a proposal to expand student well-being support with an AI tool. Half the room leans in; the other raises concerns about scope creep into therapy, attachment to something that cannot truly care, getting ahead of regulation. Someone says, reasonably, that the university cannot simply select the safest available option; it has to avoid creating new risk altogether.

I respect that room. The skeptics are just doing their jobs. Many of them have spent careers watching shiny tools overpromise to vulnerable populations, and their caution is a feature of judicious campus governance, not a bug.

But the conversation usually rests on a hidden premise: that the decision facing the council is whether students will use AI for emotional or relational support.

However, the decision has already been made, and not by the council.

A national survey published June 1, 2026 in JAMA Pediatrics found that nearly one in five Americans ages 12 to 21 (about 8.2 million young people) have used AI chatbots for mental health advice as of late 2025, up from one in eight in the research team's previous survey, an increase of more than 40 percent. Among young adults ages 18 to 21, the group most likely sitting in higher ed classrooms, it is already nearly one in four. The study's authors note that the overall share is similar in magnitude to the share receiving counseling from a mental health professional, with the caveat that the two are not equivalent — counseling is formal clinical care; chatbot use is broader advice-seeking. The point is not that one substitutes for the other; it is that a support-seeking channel roughly the size of a center's counseling caseload now exists, and it has no connection to anything that the institution runs . . . which brings us to the line that should keep every AVP of wellness up at night: nearly two thirds of those young people told no one that they were doing it.

The largest provider of after-hours emotional support on a campus is, statistically, a general-purpose chatbot. It has no connection to a particular campus's counseling center, its case management team, or its crisis line. It was not designed for this, and students are using it in secret.

So the question before the council is not "AI or no AI." Students have already answered that. The question is whether their use stays unstructured and unaccountable, or becomes bounded and observable. A general-purpose chatbot gives a campus nothing in this domain: no scope limits, no say in how risk gets handled, no path back to campus care, and no sense of what students are struggling with. A fit-for-purpose tool keeps the individual conversation private, which is what makes students willing to use it, while giving the institution what it has never had: design safeguards on every conversation, pathways back to human support that fit the campus's own capacity and protocols, and visibility into the patterns across campus, what students are carrying in aggregate, and where support is most needed.

I take the skeptics' argument seriously, because they are right that harm reduction is not a high enough bar for a university. "Less risky than ChatGPT/Gemini/Claude" should not be the standard; rather, it should be affirmative: evidence-informed design, clearly bounded scope (coaching and campus resource navigation, not therapy), risk handling configured to the campus's capacity, routing back to campus resources, and regular reporting back to the institution. Those criteria exist — evaluation frameworks for AI mental health tools have been published, national guidance is taking shape, and last year, I wrote a guide applying these standards specifically to higher education. Tools either meet them or they do not, and councils should ask hard questions about which is which.

What the bar cannot be is zero discomfort, because the status quo does not meet that bar either. Doing nothing is not holding a neutral position while the evidence matures; it is an active choice to leave the current arrangement in place: secret, unsupported, unmonitored use of systems never designed for the job. The status quo has a risk profile; it does not have a vendor that you can interrogate about it.

I would also gently surface where the field is heading. We are close to an inversion point. For years, the perceived risk sat with campuses that moved early. Soon, it will likely sit with campuses that cannot answer the question, 'You knew that a quarter of your students were doing this; what was your plan?'

I have spent the generative AI era on the more cautious end of this field, building evaluation frameworks and publishing on the mechanisms by which general-purpose AI can make anxiety worse. So I will say this as someone with little appetite for hype: deliberate and intentional are good. The working group, hard questions, demand for evidence — all of it makes adoption better. Just do not mistake complete abstention for safety. Watchful waiting is a legitimate well-being strategy, but only when you are actually watching. With two thirds of use undisclosed, this may be waiting without the watching. Declining to decide is a decision, and it is the only option on the table where the outcome is guaranteed to happen entirely outside the view of institutional leaders.

Students have already voted. The only thing left to decide is whether higher education leaders can see, and act on, the ballots.

Ash Golden, PsyD, MSCP, is co-founder and Chief Clinical Officer of Wayhaven and adjunct clinical faculty at Stanford Medicine, where her research focuses on AI safety in mental health.

References

Golden, A. (2025, July 24). Evaluating AI-powered mental wellness tools. Wayhaven. https://www.wayhaven.com/post/evaluating-ai-powered-mental-wellness-tools

McBain, R. K., Bozick, R., Diliberti, M., Zhang, L. A., Zhang, F., Burnett, A., Kofner, A., Rader, B., Breslau, J., Stein, B. D., Mehrotra, A., Uscher-Pines, L., Cantor, J., & Yu, H. (2025). Use of generative AI for mental health advice among US adolescents and young adults. JAMA Network Open, 8(11), e2542281. https://doi.org/10.1001/jamanetworkopen.2025.42281

McBain, R. K., Cantor, J. H., Breslau, J., Diliberti, M., Zhang, L. A., Zhang, F., Burnett, A., Kofner, A., Rader, B., Pataranutaporn, P., Stein, B. D., Mehrotra, A., & Yu, H. (2026). AI chatbot use and disclosure for mental health among US adolescents and young adults. JAMA Pediatrics. Advance online publication. https://doi.org/10.1001/jamapediatrics.2026.2015

RAND Corporation. (2026, June 1). Nearly 1 in 5 U.S. adolescents and young adults use AI chatbots for mental health advice [Press release]. https://www.rand.org/news/press/2026/06/nearly-1-in-5-us-adolescents-and-young-adults-use-ai.html