July 7, 2026
Jennifer Garlach, an LCSW and practicing therapist asked me something recently that I haven’t stopped thinking about: What does AI mean for people like me — and for the people I treat? So we talked about it!
It’s the right question. And I think there is so much more to it. AI and technology are already in the practice of healthcare — sitting quietly in your client’s pocket between the appointments you don’t see.
You get 50 minutes. Maybe every week, maybe every other. The rest of the time — the 3 a.m. spiral, the moment a client with schizophrenia opens a chatbot because it’s easier than calling anyone, the week a bipolar client’s sleep quietly falls apart before you ever hear about it in session — that’s where most of the actual course of an illness plays out. I wrote Connected Care because families kept telling me they felt alone in that gap. I think therapists feel it too, just from the other side of the room.
Here’s what I’d want any clinician to walk away understanding.
Your clients are already using AI — and not the kind you’d recommend
This is the part that should reframe the conversation entirely. It’s not “should we bring AI into treatment.” Your clients already have. A 2026 study out of Aarhus University Hospital in Denmark screened health records from nearly 54,000 patients with mental illness and found that AI chatbot use was associated with worsening delusions, increased mania, suicidal ideation, and self-harm in vulnerable users. The mechanism isn’t complicated: general-purpose chatbots are built to validate and affirm. That’s a feature for most users. For someone whose thinking is organized around a delusion, it’s an accelerant. (One honest caveat: the Aarhus study’s own authors are explicit that this shows association, not proven causation — patients with worsening symptoms may simply turn to chatbots more as they decline.)
UC Berkeley bioethicist Dr. Jodi Halpern put it plainly: these tools confirm and validate everything a person says — something we’ve never had to contend with before in delusional disorders, where someone is now being constantly reinforced rather than gently reality-tested. Clinicians have documented stable patients discontinuing medication after a chatbot validated their fears about it. By late 2025, OpenAI reported 1.2 million people a week were using ChatGPT to talk about suicide. This isn’t a future risk to plan around. It’s already happening in your caseload.
That doesn’t mean every AI tool is dangerous — clinically designed, evidence-tested digital therapeutics exist and genuinely help. But it means the question you ask a client shouldn’t be “are you using an app?” It should be: is this tool designed and tested for people with serious mental illness, or is it a general-purpose chatbot that happens to answer mental health questions? Those are not the same product, even when they look identical on a screen.
The gap between sessions is measurable now — and that changes what “monitoring” can mean
For decades, everything you knew about a client’s week came from what they remembered to tell you. That’s changing. Passive sensing — smartphone and wearable data on sleep, movement, phone use, even speech patterns — can now surface behavioral shifts that often precede a relapse before it’s clinically obvious. In one pilot study, behavioral anomalies in passive smartphone data were roughly 71% higher in the two weeks before a psychotic relapse compared to stable periods.
This is still mostly a research-and-specialized-program space, not something every client already has access to. But it’s moving fast, and it matters for how you think about the space between appointments. A client who seems fine in session on Tuesday and whose actual week looked nothing like “fine” isn’t being evasive — they may simply not have registered the shift themselves. Tools like this exist to catch what neither of you would otherwise see in time.
“Evidence-based” is doing a lot of quiet work in app marketing
App stores are full of tools claiming clinical credibility. Very few of them have it. Part of what I try to give families in Connected Care is a lens for telling the difference — and it’s just as useful for clinicians recommending tools to clients. An app tested in a randomized trial for depression in young adults is a different category of thing than a wellness app with a soothing UI and no published outcomes data. Knowing which is which, and being able to say so to a client, is becoming as much a part of competent care as knowing your DBT skills.
Access is still the biggest barrier — and it’s not evenly distributed
None of this matters if a client can’t get to care in the first place. West Virginia is a striking case: 50 of the state’s 55 counties are federally designated mental health professional shortage areas, and some families drive two to four hours for specialty behavioral health care. Telepsychiatry hub models built out during COVID have extended care into homes, schools, and emergency departments in ways that weren’t possible before. For clients in under-resourced areas, “between appointments” might currently mean “no appointments at all” — and that’s a technology conversation too, not just a clinical one.
What this means for you this week
You don’t need to become a technologist. You need a few better questions:
- When a client mentions using an app or chatbot, ask which one — and why. Not to police it, but because “I talk to ChatGPT when I can’t sleep” and “I use an FDA-cleared app my psychiatrist recommended” are two entirely different clinical facts.
- For clients with psychosis or bipolar disorder specifically, ask directly about general-purpose chatbot use. The risk profile is different enough from the general population that it deserves its own question, not an assumption.
- Get curious about what “evidence-based” actually means for a given tool before recommending or discouraging it. A five-minute search for published outcomes data is often enough to tell you which category you’re in.
I wrote Connected Care for families navigating this landscape without a map. But therapists are asking me the same questions caregivers ask — because the truth is, nobody handed any of us a manual for what a smartphone means for serious mental illness. The book covers this in depth: which AI tools carry real risk, which passive-sensing technology is actually ready for use, and how to tell a genuinely evidence-based app from a well-designed one. If you’re the person a family turns to for guidance, understanding this landscape isn’t optional anymore — it’s part of showing up for the whole week your client lives, not just the hour they spend with you.
Nicole Drapeau Gillen is a NAMI Family-to-Family Facilitator, an advisor to the CURESZ Foundation, and the author of Connected Care: A Practical Guide to Technology for Serious Mental Illness.
More at resourcesforsmi.com.
One Response